Steel Mountain

Introducing Steel Mountain: A Digital Guard Dog for your Connected Home

Steel Mountain is not your grandpa’s anti-virus software. It’s a new way of looking at digital security for connected homes. Today, internet connected devices are proliferating in the home, and with them, opportunities for breaches of privacy and digital theft. Steel Mountain’s flagship product, Secaura, will attach directly to a home’s wifi router, and provide comprehensive protection against attacks anywhere on its network, 24 hours a day, without impacting the speed of your internet connection. Best of all: it’s self-updating, requiring no intervention from the user to keep security up to date, all year round.

StartupYard recently announced Steel Mountain as a member of our 8th Batch of tech startups, with a Demo Day coming up on November 22nd, 2017, in Prague. 

I caught up with Will Butler, CEO at Steel Mountain, to talk about digital home security, and his plans for changing the standards of privacy and safety for the Internet of Things. Our interview is below:

 

Hi Will, first why don’t you tell us a bit about your team, how you started working together, and how you came up with the idea behind Steel Mountain?

Thomas Maarseveen and I go back seven years whilst studying engineering together in the UK. Since meeting we’ve collaborated on a number of technical projects together; websites and apps, crypto, drones, 3D printers.. We’ve done a lot of cool stuff over the years together.

Will Butler, CEO and CoFounder at Steel Mountain

The idea behind Steel Mountain originally came from Thomas’ browser-based AdBlocker not working. To solve this issue, Thomas decided to block his ads by hacking his router and filtering out the ads there rather than the end-point. This then inspired the idea of blocking digital threats at the gateway of the network- the router. The evolution of this idea was natural as Thomas and I had a history of ethical hacking and experience in cyber security and networking.

Thomas and Will, Co-Founders at Steel Mountain

What drew you to focusing on security, and particularly consumer level security?

Whilst i hold value in my security and privacy, I have always despised using my anti virus software. It slows down my PC, conflicts with other software and intrudes on my usage with annoying popups all the time. Moreover, with critical security issues fast approaching with the IoT (Internet of Things) booming, a solution is needed now.

The consumer security sector has been neglected for a long time. We haven’t seen much innovation in it for 25 years from the old incumbent companies that we all know, and with the growing complexity of home networks, we now need a new approach to deal with new threats/ That is exactly what Steel Mountain offers.

 

What makes the technology behind Steel Mountain unique compared to other approaches, like typical firewalls or anti-virus programs?

We deal with the threat at the gateway of the home network with a device that plugs into your existing home router. This means we do not slow any of your devices down, it requires only one installation to protect everything and can provide security for hardware devices that cannot otherwise protect themselves, like smart fridges and lightbulbs.

Back when consumers were first buying PCs, there were not many points of entry for intrusions and exploits. Hacking 30 years ago was focused more on social engineering, exploiting human weaknesses, and less on “brute force” attacks, or finding a bunch of “zero-day” bugs (bugs that no one has reported yet), and using those to compromise a system.

But now you don’t have one processor and one piece of software running in your home. You have dozens and dozens of them, all of which can have vulnerabilities. Your fridge can be hacked. Your nanny-cam can be hacked. Your electrical system, your phone, computer, TV, and video game systems. There are many points of entry, and many potential points of failure now.

Plus those failures are more dangerous than they used to be. Now everything you do is digital, like your finances and even your health. So you have more to lose in an intrusion. A digital intrusion could have even worse consequences than someone physically getting access to your home.

So if the vulnerabilities are multiplying like this, security software just cannot keep up anymore. It’s like you’re building a bigger and bigger city, and you still rely on just one policeman to catch all the criminal activity. You should worry more about what your software can’t see at all, which is what is going on in all those devices you have.

For this, you need a single security layer that stands between your network and the outside world. That is what Steel Mountain does: we are building a single defense point and channeling everything through it. That’s the only way to really provide protection today. This is like the difference between the local cops and the NSA. You need a much more in-depth security system today.

 

Steel Mountain

 

What are some of the dangers most consumers don’t really understand when it comes to home security in a connected world?

That they are currently completely exposed to cyber theft and monitoring if they have IoT in their home or even an android smartphone.

I’m not just trying to be alarmist here. It is this simple. Everything right now is exposed, and nobody, from the device manufacturers to the telcos, to the companies that provide the firmware for these devices, is keeping up with the multitude of threats to IoT devices.

You might think too that you don’t have IoT in your home, so you’re ok. But most of us already do. This technology is now in everything, sometimes whether you realize it or not. Today it’s just a light bulb or a fridge, or a baby monitor, which doesn’t seem so dangerous. But soon home electrical systems and kitchens, and our cars will be controlled this way too. So physical dangers will become more real as well.

 

Why do you think that despite all the publicity around high-profile data breaches and hacks, as well as digital home invasions, the average consumer is still under-protected?

The average consumer remains under-protected because current solutions are high friction and have no focus on user experience. We still see cyber-security as something geeky and obsessive, even though that idea is pretty out-of-date at this point.

It’s also about who designs and markets security products, which is to say: security obsessed people who aren’t always in touch with the average consumer.

A security product should be passive, non-intrusive and usable by anyone. Current solutions fail to make a complex security solution accessible and seem easy to install or interact with. Current security solutions always seem like “work,” but we think they should be as simple and unobtrusive as possible.

Security is traditional sold through fear mongering, and the problem in doing that is that people become resistant to it over time.

Steel Mountain Features

Even here we are talking about all the dangers, and the truth is that most people don’t respond to these triggers. What’s the lesson there? What do people respond to? 

So our view is that you have to appeal to people’s sense of responsibility and their more evolved sensibilities: taking responsibility for your family, doing your basic duty as a parent or a homeowner.

You lock your doors and check for mold and termites in your home, so the same logic should apply when you’re securing your network as well. It just has to be part of the basic package of having a safe home. Like child-locks on your medicine cabinets, or having your chimney swept once in awhile. It should be that simple.

So, people make a lot of these basic decisions without thinking too much. Do I buy the regular power strip, or the one with a surge protector? Our responsibility, as a security company, should be to make that decision obvious, and then to follow up on that, and make it simple enough that people stick with it.

 

Why do you think design is so important for a security company?

We are selling peace of mind, and in order to do that, convenience and user experience is imperative. True peace of mind in our case means you never need to think about it again after installation. (If you don’t want to.)

There is an interesting balancing act in designing something that should be easy to use and to forget about, but is also mission critical for your safety. You have to kind of make it visible, but at the same time non-threatening and friendly. This is why our approach is to create a kind of friendly hockey-puck type device that just sort of sits there and says: “I’m here… everything is fine.”

It’s sort of like something between a smoke detector and an Apple TV. It isn’t daunting or flashy, it’s just there, it’s solid, reliable, always on; nothing to worry about. In some ways this is just applying the same approach to digital security as has existed in physical security design for a long time. Simple, subtle, but not a toy.

 

You were previously located in Virginia, outside Washington DC, which is a natural hot-spot for digital security products. Why choose Prague and StartupYard as your next home?

We moved back to Europe because we saw an opportunity in the market. Europeans, generally speaking, are more privacy aware. That has been shown very much in the way regulations have evolved in Europe, now including the GDPR privacy regulations. People value privacy, and see it much more as a basic necessity than in other places.

Right now in the US, you have companies like Amazon convincing their customers to put cameras in their homes, so the company can literally watch their homes and their delivery people can enter people’s homes while they’re gone.

That is pretty surprising to many Europeans (and many Americans too). Maybe I’m crazy, but I don’t think Europeans will widely accept that kind of sacrifice of privacy in their homes for more convenience. I hope that they won’t, because it sets a scary precedent for the risks we’re willing to take with our personal security.

But also, Europeans understand that big companies are going to try to compromise our privacy even without us realizing it. These companies would like to listen to everything we say and watch everything we do, because there is money in that. But European culture more or less believes that the welfare of society is more important than the business benefit of intrusive tech.

That is why companies like Facebook and Google have so many more challenges in Europe with regulations and oversight, because they take that kind of thing much more seriously.

The reason we chose StartupYard is because they have a very relevant network which we are currently leveraging. We couldn’t be happier here.

As I don’t have to tell the Czech people reading this, Prague has a long history of security engineering and cyber-security businesses like Avast, AVG, Cognitive Security, etc. Czech businesspeople understand security and take it seriously, which makes our lives easier, and provides more opportunities.

