If you want to work in the future today then this media job at Infinite Analytics could be for you. Infinite Analytics enables their customers to use their eCommerce data to predict consumer behavior and personalize experiences. Infinite Analytics utilizes machine learning, natural language processing, random forest modeling, image recognition, and semantic technologies to drive increased sales and conversions on client sites.
The secret sauce is matching individual’s profiles to product profiles, which they call genomes. In essence Infinite Analytics is learning everything about the people that buy specific products and matching that information with the deep profiles they develop of your website visitors.
We spoke with Arjun Mendhi, the Director of Products for Infinite Analytics, who talked about all the great MIT minds that are at work finding ways to maximize sales from ecommerce stores.
But isn’t all this personalization biasing the results? Only showing your products you’re most likely to buy vs letting you explore the products? Arjun explains how it all works to Roy Weissman of Media Jobs:
| Roy: | This is Roy Weissman from MediaJobs.com, and we’re talking with Arjun Mendhi, the Director of Products for Infinite Analytics.
|
| Arjun, tell us, what is Infinite Analytics? What do you guys do?
|
|
| Arjun: | Infinite Analytics is an analytics platform for eCommerce industry. Essentially, what that means is we enable product discovery on the platform using various technical advances that are commonplace today in various other industries. Think of it this way. Today eCommerce companies spend a lot of time, effort and resources in attracting the right type of customers to their platform, but once they land at the platform, their experience, to say the least, it’s difficult, because most of the eCommerce platforms have way too many products, and product discovery is a challenge. At every stage of navigating through the website we observe that there are a significant number of people that just drop off, so we address that problem. We enable the customer to find the right product, and that’s our secret sauce, how we identify what would be the appropriate product for that particular person.
|
| Roy: | Obviously, there’s a lot of technology out there, “People who bought this also bought that,” or, “People like this also like that.” What are you guys doing that’s different, that’s not the similar type of software that’s been out there for years?
|
| Arjun: | Sure. We’re not just displaying, “You like this. You’ll also like that,” just looking at clickstream information. What we do is we focus on the machine learning aspects to provide a relevant experience all across the user interaction. That means it could be something as simple as, “You may also like” to something as complex as a personalized search engine. This may take going out of our comfort zone of really envisioning what this looks like, but we’re so attuned to thinking of a search engine as a keyword-driven search based on a lot of popular search engines that we use, that we don’t realize that there is a possibility of a search engine being completely personal where two people who go to the same search engine, type in the same query, see completely different results, because they’re individuals with different requirements, especially on a shop portal.
|
| How we achieve that is a two-part solution. The first-part solution is creating a complete profile of every user, every visitor on an eCommerce portal, whether it’s from historic clickstream information to behavioral information. We use all that data to create that profile. The second aspect, which is pretty unique to our solution, is creating a genome for every product on the catalog, and that could be anything from millions and millions of products to just a few products, depending on the catalog size. We ingest such catalogs and run them through our enrichment tools, and those enrichment tools have components such as image analyses, where we identify patterns and color densities, to object recognition, where they identify what this object actually is, and create that metadata for the catalog, which doesn’t exist.
|
|
| You may send in a catalog for a few gigabytes, and we add metadata to it that enriches it and make it like a few terabytes possibly. Then that becomes the basis of this matching algorithm that we then run. Did that make sense?
|
|
| Roy: | Well, kind of. It sounds like it’s very personal. There’s two things I want to ask.
|
| Arjun: | Sure.
|
| Roy: | One of them, which I’ll ask the second but I’ll get back to this in a second, just that, when you make something very personal, isn’t it almost biasing the results to a self-fulfilling prophesy? Don’t answer that until you answer this. You mention machine learning. We’ve heard a lot about machine learning. We’ve heard about artificial intelligence. How would you define machine learning, and what is the difference between machine learning and artificial intelligence?
|
| Arjun: | Sure. In very simple terms, artificial intelligence is the user infraction layer, and machine learning is the backend system that makes artificial intelligence possible. When you interact with a system that seems to be intelligent, it’s called the Turing test, where you interact with the system, and if you’re unable to distinguish or clearly identify whether you’re interacting with a person at the other end or a machine, then that is artificially intelligent as a system. That interaction layer, I believe, is artificial intelligence, and the systems that perform the learning and training of models is the machine learning layer.
|
| The second part of your question, could you repeat that?
|
|
| Roy: | With the machine learning, you’re saying the machine learning is a primary component of artificial intelligence.
|
| Arjun: | Yes.
|
| Roy: | It’s a subset, but artificial intelligence is a bigger picture. It has more capabilities. Is that what you’re saying, the machine learning?
|
| Arjun: | No. I’m saying there are two layers to an application, an intelligent application. One is the layer that learns from training data or learns, that learning can be supervised or unsupervised. That learning layer is the machine learning layer. Once you have models that have that learning and the training embedded in them, you can create applications on top of them which are seemingly intelligent. That is artificial intelligence.
|
| Roy: | Okay. Machine learning is just learning it, and intelligence is applying that learning?
|
| Arjun: | Yes, you can say that.
