Meet Behavee: The Future of Recommendations is Open Source

StartupYard Batch 9, has been defined by “tech with a soul.” That trend continues even when the focus of the startup is on what people traditionally think of as a dry and boring topic: big data, behavior analysis and product recommendation.

That’s because Tomas Pluharik, CEO and Co-Founder of Behavee, believes strongly that the future of recommendations is at once more personal, and more private. Behavee is focused on providing e-commerce and content companies with the tools to truly understand their customers, but without breaching their privacy or mining their personal data and shopping history. They do this by training their machine learning systems to understand how customers behave, and use that knowledge to create instantaneous micro-segmentation of customers based on what they are interested, and what they really want – not just what a company wants to sell them, or what they’ve bought in the past.

To do this, Tomas has arrayed a team of experts in big data, marketing, machine learning, and software engineering with a strong corporate background, who are tired of the way big data is being used for marketing, and believe there is a better way. Behavee is an open-source recommendation engine that learns from its experiences, but doesn’t violate your privacy.

I sat down with Tomas to talk about his vision for the future of marketing in an age of privacy and a post-GDPR world. Here is what he had to say:

Hi Tomas, tell us a little about yourself and your team. How did you all end up together founding Behavee?

We’re a team of corporate refugees who have all worked many years at the Big 4, banks, telcos, and pretty much every big corporations you can imagine in Central Europe.

Behavee, Tomas Pluharik, StartupYard

Without getting into too many of the messy details, I originally hoped to form a spinnoff from one of the Big 4 consultancy firms, to pursue what I found most interesting about big data, which I’m sure we’ll talk about here. Anyway, that experience showed me that my colleagues and I were living on a different planet from our superiors, and we needed to escape to startupland as soon as possible so that we could really do something that big companies don’t have the appetite for.

We teamed up, David, Jan, Richard and myself (team bios are here), and banded together with Michal and Juraj from the banking sector, and Marketa from Microsoft, and we formed what started as a kind of big data agency. I have really enjoyed the fact that the team has been very dynamic, and a harsh selection process outside the cushy corporate environment, which has molded us into a team of hardened individuals who can stay lean and scrappy.

Behavee focuses on user behavior and micro-segmentation for recommendations. What is the unique approach you’re using, compared with other recommendation services?

Ok, how much time do you have? Just kidding, I’ve learned now to make this as clear and straightforward as possible, although in the background it’s tremendously complex.

First of all, we must understand how content and product recommendations currently work online. Or rather, how they don’t really work. What happens mostly today is that e-commerce companies and content companies hoover up a huge mountain of personal data about their users through the use of memberships, cookies, browsing history, purchase history, and data collection services.

With all that data in one hand, and with a pile of products or content to promote in the other hand, they do what you might expect anyone to do in this situation: they figure out the best way to get the most people they can to buy the products they have or look at the content they own.

Now, in this process, they like to think that what they are doing is helping their customers to find products that they need and want. That is sometimes true. However, as you may imagine, it also produces a huge number of recommendations for things customers not only don’t want, but which are actively annoying or nonsensical for them.

A good example of this approach is if say, I bought a phone on an e-commerce platform. The e-shop can recommend me a case for the phone while I am shopping, or even after I purchase the phone. So far so good. Now say I bought the case. Great! Now what does the recommendation system for that e-shop do? It spends the next 6 months advertising that same case to me, over and over again. That’s a true story by the way. Many people, if not most people know it well. 

Why does it do this? Because it is not looking at *me* and thinking: “Tom already bought that case… let’s find something else Tom would like based on what he is actively doing on our platform.” No, instead the e-commerce platform is saying: “We have a couple hundred of these phone cases lying around… we need to sell them to anyone who bought the phone.” That includes me, even if I bought the case already. Who knows- maybe they get lucky and I buy it again? 

I know that sounds pretty dumb actually, but even the most advanced e-commerce companies are still doing this. Even Amazon does that. They can’t stop.

Wait… why can’t they stop doing this?

Because, as I said they have this pile of user data, and this pile of products. They have to move the products using the data. So these little annoyances like seeing the same promotion 50 times after you bought something are not their cheif concern. Their chief concern is to sell the products they need to sell.

What do you do differently?