Plus, Czechia has a global reputation for security knowledge and prowess. We have definitely seen that this is based in fact.

 

What would you say is the most important thing you’ve learned in your time at StartupYard? What was your hardest lesson? Have you grown as a company and as founders because of that experience?

This is my first company. Obviously, you have ideas about how easy it is to make deals, and negotiate with people based on your common interests. When you really believe in your ideas, it feels like you can convince anyone you’re right.

One thing I’ve really appreciated in the process of mentorship at StartupYard, and then talking with corporations and partners about future plans, is that corporations are not one entity at all. They are a collection of different power centers and objectives, mixed with a lot of personal and political motivations as well.

When you start meeting with these companies, you realize that you have a lot of work to do to show many different people that it is in their interest to help you. Having people sponsoring you in a corporation is great, but it only takes one person in a position to block you, and you’re stuck.

As a startup, you move fast every day, and you don’t ask for permission. That’s how you build something new and different, but at the end of the day, people from the older world of business have to buy into what you’re doing. You have to respect the fact that these people have been in the market much longer than you, and they know a lot of things about it that you don’t.

I hope, and I think, that we have become more humble in that regard.

 

 

What can we expect from Steel Mountain in the near term? When will people be able to buy your products and use them in their homes?

Steel Mountain will be making it’s flagship product, Secaura, available for pre-ordering from March 2018. Subscribe to our website and we will keep you updated.

We have many details to take care of between now and March, and we are tackling the complex process of manufacturing and distributing the hardware, as well as maintaining the security service that comes with it.

Our aim is to be a part of people’s homes for the foreseeable future, so it’s critical that we build everything now on a solid foundation. Security and privacy is very dependent on trust, so we are building partnerships with trusted companies that have a good track record of delivering on their promises.

Exclusive: Cyber-Security Guru Vlastimil Klima Talks Blockchain and Cryptelo

This week I sat down with the renowned cyber-security expert and co-founder of StartupYard alumni company Cryptelo, to talk about a topic we’ve covered a lot lately: blockchain, and security. 

We have informally dubbed Cryptelo. “The Unbreakable Dropbox.” You can also check out a previous interview with his fellow co-founder and CEO at Cryptelo, Martin Baros, or visit their website to learn more about their products.

Hi Vlasta, tell us a bit about your involvement with Cryptelo. How did you and Martin Baros start working together?

Martin actually came to me after encountering security issues himself. He wanted to create a secure storage solution that didn’t exist anywhere on the market.

As an entrepreneur, he had a natural instinct that caused him to seek me out. When he proposed the idea, I realized: “yes! Why hasn’t anyone done this?” I joined as his chief cryptographer, and together we built Cryptelo.

Cryptelo Co-Founder Vastimil Klíma

You have a fascinating background in Cyber-Security, and are named among the top cryptologists in the world. You’ve worked for the Czech government, and you were among the few to seriously break SSL as a whitehat. What drew you to cybersecurity?

As a little boy, I was a very good chess player and a mathematician in high school. I also took part in the International Mathematical Olympiad. Later, I learned/realized that since that point I had been watched by the “head hunters” of the secret service.

That sounds like something from the movies, but it really happens!

Once I graduated with a mathematics degree from Charles University, I ended up working for the state, in a secret department for censorship and cipher development. As I discovered later, there are many great mathematicians and participants in the international mathematical Olympiads working in the secret services of the various states.

One of the big attractions for somebody like me to this kind of work is the opportunity to solve very complex problems that no one else has done before. You have a sense of tackling the unknown, which is very rewarding.

In my work I dealt with the development of cipher and cryptographic devices as well as cryptanalysis. Later I was also in charge of the ciphers for our agents abroad. After the Velvet Revolution in 1989, I was entrusted with the development of ciphers independent of the Soviet Union.

For almost two years I worked for the General Staff of the Czech Army, and then I went to the private sector. The pearl in my story is that I did my first private-sector job together with Eduard Kucera and Pavel Baudis (nowadays Avast’s vice-presidents) for their company, which is now among the top antivirus companies in the world. I’m quite proud of that.

Then a number of security companies followed, for which I developed different cryptographic products or did security and cryptological analysis or cryptographic designs. Some time ago I worked for the Czech National Security Authority on the design of cipher and cryptographic devices already in operation for five years. I was very fortunate to have always been able to work with the most advanced technologies or even the “upcoming” technologies, both in cryptanalysis and in cryptography.

Let’s talk about blockchain. Today it’s often described as highly secure. As an expert, what is your view on this?

The “blockchain” concept is very good and very safe compared to other [data verification] concepts. It is based on distributed security and responsibility, which is great.

But it’s just one building block in the whole system. Much depends on the other parts of the system. Surely you remember the lesson that an attacker chooses the weakest link in the chain. In security, you are only ever as good as that weakest link.

 Vlastimil Klima, Cryptelo

Why is it that despite the integrity of bitcoin’s ledger, there are still so many bitcoin heists and thefts?

Bitcoins are based on the blockchain principle, but paying with them requires the protection of cryptographic keys. In all major world bitcoin thefts, these keys have been stolen. The thieves then simply transferred the bitcoins to their bitcoin accounts.

So this is something like building the most secure safe in the world, with keys impossible to copy and locks impossible to crack, but then having it breached by the thief simply taking the keys off your desk. The whole concept of the unbreakable safe is not much good if getting into it is so easy.

Let us note that there has been a shift in our collective understanding of security – we are not talking about cryptographic techniques, but only about keys, their creation, distribution and protection. In many respects, we have figured out cryptography quite well. Information can easily be made very secure in terms of encryption. But that does not mean we have “solved security.” Far from it.

People think of Bitcoin and other cryptocurrencies as anonymous. Is that a mistake?

Bitcoins may be anonymous, but they may not.

The advantage of bitcoins and other blockchain-based coins is that transactions with these coins can be verified. For the same reason, it is possible to see how the coins “travel” on different “wallets”.

If someone makes a mistake, you can determine who they are, and what they bought for bitcoins.

I worked as a forensic expert on investigating several bitcoin thefts involving illegal drugs and arms markets, and managed to prove who controlled the marketplace and who stole the bitcoins. These are not truly anonymous platforms.

If I’m a regular guy wanting to buy crypto-coins of any kind, how can I protect myself from theft?

Every security breach up till now has consisted of theft of cryptographic keys, which were inadequately protected.

Here comes the simple advice: protect your keys and do not give them to anybody else. At big markets and shops, it is common that you have to give them their keys to make deals for you. Here you have to be very careful, because the purses of the big stores are the most threatened. Just give them small amounts or at best trade peer to peer.

As a cryptologist, what are one or two ways you wish every software company would think differently about data security?

This is very difficult. We all do just what we have to do. It is natural that we do not perceive security as important until we become a victim of a security incident. I have experienced this myself, so I know what I’m talking about.

Most of the time, data security problems arise from a lack of time and money to do the work properly. And attackers choose just this kind of company to attack, because it is vulnerable. So the best defense against security breaches is to maintain a high standard – higher than your competitors.

Predators prey on the weak. As we say: the gazelle does not have to be faster than the cheetah, it simply has to be faster than the other gazelles.

SY Alum Decissio Uses AI to Accurately Predict StartupYard Investments

You may remember Decissio, a Batch 7 StartupYard alum that has been working on the “Jarvis for Investment Decision Making.” Earlier this year, the company announced its kick-off product, an intelligent dashboard for VC investors and Accelerators to evaluate and monitor companies they invest in.

Decissio aims to go beyond a typical investment dashboard by combining up-to-date company data with complex big-data based probability models and machine learning algorithms, helping investors to continuously evaluate their investment decisions.

As Decissio and founder Dite Gashi continues to gather data and build the company’s flagship SaaS product, they have focused on piloting their approach with small controlled experiments.

One such pilot has been in partnership with StartupYard. Decissio’s Mission: to process all of StartupYard’s applications for Batch 8, our latest batch starting next week, and deliver predictions on their success based on a variety of factors, including written applications, founder profiles, founder/market fit, and the current state of the company.  

Dite Gashi

Dite Gashi: Founder and CEO at Decissio

The numbers are in on this pilot, and they’re very promising. We’re not ready to stop reading applications or doing our own research just yet, but we’re now confident that Decissio can be a big part of making our application process better, fairer, and more efficient.

The following case-study is a co-production of Decissio and StartupYard, written by Dite Gashi, and Lloyd Waldo. A more detailed write up and analysis will appear shortly after publication at Decissio.com. For more info on the technology and related work, please visit Decissio.com.

Warning: This post is long and contains big words. Skip to the bottom for a bulleted Tl;Dr 

Good Small Decisions = Big Positive Outcomes

The StartupYard application process doesn’t happen all at once. It involves a long series of smaller decisions. Does a startup have a unique idea? Does it fit into our mentor group and experience? Do the founders have enough experience? Is there strong competition in the market?

Some decisions are even more granular: did the founder answer questions thoroughly and clearly? Were they responsive in detail?

Small details often reveal big trends. But a human mind isn’t set up to think in that direction. We aren’t programmed to carefully add up small decisions to make big ones. Enter Decissio, whose mission was to apply a machine-learning approach to small decisions we make in the application process, not to override the judgement and experience of our evaluators, but rather to augment it with important insights.

StartupYard Alum Decissio.com uses #AI to accurately predict future StartupYard startup investments... Click To Tweet

The Framework

An application to an accelerator consists of a relatively small data set. We have a written application, founder profiles (on LinkedIn), sometimes a website, and whatever has been written about the company online.

Rarely do we have hard financial data on the companies, in some cases because there is no company in existence, and so the founding team has no financial data to look at. Nor do we have much access to the IP teams are working on. We have to rely on what founders say, and what they have done in the past.

But a bunch of small data sets together make up a bigger data set. Decissio examined over 1300 previous applications to StartupYard, along with the rankings our evaluation committee has generated, and used that data as a benchmark for incoming applications.

They found a number of statistically significant trends in that data. Startups that were successful as applicants to StartupYard could be ranked point-by-point, according to the following framework:

  • A Completeness Score: how thoroughly the application is filled in, and with how much quality information.
  • Effort Score: The quality of the writing in the application, particularly the responsiveness of answers, and the scope and variety of detail provided.
  • Relatedness Score: how closely a founder’s profile and experience matches the content of the application
  • Founder Linkedin Score: The completeness and quality of a founder’s LinkedIn profile
  • Media Mentions: The number, quality, and sources of mentions of the company or product online, along with sentiment analysis
  • Money/Work/Revenue Generated: The ratio of previous investments and time spent on the project to real revenues (if any).
  • Spell Check

Believe it or not, Spell Check is powerfully predictive of application quality. Note to founders: always use Spell Check.

The Analysis

This is where the historical data from previous StartupYard applications comes in. While it’s not very useful to directly compare older applications to newer ones, because the topics and ideas in them are often so different, it is useful to weight the importance of the different factors in the framework according to their impact on previous decisions.

Furthermore, the final analysis includes proprietary algorithms by Decissio that can dynamically weight the outcomes for individual teams, based on cross-referencing between different data sets. For example: Decissio’s AI can adjust its expectations for the Effort Score, if the founders are experienced in marketing and sales, or have no such experience. Thus each team is examined according to its own merits, and not an evaluator’s less informed expectations.

As “calibration,” or maintaining consistency and fairness of scoring across a large number of applications is a significant problem with humans, Decissio can re-calibrate an evaluator’s judgement to keep them from penalizing teams for the wrong reasons. As the standardized testing field has long known, human scoring can be so inconsistent that a significant amount of scoring time (even up to half) must be devoted to calibration in some cases.

Since our evaluations involve multiple rounds with a Pass/Fail outcome, each examining more and more detailed information, highly predictive models can be built for an application that will make it through round 1. A less predictive but still strong model can be built for round 2, and a much less accurate, but still useful model can be built for round 3, and so on.

The chart below shows overall predictiveness of the approach over multiple rounds. StartupYard uses a “first past the post” system of ranking, where the ranking cutoff for each round is smaller. This means that in round one, 70-80% of applicants are rejected. In round two, just over 50% of the remaining applicants are rejected, and in round 3 (which are day-long in person interviews), only 20-30% are rejected.

Decissio False Negatives

None of Decissio’s bottom-ranked 63 startups were ultimately selected, meaning that virtually all of the first round of evaluations could be handed over to the AI, leaving a much smaller pool of applicants to evaluate, and allowing the human evaluators to use a much lower cutoff, in a smaller, better initial pool. In this scenario, only 20% of human evaluated startups would need to be rejected in the first round.

We would expect false negatives to rise, as Decissio gets only one pass at the data, and with each round, human evaluators gather more data, which causes their behavior to diverge from the model.

For example, if use of Spell Check is 90% predictive of the Pass/Fail rate for round 1, it may be only slightly predictive of the success rate of round 2, and by round 3, it may lose its predictive power altogether. By the time an application involves a detailed look at a founder’s CV, and personal interviews with that person, other factors can arise that vastly outweigh any minor inattention to detail, like spelling.

Or the predictiveness curve can go in the other direction as well, with certain data only gaining predictive power in later rounds. Media mentions may have a low predictive power in the earlier rounds, and become more powerful later on. This can be because a company with a low early round score for Relatedness or very high Money/Work/Revenue ratios, can have many mentions in the media, but also fatal problems in their business, team, or technology. Thus, hype is not strongly predictive in Round 1, but by Round 3, it becomes a major asset to an applicant. Once all other factors are examined, media exposure becomes an affirmation of market fit, demand, or interest.

How Well Does This Work?

Decissio’s Success rate in the first round of applications (the on-paper evaluations), was 73%, far exceeding random chance. The accuracy dropped as expected in subsequent rounds where evaluations focused on personal interviews, from 50% in the 2nd round, to 20% in the final round. Still, this means that exactly half the time, a startup that passed the first interview with our selection committee was predicted to do so by Decissio, based only on their written application and profile.

There are two ways in which this kind of analysis can be useful. Either it can be used to identify applications that have a high likelihood of success, or it can be used to filter out those with the lowest likelihood of success.

Decissio Picked the Top 2 Ranking Finalists

We don’t have enough data to be able to confidently say that an application will definitely fail. However, on the opposite side of the scale, the results from Decissio’s analysis did correctly identify StartupYard’s two highest human-ranked finalists, and placed both in its own independent top ten prediction.

Decissio Picked the 100 lowest-rated applications with 89% Accuracy.

Still, the most immediate benefit of Decissio’s approach is in the earliest rounds, where pass/fail decisions are by design based on less human-focused information than the pass/fail decisions in later rounds.

This theory holds up with Decissio’s results: their bottom 100 applicants in this pool of applications (out of around 130), was 89% accurate, meaning that only 11% of the time, we determined a startup to be worth advancing, while Decissio did not. Clearly, in terms of identifying a lack of potential, Decissio’s approach is already very effective.  

Further mining of the available data could produce a much more precise prediction. For example, by analysing co-founder and founder/investor fit according to the work histories and digital footprints of both can theoretically yield very reliable predictions of compatibility, which in turn raises the chances of success or failure for a startup.

These factors would require a different kind of data to solve; a kind of data we don’t collect systematically right now. But this kind of approach, which treats people as nodes in a system that has its own features beyond those of individuals, has been deeply developed already, particularly on the level of enterprise management consulting involving things like the Meyers-Briggs Type Indicator Test.

It may prove true in the future that a set of personality tests of some kind are more predictive of success in a particular accelerator program or industry, than the content of an application, though we don’t know what that test would look like, or how it would be used.

SY Alum Decissio.com predicts first round StartupYard application decisions with 89% accuracy, picks two finalists. Click To Tweet

 

Potential Applications:

Time Saving

Decissio was able to predict with strong accuracy (73%), the likelihood that a startup would make it through the first round. This means that evaluator’s mental resources can be focused more on rounds in which more human-level data is being examined, particularly personal interviews and meetings.

An evaluator can spend relatively less time making early-round decisions, because Decissio can compare cursory evaluator consensus to its own scores, and “call out” the circumstances in which these do not match for further study. There is less of a chance that a good application will be “overlooked” in this way– a constant fear among startup investors dealing with many applications.

Bias Reduction

While a human with experience can “skim” an application and be able to tell it isn’t strong, that subjective evaluation is highly prone to error and internal biases. Very poor spelling could cause a human evaluator to give up on an application, whereas an algorithm might see past this issue and find more value in the startup than a person would look for.

This process could also serve as a check against more latent biases, such as gender, age, nationality, and sexual orientation. While it’s difficult for a human to differentiate between their instinctive reactions to people based on conditioning, and their objective evaluations of people in a professional context, an algorithm can demonstrate more consistency in that regard. Biases can’t be eliminated even this way, but they can be better controlled.