Ok, now we’re getting somewhere. Behavee is different because we don’t need this giant pile of customer data. We don’t need to know the history of any single individual and what they have bought, or what they are like. We instead anonymously study the behavior of many people, and from this we derive an understanding of how people really work, and what really motivates them. What we do instead of focusing on the past of someone’s shopping activities, is to focus on how a customer behaves, what she or he does when they are visiting the e-commerce or content site, and use that behavior to understand what they are looking for right now. Not a month ago, but right now today.

In order to do that, you must break this addiction to looking at a customer’s past, and focus on a customer’s present. The fact that I bought a phone 6 months ago is irrelevant, if I am in no way motivated to buy a case. So don’t show me something I am not ready to buy. Instead understand me by the way I behave, and show me something I really want. That is what Behavee is doing. Behind this is machine learning technology, analytical tools and automation software to help e-commerce and content platforms make better recommendations automatically, based on dynamic micro-segmentation, which basically means making offers that work for a customer based on what they do, and not who they are.

How can you understand what a person wants by how they are browsing a site?

Let me give you a nice analogy. My grandfather ran two pharmacies during the second world war and before the period of socialism. He said that after many years, he could tell what a customer was looking for in his shop, just based on how they looked, what they did in the shop, and generally how they behaved. Actually it’s very intuitive, because in-person sales is all about reading body language and judging a person on how they are behaving and how they present themselves to you. Are they confident? Unsure? What items are they interested in? You get that all by observing.

In fact my grandfather said this is the trick to serving people is not to judge based on the past: you pay attention to how they act right now, and not always to what they say about themselves. Very often a customer does not know how to get what they really want. You must understand them and help them to understand and see the products that they really need. Maybe a person who has bought the same thing every week for years doesn’t actually want that same thing again. If you presume too much, you will miss an opportunity to sell them something else.

So this is what we do at Behavee. Our recommendation engine first of all observes how many thousands of people behave on a website, and then just like grandpa in the 50s, we train our machine learning systems to recognize what people need by how they behave: what they look at, where they go, how much attention they pay to something, etc. Then we can actively micro-segment and offer a customer something specifically for them, that we know with a high degree of certainty that they will want.

The best way to give an amazing customer experience is to make it feel like the person magically found exactly that thing they wanted. It feels as if that thing is there waiting just for them. That is the experience we want for customers of e-commerce. Less flashing lights, and more understanding of the individual, but without being creepy and pushing against their privacy.

Our solution looks on the whole customer journey from beginning to the end. We do not ask how we can sell a pile of products we must sell; instead we ask “what does this person need, right now, and can we offer that?”

To do this we are an open-source, open platform company. The more our recommendation engine learns from customer behavior (all without keeping any personal data about individuals), the better we can be at helping e-commerce and even content companies to recommend the “next best thing,” for any individual, based on what they themselves do.

So this is ethical big data?

Of course, privacy and anonymity I believe are fundamental rights that we ignore at our peril.

However, I want to stress that even if I didn’t feel that way, this approach to recommendations just works better than looking deep into someone’s past and trying to predict their future that way. If we keep doing that, using a person’s history to define their future, then we will always miss on the best opportunities, and we will never recognize when our customers have changed and are not interested in the same things anymore.

Using someone’s history all the time is like never letting them grow up. You show you don’t understand or even care about someone by doing that. You bought jeans? You are the jeans guy forever. That is somehow both deeply personal and dehumanizing at the same time. It doesn’t work. It will never work.

Just imagine in my story about my grandfather if he had, instead of looking at his customers as people, just looked at them in some spreadsheet with all their past purchasing activities. That is ridiculous. Maybe he could sell them something that way, but could he tell from this if the person in front of him has a cold right now? No. You can see how this approach can be taken to such extremes that we end up not being able to help people anymore. Ultimately people are looking to be understood, not to be treated as pieces of data.

We work with pseudonymized and anonymized data. We don’t keep data of individuals after using it to train our algorithms. If someone came to me and asked for data deletion (as is anyone’s right now with GDPR), I would have no trouble because I don’t even have it, and I don’t want to keep it. Behavee has no knowledge of individuals, and that makes us better at recommendations than if we did.

As for our partners and clients, this will be an adjustment, but a much needed one.