Thus, Decissio can be a check against the human decision making process, enhancing it without replacing it.

Fighting the “Best Horse” Problem

Decissio’s approach can also serve to fight the “best horse” problem, whereby a candidate with a strong outward appearance can advance well into the selection process without revealing sometimes severe deficiencies.

The best horse problem is one of reinforced selection bias. Imagine you have 10 horses, and you send them all running around a track. Then judging by the outcome of the test, you give special care and attention to the fastest horse, believing that it above the others has greater potential as a champion.

In this way we sometimes pick winners for all the wrong reasons. The horse to finish first can finish first for a number of reasons not having to do with potential as a racehorse. Cheating for example, or luck. Likewise, the last horse around the track can be the one with the most future potential.

In our application process, a very strong written application or interview performance can mask a basic weakness in the founding team’s experience or ability. It’s only much later that these weaknesses reveal themselves in a lack of tangible results from the company.

Startups can and do advance very far in accelerator programs while still lacking the core abilities and disposition needed to thrive. It can take a long time to recognize a fraud or a fish out of water.

Creating More Useful Feedback

Another thing this big data approach can solve is the information problem. What happens frequently with accelerator applications, as we suspect happens in many fields, is that successful written applications contain a near-perfect mix of description and data. Something like the “golden ratio” often described in mathematical analyses of artworks and natural proportionality.

The human mind likes a certain level of balance in the information it receives. When a person writes, they tend to favor either information or analysis, but only experienced writers know how to mix the two into pleasing and easy to read narratives. It’s a problem even good writers frequently struggle with. 

Too much writing about ideas, and the application seems too “light.” Too much data, and it seems too dense or too technical. In formal writing analysis, this formula is often used to describe balance between facts and ideas, where the value a is descriptive and creative writing, while b is supporting data and factual information. Those familiar with the classic “5 paragraph essay” often taught in schools, will recall the same proportionality. About 3 parts of persuasive writing, for every 1 part of factual basis. 

This type of training is not universal even among professionals, which sets up an arbitrary test of writing skill that may not be as relevant to the outcome as we tend to believe. If our job is to train people how to be better entrepreneurs, then we fail at that mission from the beginning if we can’t differentiate between someone who deserves our help, and someone who doesn’t.

By offering feedback on the strength of an application according to the above mentioned metrics (Completeness, Effort, Spelling, etc), Decissio could potentially improve the chances of failing applications where the main problem is poor writing.

An opportunity to improve an application is also an opportunity for us to see value where it is hard to spot. Telling an applicant that their application is failing because of style and substance can help those applicants to better express themselves, and thus deliver us more opportunities to find quality teams.

Conclusions

StartupYard and Decissio pilot project shows that AI assisted investing can improve results quickly. Click To Tweet

The results of this pilot clearly show that there is great potential in enhancing our decision making process with machine learning and data analysis.

We are not at the point where we’re ready to let a machine determine our investment strategies on its own- the way machines already do some forms of investing without human inputs.

Unlike an investor in securities, or a high-frequency bond trader, an accelerator’s main advantages are as a first mover. We invest in companies that don’t exist yet, have limited information on their markets, and have a limited history, or no history. So we invest in people – and people are inherently hard to quantify.

Our anecdotal experience of meeting teams in person *before* evaluating their applications, consistently reveals that the application process cannot identify many important personality traits. For an accelerator, success comes only when we are right about a trend, and a particular person, at just the right time.

So employing an AI powered decision-making approach cannot mean abandoning the unique advantages we have: the ability to see things others don’t see. Expertise (and hard work) is still the core of sound early-stage investing, but AI can help us to focus that expertise on the “creme de la creme” of potential investments.

It can save us from becoming jaded by the junk applications that routinely swamp our inboxes.

A startup is not an individual, it’s a team. And it is not in our interest to arbitrarily eliminate applicants who are not good at writing applications, or have other deficiencies more visible on an application than in real life. However, it is in our interest to conserve and spend our resources (including our time and energy), where the potential for gain is highest. 

This approach can benefit higher-dollar investors too: later stage investors have many of the same problems accelerators have, but on a different scale. A Seed or Series A investor makes decisions involving 10-50x more money than any single investment from an accelerator, and they also receive more requests, on average, than a small accelerator does.

Currently the most obvious and most immediate advantage of using Decissio’s AI is for very early stage investors with many applicants, such as government innovation programs, and big accelerators like TechStars, Y-Combinator, and 500 Startups. 

Tl;dr:

  • StartupYard alum Decissio analyzed our past applications over a 6 year period.
  • Decissio used this data and their own AI to predict which applications to StartupYard would succeed.
  • Two of their top 10 picks were also StartupYard finalists
  • They accurately predicted the bottom ranked half of applicants.
  • This approach can be used by accelerators to:
    • Improve applications overall
    • Save time on the poorest applications
    • Reduce systemic biases
    • Get better information on applicants
  • Decissio’s AI could be applied to other early stage investors, such as Series A and Seed Investors, or to large accelerators, particularly Tech Stars, Y-C, and 500 Startups.
  • At the end of the day, AI will help early-stage investors to get better information, and spend more time focusing on the human-focused side of their work.
Central Europe Accelerator

11 Things We Say All the Time to Startups

11 Things We say ALL THE TIME to Startups

“You Just have to…”

Paul Graham has an amazing post on his blog, called “Before the Startup.” You should read it. I’ll wait.


Ok, for those who haven’t read it: he talks about how his role as a startup mentor is often just to repeat the same things. After a while, he realized that the problem wasn’t that startups didn’t know things- it was that they were asking the wrong kinds of questions. And they were doing it because they’d been trained in life and education to do it that way.

Instead of asking “what do you think about…” startup founders ask: “how do I…” They do this because the education system and tech culture itself tell them that there are “secret answers” or “key learnings” that apply to almost any situation. Like the college student who asks if a piece of information will be on the test, startups look for “tricks,” asking what they should be doing, instead of asking for mentors to react to what they’ve actually done.

We have much the same experience at StartupYard, and so we thought it would be useful to break down those things we say so often, and explain why it is we say them.

For this piece, 11 Things We Say All the Time to Startups, there will be two contributors: StartupYard Managing Director Cedric Maloux, and Community manager Lloyd Waldo. The original version of this post appeared on our blog in March, 2016.

“It’s Not About You”

Cedric: Startup founders tend to focus, particularly at an early stage, on what they want, and what kind of company they want to be, instead of the problems that they will solve for their users. When they first start talking to their customers, they will talk about “we,” and “us,” instead of “you,” and “our customers.”

So it’s almost always necessary to refocus your messaging early on to make sure that you’re focusing on your customer’s problems, and are bringing them something of value, not just attaining a goal that you have as a company. Less “we need your support,” and more: “you need this product.”

“You Are Not Your Customers.”

Lloyd: Banish this aphorism from your speaking vocabulary. You created a business and risked everything to run a startup on the strength of one idea. You are not like the people who will be your customers. You may know a lot about them, and you may even use your product, but you also created it yourself. That does not give you an excuse to not talk to users, and try to understand them better than you do.

“No One Will Believe Your Projections”

Cedric: When we talk about projections (user growth, revenue), founders can get too caught up in how to make projections that investors might believe in.

But that’s backwards. Investors will never believe in projections, because they are just that- projections. Instead, you need to develop a plan that makes those projections seem attainable. Investors don’t invest in your projections- they invest in your plan, and if that plan makes sense, it doesn’t matter whether the projections are believable or not.

“How Will This Help You Grow?”

Cedric meeting with StartupYard Startups in 2015

Lloyd: Startups come across a lot of ideas about things that might help them grow. It’s important to keep in mind the goal of doing anything connected with so called “growth hacking,” which is to actually grow.

Vanity statistics, like Facebook likes and Twitter followers, are not growth (not alone). But the logic often goes like this: Step 1: Likes on Facebook (or whatever), Step 2: …? Step 3: Growth. Focusing on how something will lead to growth is important- you can’t go from step 1 to step 3 without taking step 2. So what is step 2?

“What’s the Next Step?”

Cedric: Just like startups have to focus on how doing things will help them grow, they have to also make sure that every step they take has something after it. Everything has a desired result. You got a meeting with a potential partner? Great. What’s the next step? If there is no next step, then what will that meeting accomplish? What will that partnership accomplish without a clearly stated goal?

“Where’s the Call to Action?”

Lloyd: Simply put, you shouldn’t be communicating with your customers if you aren’t giving them something to do, or something they value.

Startup founders usually get a sense that they have to be activating their customers, but they also have to activate them to do clear and understandable things. There also has to be a clear way of measuring whether that activation is actually working. Enter the Call to Action: if you don’t have one, in an email, a post on social media, or a landing page, then you are wasting your users’ time and attention for nothing.

“You’ll have to test this and see”

Lloyd: Founders are prone to confusing advice with directions. As a mentor, I can give good and actionable advice, but just because I think it will work, doesn’t mean it works. The only way to see if the advice is sound is to try it, and pay attention to the results. Testing can’t tell you everything, but not testing tells you nothing.

“I can’t hear you.”

Cedric: When you meet with employees, with investors, or with anybody, remember that you’re the founder of a company. Your opinion matters, and you need to be heard, loud and clear. Some founders just don’t know how to make themselves heard, and make their presence felt. Instead of owning the room and controlling the conversation, they react passively, and let others lead. Instead, be the boss, and say what you think in a clear, audible voice. You’re the boss.

“It’s Your Company.”

Lloyd: Leading from that, remember that whatever you’re doing, make sure you believe in it. In an accelerator, you get a lot of advice, and a lot of direction. But it’s your company. If you aren’t happy with things a certain way, then the last word is ultimately yours. You should listen, and be open, but you shouldn’t do things just because people tell you to. If something doesn’t feel right, ask for help, but don’t “go with the flow.” It’s your company.

“Stop Selling, and Start Creating Visions”

Cedric: Selling isn’t about getting money from people. It’s about giving them something they can believe in, and are willing to pay for. To sell in the long run, you have to build a vision that people can relate to, and that people want to be a part of. If you focus on your vision, and on communicating that vision to people, then the money conversations -the selling- are just a detail. A small part of the overall experience, and not the focus.

“You Need to Control This Process”

Cedric: B2B sales are very different from B2C sales. In B2B sales, you need to remain in the driver seat, not waiting for the customer to decide that they’d like to work with you. You need to own the process: move each separate piece like a conductor, anticipate every question and issue, and close the deal. Going at the pace that the client picks is, in effect, accepting the client’s own objections and doubts as your own. If you don’t set the pace, then no one will, and many deals that could happen just won’t.

Ouibring, StartupYard

SY Batch 7 Alum Ouibring Gains Investment – With a Twist

Good news often comes all at once. Yesterday we announced that Neuron Soundware had raised €600,000, and StartupYard has raised €1 million in a record breaking investment round.  Today we’re able to announce that Ouibring, a StartupYard company (Batch 7) that helps travelers and shoppers to bring joy into each other’s lives by bringing rare items home with them from abroad, has also raised seed investment.

StartupYard, Ouibring

OuiBring Founder and CEO Joel Gordon, signing a deal with Busyman.cz

The Details

The seed investment comes from the Czech incubator Busyman.cz.

Since joining StartupYard in late 2016, Ouibring has quickly built a following of more than 60,000 Facebook fans. Filip Major, the founder of Busyman commented: “Ouibring has the potential to change the global consumer goods logistic system as UBER is changing the way people move”.

The investment will power global expansion, as Ouibring connects more of the 30 million flights carrying almost 1 billion travelers each year with shoppers all around the world.

Ouibring connects shoppers who need help sourcing hard to find products, and travelers with spare luggage capacity to create a win-win situation. On Ouibring’s platform it’s possible to order hard-to-find goods from your home country, or discover new items that travelers can then bring with them when they visit a city near you.

Ouibring, Startupyard

The Twist

Busyman.cz has acquired a minority stake in Ouibring using a digital commodity, “Crown,” which is a “non-pre-mined” digital currency.

Crown has a market cap of more than $13m USD, processes hundreds of transactions per day on its blockchain and provides powerful security features. As part of this deal, Ouibring will move its client-to-client transaction settlement onto the Crown blockchain, making every transaction easily trackable, efficient and transparent. Ouibring also aims to emit its own token of exchange on the Crown blockchain.

“Our customers care about security and compliance. Using the Crown blockchain to create unique new features will help make Ouibring even more reliable and easy to use for our customers” says Joel Gordon, CEO and founder of Ouibring.

About Ouibring:

In our interview with him earlier this year, Joel told us the story behind Ouibring as a new online shopping experience:

 

” The idea for Ouibring came from experiences gained living and working abroad for the last 15 years. The fun and excitement when a special package delivered by a friend arrives is the inspiration for Ouibring’s tagline – ‘Bring a little happiness’.

As any expatriate knows, living abroad can give you a special appreciation for things that those at home just take for granted. You look forward to that time when a friend will bring a special something you’ve requested from your home. That’s a magical feeling, as if you’re the only person in the world that has what you have. We wanted to capture that feeling, and make it something anyone could enjoy. A special moment of joy only for them; an experience no one else is having.

At the same time, we can give others the chance to make a bit of money, and reduce waste by sharing their spare luggage capacity.

One story I really like is how even a small, generic item that is plentiful in one location can provide a whole lot of pleasure and luxury when it appears in an unexpected context. When a Ouibringer arrived with three massive bags of Monster Munch Pickled Onion and delivered them to a travel blogger living in Bangkok. They really made her day.” 

Joel Gordon
Joel GordonCEO, Ouibring

 

What’s Next for OuiBring

CEO Joel Gordon moonlights as a user of his own product: here he delivers some treats to customers in Thailand.

Ouibring has already attracted hundreds of shoppers and travelers from around the world. The company is continuing to explore alternative approaches to shopping and fulfillment for adventurous people everywhere.

So, to celebrate this big step for the young company, why not jump over and order something for yourself, or sign up to bring a little happiness into someone else’s life?

You can now apply for StartupYard Batch #8.

  • Robots
  • Artificial Intelligence
  • VR/AR
  • IoT
  • Cryptography
  • Blockchain
Applications Open: Now
Applications Close: June 30th, 2017
Program starts: September 4th, 2017
Program ends: December 1st, 2017
Pavel Konecny, Neuron Soundware, StartupYard

SY Alum Neuron Soundware Closes €600K Investment from J&T Ventures

We are very pleased to announce that Neuron Soundware, or 2016 Alum, and winner of “Vodafone Idea of the Year 2016” has closed an investment from Prague-based J&T Ventures, of €600,000 to grow their team and expand their sales to capitalize on early traction with clients like Siemens.

The story broke first on Euro.cz this morning.

Congratulations @startupyard alum @NeuronSW on exciting progress, and fundraising €600K to expand operations! Click To Tweet

The Details

Pavel Konecny, Co-founder and CEO of Neuron Soundware, made the announcement today in Mlada Fronta, together with Adam Kocik, Managing Director of J&T Ventures The investment will help Neuron Soundware to beef up its team, refine its technology, and expand its customer reach to include aerospace manufacturers, rail operators, and automotive companies.

Neuron Soundware, founded in 2016, joined StartupYard the same year. There the founding team, a group of AI experts led by Konecny, conceived of a device which can listen to heavy machinery, and over time, learn to recognize mechanical issues and predict when the machinery is likely to fail. Since attending StartupYard, they have developed a device employing high-end sensors used in aerospace, and audio processing software that can be plugged directly into heavy machinery and can warn of future mechanical problems. The company announced a cooperation with Siemens in 2016, and was invited to join the Airbus Innovation Lab the same year.

“We are continually impressed by the Neuron Soundware team’s technical prowess and ability to attack very complex problem sets with novel approaches and technology,” Kocik commented on the investment, “this technology is going to be even more essential as the IoT [Internet of Things] matures. Neuron Soundware will help to make machines safer, more efficient, and longer lasting.” The investment, a cooperation between J&T Ventures and a private investor, will be used to refine the engineering of Neuron Soundware’s physical devices and software, and to support its outreach to large industrial machinery firms, where demand for the technology is already growing.

Neuron Soundware, StartupYard Accelerator

According to Konecny, the technology, based on “deep neural networks,” learns from the sounds machinery produces, and can detect patterns too faint or complex for a human to hear, diagnosing issues with machinery well before they become catastrophic. Konecny says of the technology: “Sound is a rich source of data, and also quite universal, which is why mechanics and engineers rely on it so much. But a human cannot listen to 100 airplane or diesel engines for 1000 hours each, and make sense of it all. A machine can do this, and when one engine fails, it can apply that learning to all it has already heard, thus greatly enhancing our ability to detect and prevent future problems.”