So you see GDPR (General Data Protection Regulation), as an opportunity to change the way online marketing works?

GDPR will clear the marketing space of the random advert calls for a while and will partially reset some marketing functions (like loyalty programs). The opportunity is in the approach to lead generation. It will be hard to generate direct leads from endless databases anymore. Now you must be smarter and more proactive in thinking about the customer journey.

Though anyone who knows me can say that I am not a big fan of too much regulation, I believe GDPR for the most part did a necessary amount of damage to the current system. It made it so that these ways of using big databases to sell are less attractive, and that is ultimately better for consumers and businesses.

You’ve already launched the service with some pilot customers. Can you talk about that process? What surprised you, both about the client’s needs and their reactions to Behavee’s capabilities?

Yes we started and we expected to collect the data for a month to make something beneficial for our clients. We still do need a lot of data, to be clear, but we don’t need to keep that data for very long at all to train the algorithms, and that data does not need to be linked to any specific person.

To our surprise, in only a week, we were able to generate interesting reports for the client that had an impact on their sales right away. I did not believe we could provide such a benefit so quickly, so that is a very welcome shock. Actually one of our clients is now calling for more help after he just saw the data we are providing. We are becoming very busy!

On which kinds of projects or in which industries do you see Behavee having the biggest impact in the near term, say 1-2 years?

What we can do right now, and for the near future is to provide a much better insight for our clients into how their visitors behave, and how to work with them. On top of that, we can provide, after some time of monitoring customer behavior, a way of automating recommendations of content and products, and also offering customers relevant products from other sellers, with revenue sharing for the lead originating site (referral marketing), if that is desired.

In short our focus is mostly on e-commerce and media. However we see great potential in interconnecting verticals and businesses over the whole internet. Realtime behavioral analysis of users can be used even in industrial segments, but as I said our main focus is now on e-commerce and media.

Tell us about your decision to join StartupYard. You’ve got many years of corporate experience on your team, so what was the biggest motivator for joining the accelerator?

To be frank, we were initially not sure if we should join. But I knew the product is too fat and unfocused, communication too complicated and clumsy and our sales cycle too long 🙂 . So at the end of the day we give it a shot.

I must say that it surpassed my expectations. The network we have through StartupYard is very strong, even stronger than before, and there is a level of trust that we bring with us into meetings and new potential partners and clients because they know we are an “SY Company,” so it means we are serious and have backup on our side.

The StartupYard management team has been great. It can be very helpful to have someone always with an outside perspective, bringing past experience and knowledge of the network to help us stay realistic but also to be more open to new possibilities too. I can only say I recommend it, even if you have lots of previous experience. This is not about someone telling you what to do, but rather having a support system that you couldn’t buy otherwise.

What kinds of partners and customers are you looking for in the near future?

Ok, so I could tell you that we are looking for the most forward-looking and most progressive e-commerce companies, financial institutions, media, etc. That is the startup thing to say!

Actually these businesses are generally very conservative, and the most conservative players can use our help the most. Some of these companies are trying to build their own closed solutions, and that is ultimately not going to be enough, or is frankly just a bad idea. Smaller startups and younger companies are already set up to use data more wisely, but bigger institutions are big machines that have only a few speeds.

Thus we are looking to work with companies that deal with lots of customers in a pretty conservative, by the book big data approach. That is where we can do the most good, because generally these companies have a lot of tools and a lot of data (and with this they do make a lot of sales) but not much in the way of connecting all those tools to the data to actually improve their customer experiences. Behavee isn’t about “press this button and user your data to sell more.” It’s about collecting the right data, and using it the right way to ultimately grow your business stronger, with better customer experiences. That will generate more sales, and not only in the long run.

What we have learned recently is that being open-source is not a deal breaker for big companies and conservative institutions if they understand what is actually being shared, which is not their own operational data or the data of their customers, but the deeper understanding of how customers behave in general. In this way a bank that is using Behavee can benefit from our understanding of an e-commerce company, and vice-versa. Everyone can become stronger this way.

So if you have a lot of data and you somehow feel that you aren’t quite giving your customers what they are really looking for, then we are open to a discussion with you. Let me convince you that open-source is the future, and that understanding customer behavior, not just mining your data, is the key to being competitive today and in the future.