“When Neuron Soundware joined us for our 6th program [out of 8], their approach to understanding sound had never really been tried before,” commented Cedric Maloux, our CEO, “leveraging StartupYard’s mentor network, locally and abroad, they were able to very quickly prove that there was a huge need for this kind of technology.” The company notes that future applications for machine learning and sound reach beyond machine maintenance, to product testing, autonomous navigation, green energy solutions, and even security. “Sound is everywhere,” remarks Konecny, “and we’ve just started to see how we can use it to understand more of how everything works.”

About Neuron Soundware:

Neuron Soundware is a deep tech startup, exploring the use of self-teaching, constantly learning neural networks in a wide range of audio analysis and audio manipulation applications. Since 2016, Neuron Soundware has focused on technology to monitor and diagnose industrial equipment to predict failures and increase efficiency. They include Siemens and a number of other leading industrial and transportation equipment manufacturers among their clients.

About J&T Ventures:

J&T Ventures is a Venture Capital fund based in Prague. The fund invests up to €500 000 in technology firms at the seed stage in CEE region. Since 2014, J&T Ventures has been invested in 11 growing and promising innovative startups with the goal to contribute to their dynamic growth and value creation. The fund focuses mainly on B2B sector with a particular interest in FinTech, IT (Big Data Analytics), IoT/IoE & Smart City IoT and Retail.

 

You can now apply for StartupYard Batch #8.

  • Robots
  • Artificial Intelligence
  • VR/AR
  • IoT
  • Cryptography
  • Blockchain
Applications Open: Now
Applications Close: June 30th, 2017
Program starts: September 4th, 2017
Program ends: December 1st, 2017
Rossum, Startupyard

Exclusive Interview: Rossum – AI for Documents

Rossum is a deep-tech startup that joined StartupYard with an understanding of the technology they were developing, but no clear use-case for that technology. The team, a group of PHD candidates in artificial intelligence and machine learning, were exploring applications for machine learning in routine tasks like image categorization.

Through the acceleration process, the Rossum team found opportunities in applying machine learning to corporate accounting, where automation technology has failed to make serious gains in productivity in recent years, due to the sheer complexity and variability of financial documentation. Even invoices alone, of which over 1 billion are sent each day worldwide, are highly resistant to automation due to technical limitations imposed due to historical factors. Rossum began to apply their technology to this problem, and have already made major breakthroughs with the help of a few early key partners.

I sat down this week with Co-Founders Tomas Gogar, and Petr Baudis to talk about Rossum:

Hi Tom, tell us a bit about Rossum. Where did the idea come from?

Tomas Gogar: Hi! The idea for Rossum dates back several years, and it started out as something very different. Our team is a group of PHD candidates who have been studying AI/Machine learning and semantic understanding for a long time.

My Co-Founder, Petr Baudis, is quite well known in AI circles for developing the leading open-source Question Answering AI, something similar to IBM’s own Watson. We actually know that some members of the IBM team periodically test his platform to see how it performs.

Petr also laid the groundwork for Google’s AlphaGo project (the AI that beat a world master in the game Go, many years before it had been predicted that a computer could beat a human). Google cited Petr in their landmark paper in the journal Nature.

The Rossum Team: Petr Baudis (left), and Tomas Gogar (middle), with Tomas Tunys (right)

The Rossum Team: Petr Baudis (left), and Tomas Gogar (middle), with Tomas Tunys (right)

I also have published groundbreaking research on the use of neural nets to parse and understand complex semantics in written text (such as semi-structured documents). It was actually this research that led us to the underlying idea behind Rossum.

Like everybody in this field, we’ve been fascinated by how neural networks and machine learning can be applied to massive volumes of data online. We started with a very popular objective, which is recognition of images and their contents, commonly called computer vision.

When we applied to StartupYard, we had the idea of providing a kind of “AI as a Service,” platform that could be used by researchers or data scientists to apply machine learning to their own problems. One of our innovations was being able to train a neural network on very limited data, which made it very useful for this purpose.

In mentoring though, we quickly discovered that there is actually a huge need for AI that understands and can parse text, in many different formats. We had assumed that problem had been solved by OCR (Optical Character Recognition), but in fact automation based on OCR alone has not really improved in effectiveness in many years. It remains a very hit and miss process.

You’re solving what seems like an unlikely problem- processing invoices. How can AI be applied to that problem?

Yes, it seemed unlikely to us as well! But what we found when we met with mentors from the big 4 accounting firms is that, even in 2017, invoice handling is largely still a manual process. This is despite OCR being around for 25 years already, and electronic invoices existing for over 50 years.

That’s mainly because despite all attempts to standardize and streamline invoicing (including newer techniques like using QR codes to allow machines to read them), it remains a problem that is really too big for any one player to solve. No government or company has been able to get enough others to agree on a single standard, and that has meant, effectively, that the complexity of processing invoices has not changed much for decades. It may actually have gotten worse, thanks to the increasing complexity of the products and services for which we receive invoices, and the increase in the total volume of invoices.

Rossum_Homepage

So just imagine you are an accounting company. You receive thousands of invoices a day. Maybe hundreds of thousands a year. If every invoice takes your accounting people just one minute to process and confirm, that’s hundreds of hours of work that needs to be done. And that is aside from the possibility that you are audited, and have to check all that work for a second time. The human error rate for invoice processing is low, but mistakes are very expensive. Fraud is surprisingly common as well, as bad actors take advantage of the complexities of billing to trick people into paying for non-existent things.

So we began to realize that one answer to this seemingly intractable problem was AI. The big hurdle for automating invoice processing is that no two invoices are ever exactly the same. Getting a machine to recognize one type of invoice, and correctly pull the right information out of it is non-trivial, but certainly possible with existing technology. The problem is that at any time, companies are receiving invoices from many, many parties, a company can change the way it formats invoices, or an invoice can contain a mistake or can be fraudulent.

So that means that no matter how streamlined your process may be, it still requires human-level judgement to confirm that everything is correct. That creates a bottleneck, because human-level judgement takes a long time to apply. People are very good at semantics, but we are also very slow. And you have the influence of many human factors: being tired, misreading a number, forgetting to double check.

Rossum: Human level judgement is slow. And most AI is dumb. Read the exclusive interview: Click To Tweet

AI that can read and understand an invoice at a human level can also be made to do the same work many, many times faster than a human. So if you train a neural network to be able to recognize and work with invoices it has never seen before (just as a human can do), then you can turn hundreds of hours of tedious work into a matter of a few seconds or minutes of computer processing time.

Who would use this technology? Why hasn’t it been developed already?

Well, neural networks are really just starting to be applied to these problems. One of the reasons this hasn’t been done already is that processing power and computer architecture hasn’t been powerful enough to make it possible. The other side of that equation is the data itself. Until recently, much of the data needed to train a neural network didn’t exist in a form that a network could actually handle. Invoices were on paper, or they were “electronic” and thus in a form that could be handled automatically by a hard-wired program.

In the past decade though, the volume of data that we can apply to machine learning has exploded. That’s why AI and machine learning have suddenly become hot topics again.

Rossum_technology

The other issue has been that the specific methodologies for training and checking the accuracy of neural networks have been evolving, and are just starting to become really useful for this kind of work. You can just show a million invoices to a neural network, but getting it to focus on what is important is not something you can just ask for. You have to be able to train the algorithms in ways that help it eliminate useless information and focus on what you want it to focus on.

The process is analogous to the way a baby learns. When a baby is born, its ability to sense information around it is very limited. It can’t see or hear very well, and it can’t process what it does see or hear. Slowly it becomes more able to sense information, and it begins to use that new information to construct an understanding of the world around it. Then comes language: the way that a human mind is able to abstract information and complexity, and imagine new things it has never seen before.

If we are to use this analogy, today neural networks are operating mostly blind, and with little understanding of language to create or understand abstract ideas. As we expose them to more information, we also have to teach them a “language” that they can use to extract something useful from that data. Rossum is a part of that language: we are helping neural networks to understand what we want them to do, and why.

The answer to who would use this technology, is, well, everyone! Human level judgement in understanding documents, even of only a very specific type, would save huge amounts of routine work for humans, who can spend that time doing things that are more natural to them. There is nothing natural for a person about spending their days processing invoices. We can learn it, but we never love to do it.

If you consider how many of these kinds of tasks exist, you realize that we spend massive amounts of time doing things that just don’t bring us anything of value, and in fact waste our time and demoralize us. That is the promise of AI in the near future: this ability to free people from having to deal with things that we just don’t get anything out of, but that just have to be done by someone.

What will be your near-term strategy for bringing Rossum’s technology to market?

As mentioned, we are already in contact with representatives of the major accounting firms. They have the biggest immediate need for Rossum, and they also have the data that Rossum needs to be able to train itself and understand invoices it has never seen before.

In the next few years, we want to have a platform that can understand and work with invoices from literally anywhere in the world, many times faster than a human can, with an error-rate lower than any human. Then the work becomes helping these companies to implement the solution, and find ways for Rossum to interact with other systems so that it can help companies streamline their document handling operations.

It is not much use to an accounting firm for an AI to understand all its invoices, if the AI doesn’t also know how to connect with their other systems, and give the outputs they need to take action. That in itself will be a challenge, and one we anticipate will take some time and development. Still, once you have the ability to process invoices with human or above-human accuracy and speed, then there is a huge incentive on the part of companies to integrate that solution into their systems.

We believe that Rossum will be a must-have for accounting firms in the very near future. And once that value has been clearly demonstrated, we can apply the technology to many other processes that are similar in nature. Auditing, analysis and processing of other documents, etc. Rossum could be a backbone for a suite of intelligent applications that takes care of a wide range of tasks that are complex, and repetitive.

We also want to open up Rossum as an online platform, to allow small and medium sized companies to find it and gain the same value from it. Currently, invoice processing services online come at a heavy cost- over a Euro per invoice in many cases. Rossum can do the same work for a tiny fraction of those costs, and it can do it instantaneously, with a very high degree of certainty.

An automatic platform solution is faster, cheaper, and safer for companies that have confidential information they don’t want others to have access to.

How has your experience been at StartupYard? What surprised you?

Petr Baudis:  We were rather hesitant about joining StartupYard actually, even though we received several personal recommendations from some of the top alumni founders. We were thinking “hey, we have an office and our own good network of contracts, does StartupYard make sense for us?” and we applied at the last minute and very tentatively.  

FullSizeRender 8

Tomas Gogar (right), works with another co-founder at the StartupYard voice workshop

Oh boy, we were in for a real ride when we did decide to join.  Mentoring sessions gave us a much wider scale of perspectives than we could ever gain from our own professional network, and a real and much needed shift of focus from the technical to business.  That we expected a little – but it surprised us how eager the core StartupYard team was to help with their experience and feedback, these few people (including you, Lloyd!) really became an important part of Rossum’s story.  

And most importantly, StartupYard finally gave us the impulse to really focus on one single thing – we were busy people before, but now we had the reason to finally drop all the side projects for good.  We thought the first mentoring month would be the most intense phase, but the pace is only picking up since, and without the “little” pushing by the StartupYard team we would be much more comfortable, getting a good eight hours of sleep a night, but still at the beginning.

Chatler: AI for Conversations at Scale

The past year has seen a boom in chatbots, which have become a buzzword in the tech industry, most particularly with retailers and big brands. StartupYard this year handled dozens of applications for chatbot startups, but despite the buzz, none of these seemed to us to have really discovered the inherent value of automating customer interaction on social media, and in customer care. Chatbots are not a new idea, after all, and much of the recent hype has come thanks to Facebook opening its platform for 3rd party developers, which has spurred renewed interest in new applications for chat.

Facebook’s strategy is bold, and we believe that it may yet yield positive net benefits for end-users and customers, but the chatbot arena is still very immature. And so Janos Szabo gained our attention immediately when we met him and his team during our roadshow, in Budapest last summer. Janos convinced us that the future of chatbots, at least for now, is not in full automation, but in AI integration. He founded Chatler late last year with a team of friends, including strong experience in brand management, to prove that the way forward for chat with big companies is in helping human beings to do what they do best:  be human,

I caught up with Janos this week to talk about his vision for Chatler. Here is our discussion:

Hi Janos, tell us a bit about Chatler. How did you come up with the idea?

I’ve worked a bit with chatbots. I was actually on a team that was creating them. But it quickly became apparent that chatbots aren’t the massive game-changer they’ve been billed as. They aren’t a “fire and forget” sort of thing. They need a lot of care, and a lot of time, and as a result they never end up being as good or as efficient as we hope they will be.

There’s a cycle to AI products in general. It starts with hype: a new way of doing something has its day in the sun, like chatbots, or language translation, and early demonstrations are incredibly promising. But when they’re released into the wild, these things just never work as well as we hope they will. They always have their limitations, and those limitations become apparent relatively fast.

Janos Szabo, CEO and Co-Founder of Chatler

Janos Szabo, CEO and Co-Founder of Chatler

In something closed and strictly defined like Chess, or Go, or even driving a car or landing a rocket on Mars, the rules are clear and the inputs are fairly simple to deal with. But humans are organic, and human conversation is chaotic. Subtlety and context that is easy for a human is incredibly difficult for a machine. I couldn’t land a rocket on Mars, but I can probably handle a customer interaction better than a bot that cost a billion dollars to make. So we have a long way to go.

The problem with chatbots today is the same as it was 20 years ago, when chatbots first were popularized. Some may remember Aimbots, that could answer simple questions according to a hardwired answer-tree. Essentially, modern chatbots work the same way, but pull information from more sources, and understand the intent of questions, perhaps a little better than they used to, when the only thing they really understood were keywords.

Still, 20 plus years, and chatbots can’t do basic things. They can’t judge a person’s reactions to what they say very well. They can’t tell if a person is annoyed, or is making a joke. They can’t think like we do. And they can’t improvise the way a person can.

You wouldn’t trust one to handle your high-value customers, and you might not even trust one to run your Twitter feed. We’ve seen from live experiments from companies like Microsoft that chatbots are still poorly understood. They created a Twitterbot that quickly became an anti-semitic Nazi sympathizer. Of course, it was just doing what it was programmed to do, but that was part of the problem. A chatbot, at the basic level, doesn’t have any sense of morality or of right and wrong. Even backed up by fairly sophisticated AI, it doesn’t have a conscience to guide its actions. It just reacts to inputs.

Now, AI is about learning. We can teach bots how to simulate our own sense of right and wrong. But that’s still on the horizon for modern technology. Today, we still very much need humans to be at the center of communications with other humans. So I started to realize that what we needed to do was, for now, let go of the idea of building bots that can be fully automated.

Chatler_insert_1

We need bots that can learn from us, by showing them how we interact with other people. In time, they can begin to take over tasks that are relatively simple. For example, answering a common question, or accessing an internal system to check a customer’s account information, and other routine tasks that bots should theoretically be very good for.

AI can do a lot of things better than a human can. Particularly routine tasks, but also more complex ones, if there is enough opportunity for the AI to learn them. That’s the assumption that Chatler is built upon. But humans aren’t going away. Until we build an AI that can really feel as we do, there will always be a need for real people.

You’ve actually been selected for StartupYard once before (but didn’t attend), as part of a now defunct startup. Tell us a bit about that experience, and what you learned from it.

Yes, I was one of the founders of ClipDis, which was a tool for message platforms. It was a really catchy product. You could type something in, and it would construct a short video of the sentence using clips from TV and movies. People loved it, and it was fun.

It was a great opportunity for us, but we didn’t fully understand our market, or how to create a real business out of the idea. You have to establish your fundamentals earlier, and spend time on coding later. We were a bit naive, and a bit arrogant, because we had such a cool product, and we thought we could just worry about getting lots of users, and take care of the business part later.

But, especially in Europe, that just isn’t something investors want to be involved with. A hard lesson for us, and something I’ve taken with me to Chatler. I’m still deeply interested in the way people message each other, and how we can understand and enhance our ability to effectively communicate. That has not changed.

What is the market getting wrong about chat, and how does Chatler get it right?

The market right now knows that chat, in some form, is a huge part of the future when it comes not only to customer care, but also to sales and even business development, research, and other fields. So there has been a rush into this space, starting all the way back with Facebook’s acquisition of WhatsApp 4 years ago.

Chatbots seem like an elegant solution to the problem of chatting with your customers at a large scale, but as I’ve noted, they have some really basic limitations. They are good at tasks, but in authentic conversations, they are totally lost. That’s why I see Human/AI collaboration as a key step forward in chat technology. If chat is indeed going to dominate as a future medium for customer communication, then AI is going to play a big part. We just don’t believe it will be the driving force.

Chatler helps big companies and brands make their chat channels more efficient, more responsive, and easier for their customer care and sales agents to use. It does this by learning from the way real people talk to customers, and teaching itself to do routine tasks that real people do. For example, Chatler will quickly learn the best answers to common questions: “What time are you open?” It will then be able to progress to more complex topics: “How do I get to your offices from my home?”

Chatler_insert_2

Chatler will be there suggesting answers for a human to decide upon and use. But eventually, Chatler is going to make that human better at giving answers. Humans make dumb mistakes, like typing a wrong letter, or misreading a map. An AI doesn’t make those kinds of mistakes, so the two working together create a better overall customer experience. Fewer mistakes, faster responses, and better information.

There are a few paths we can take from there. We can more fully develop a recommendation engine that can be applied horizontally to different modes of communication, and/or, we can integrate Chatler more deeply with a company’s operations, enabling it to help a chat agent to accomplish tasks that would take a person a lot of time. Things like looking through databases to find one particular item, or checking multiple sources for one key piece of information. An AI can do in a moment what it can take a person hours to do on their own.

For example, an AI could search your entire communication history with a company instantly, and find out exactly what problems you have had in the past, while a human would take a long time to find and understand all that information. We think humans should be left to do what they are best at- which is caring about other people.

Humans have emotional intelligence. And machines don’t. Certainly not yet, anyway. So the goal shouldn’t be to fake emotional experiences, but to enhance a human’s ability to focus on what they’re best at.

Humans should focus on what they are best at: emotional intelligence. @ChatlerAI Click To Tweet

Why are big brands so interested in moving their customer care operations to a chat format?

A few reasons. FIrst of all, chat is where people are now, particularly younger people. People under 30 now rarely talk on the phone, and school-aged kids today essentially never do, unless forced to. They now see voice-calling as invasive. Some people don’t answer voice calls anymore, directing people to message them instead.

Messaging has become a very visual and creative medium thanks to innovations from Snap, Facebook, and others. It has become a prefered way of communicating between friends and family, and even with some businesses. It’s convenient, and flexible, not requiring two people to pay complete attention to a conversation at one time, and creating an easy record of what’s been said.

A messenger code. Scan it to talk to Chatler.

A Facebook messenger code. Scan it to talk to Chatler.

And for those who remember the early days of ubiquitous mobile phones, chat is more private, and it is also less intrusive to people around you.

Secondly, it’s cheap. This is partly why chat became so popular to begin with, because sending electronic messages uses less data or “calling minutes” than making a call or using VOIP. A customer service person can also service multiple inquiries at once using chat- something that a voice-call center can simply not do. That means that fewer overall people are needed, because the attention of a chat agent is more divisible.

It’s much more important to have more agents available when there are a lot of customers contacting you. And it’s much simpler technologically- without the need for expensive equipment, phone switches, and complex phone-trees and recording equipment.

There is a universe of Customer Relationship Management (CRM) tools and platforms out there. How does Chatler fit into the broader market for these products?

Chatler is focused on chat-based CRM. It will take the form of an end-user solution, but we also plan to offer it as an API for existing CRM providers. That way, if a company has invested deeply in a system that closely matches their needs, Chatler will be easy to integrate in order to make that system more effective and efficient.

Chatler is data driven. Many CRM systems, even those designed for chat, are driven mainly by solutions to logistical issues. We don’t plan to disrupt that, and we don’t pretend to be able to completely replace existing platforms. Instead, we’ll bring a data focus that will help companies in any stage of their chat development. In many cases, our aim is for Chatler shorten a 3-step process into a one-click process, or a 10 step process into a 3 step one, Moreover, Chatler will learn continuously, meaning that what it can’t do today, it can learn to do tomorrow.

2_TEST_Topics_mo

Chatler understands, customer interactions,and recommends effective answers and solutions.

For companies that have never used chat, they can start with Chatler. For companies that are deeply invested in chat already, even better. Chatler will seamlessly enter the flow of existing systems, and save chat agents time and stress throughout the day.

You’ve recently worked on a set of trials with real brands. What did you learn from this experience?

Well, I will be diplomatic and not describe the processes that we’ve observed. We can say that some of them are archaic. Copy paste and spreadsheets are not uncommon. In many ways, the problem is typically that big companies haven’t taken chat seriously from the beginning, and now demand has seriously outweighed their ability to leverage chat. This keeps many brands from even using chat, and it keeps others from using it effectively, or investing in it further.  

Companies are just coming around to the reality that chat isn’t going away, and that it will have some fundamental effects on the way they do business. But that’s just starting. As they did when social media first appeared and changed the way businesses communicate, big companies are constantly weighing the risk and the reward of diving into a new form of communication.

That has helped us to understand what companies need from us. And often what they need is someone to tell them that there is a place to start- a way to make chat an organic part of your business, just as telephones became a part of business a century ago, and the internet became a part of it 30 years ago.

How do you see Chatler growing within the next year? Who are the obvious first customers?

As a standalone platform, SMEs can use Chatler as their primary solution for chat. It will always be better than a native chat platform, because it will be dedicated to learning exactly how your business, and only your business runs, and what kinds of things you communicate, and how you do it. This will extend to mobile as well, as more and more, customers expect to be services on mobile chat.

Large enterprises will integrate Chatler with their existing CRM platforms, and for that we’ll need a global sales team, and a richer, fuller analytics product that helps companies to understand the value that chat is bringing them, and gain actionable insights on their whole chat-oriented customer care operation.

How do you think Chatler will play a role in the further future, 3-5 years from now?

Eventually, consumers will come to expect a chat experience from businesses that mirrors the experience we currently have of the web. Chat will be highly automated, and “explorable,” working together with the customer to solve their needs and find new opportunities for them.

We won’t get to that state by building chatbots. No amount of planning can really tell you how people will use chat. Instead, it will be based on an accumulation of experience, reflected in the complexity of the machine learning algorithms behind Chatler. Over time, Chatler’s AI will become adept at learning about a new company, and finding ways to help take over tasks that it can automate, and interactions that it can handle on its own, then only referring to humans when it is unsure of how to proceed.

This has deeper implications than just making businesses better at chat. Websites and social media both also transformed the activities that businesses actually engage in. It affects the products they create, and the way they do business. Chat has that same potential. The data and experience that Chatler’s AI will gain from conversing with real people will translate to actionable business data, helping companies make better decisions with products, services, customer relations, and more.

Chat will become an opportunity, rather than a liability.

Tell us a bit about your team. How did you all start working together?

We know each other from our previous startup projects, years spent at multinational companies,  agencies, and several meetups.

András Reder has 11 years of experience at Coca-Cola,  leading cross functional teams in a matrix organization. He launched over 17 products/brands into international markets with Coke. Having seen the industry from the inside, András wanted to challenge himself in new projects where A multinational background doesn’t necessarily mean success. He brings the business oriented mindset into our team and makes sure that we all line up from operational POV. 

IMG_6021

The chatler team: Andras, Andris, and Janos

Andris is a freshly graduated coder and bot enthusiast. He organizes the Chatbot meetup in Budapest, supported by Microsoft Hungary. We met at one of his meetups and we immediately jumped into a passionate discussion about chatbots and the future of chat communication. It was an obvious fit for the team. Beyond coding he is also contributing heavily to conversational UX related topics.

With Gábor, Bence and János, we created one of the first chatbots for KIK Messenger’s global bot shop launch (other launch partners were Vine, H&M, WeatherApp). This was the time we started to experiment with recommendation engines, and we realized that even the best algorithm is useless if it doesn’t starts with a clearly defined use/business case.

You’ve known about StartupYard a long time. How has the experience been now that you’ve finally joined us for acceleration?

First I wanted to stay that I should have gone to Startupyard with my previous project. But now I feel lucky that we joined with this project. Having learned the lesson the hard way helps me to appreciate and value the feedback even more. Maybe I would have been more stubborn without  that experience behind me.

The Chatler team during mentoring.

The Chatler team during mentoring.

StartupYard really helped us explore many possible futures, and grow confidence in some of our ideas, while letting others go for now. Instead of sitting in classrooms we were pushed to go out to the jungle and meet possible clients, and investors. This helped us to shape and sharpen our vision through real life feedback. I’m very excited about the progress we’ve made, and where we’re going.