|
| Roy: | Okay. The other question I ask is we always talk about personalization, and Google has tried to do some of that by when you sign into your account it keeps track of the kinds of things you like to see, so it tries to show you search results that relate to things you want to see. Sometimes when you talk about personalizing search based on my behaviors on a website or whatever, one of the things that intrigues me about this is I always feel like aren’t you effectively creating a self-fulfilling prophesy? You’re limiting the results to what the computer thinks you want to see.
|
| It’s kind of like having an office full of “Yes” men. Nobody every disagrees with you, so you start to think you’re right. If you start searching for something and it thinks, “Oh, well, he really wants golf shoes,” and he starts showing me golf socks and golf shoes, and I’m searching for something different. I just happen to initially out of curiosity look because a friend of mine bought golf shoes. Aren’t I limiting my results? Isn’t this personalization limit my ability to maximize, or do you find a different value to it?
|
|
| Arjun: | Yes, that’s a great point. That’s also the boundary between being supportive as platform, being helpful, and being creepy. That’s something we think of very carefully in how we design our systems. The best way I can put it is we make the process of discovering the right product a seamless experience and not an intruding experience. What that means is we take live cues from what the user is doing and support that process as opposed to make our assumptions and consciously direct the consumer in a direction that the consumer may not want to go into. We’re not walking in with a predisposition. We’re not trying to drive the consumer in a direction which is dis-joined from the consumer’s intentions. This brings me back to our machine learning systems, where we’re able to match the user’s real-time behavior with what we think would be a complementary showcase of products, and that’s definitely based on data.
|
| To summarize, yes, that is a concern, but I think it’s a delicate balance between whether you are imposing products that you actually want the user to look at or you’re just letting the data do its job and you’re providing the user insight into products that might have probably taken him or her various clicks and various other pages and searches to get to.
|
|
| Roy: | No, that makes sense. I think that makes a lot of sense. When we talked earlier, you had mentioned that you guys had done some tests, and had been a competition or something where they put your software up against other software. Maybe you could just explain that a little, the results.
|
| Arjun: | Sure. That’s a simple AB test where without giving specifics of the client or the other contenders, it’s essentially one of the biggest eCommerce platforms in the world. The platform had consciously decided to improve its personalization aspects and brought in some of the leading vendors in the world, including the incumbent player. We had ABC essentially, an ABC test, with the three short-listed vendors getting a portion of that traffic. The test ran for three weeks live on their website, and every aspect of the performance of these personalization systems was monitored. We were by far the smallest company in the competition. The others were fairly well-established, but we won that competition, that test, in every single parameter that had a business significance to the portal that included metrics such as average order value, total revenue generated from our recommendations and personalization engines, total traffic generated, click-through rates and things like that.
|
| Again, it’s important to remember that these systems are, they’re basically machine learning systems which get better and better with more and more training data. If you bring in an outside entity and give them a small percentage of your data and put them live on your website and compare that engine with the incumbent engine, which has been on the website for, say, years, then there’s a huge difference in how well-calibrated or learned these models are. Despite that difference, we were able to win the competition.
|
|
| Also, one of the biggest challenges we had was that the portal, the eCommerce engine, had customers that were using the portal in a non-English language, so not only did we have to understand the behaviors of a completely different demographic than our core markets, we also had to translate and understand metadata in a different language. Despite all these challenges, we were very excited that we won that test, and it speaks volumes to our unique approach and how that is not just theoretically superior but also delivers results.
|
|
| Roy: | Did you have any calculations of how much that could have been worth to the company on a global basis or in dollars and cents or something like that?
|
| Arjun: | Yes. The company themselves did that. They were monitoring it very closely, and they were extrapolating the fractional data that we were getting to a scenario where each company would get 100% of the data. Given that, yes, all I can say is, it was worth a lot. It was worth a significant chunk to the organization. They’re still working on finalizing their next steps, so it’s a fairly recent test that we came out of.
|
| Roy: | That’s very exciting.
|
| Arjun: | Thank you.
|
| Roy: | Hopefully, you’ll be able to share that with a few other potential customers, those kinds of results.
|
| Arjun: | We’re hoping to, yes. Once this is public, we are hoping to also share more information about it. Check back with me in a few weeks.
|
| Roy: | Do you guys have any sense of the size of the potential … Obviously, you’re going to license your software, I assume, and people will pay you a fee. Is that going to be based on a percentage of revenue, or it’s just a fixed fee? How do you envision pricing it?
|
| Arjun: | Yeah, that’s a great question. One of the things we observed about this market is that pricing is very opaque, and it’s very, very customized and extremely negotiated. One vendor has no clue how the market is pricing such services, so we try to take a completely different approach. Our approach is as transparent as it gets. We talk about a value we can deliver. We analyze their existing performance, and we look at what’s that delta that we can create.
|
| Let’s say currently take a hypothetical scenario where a customer has a recommendation engine that’s providing, say, X percentage of their revenues through certain click-through rates to certain average auto value. We compare that with their generic traffic, and so if the other data that we have from our other engagements, we know with very little or, let me say, with a well-defined margin of error, we know what is the difference in performance we can deliver. We take that difference in performance that we can create in the top line for the customer as the total value created, and we capture a certain portion of that value for our services. The rest of it is left on the table for the customer, which is a comfortable margin.
|
|
| Now we give the customer a lot of flexibility in how they want to structure that value that we capture. They can structure it in various ways. They can award that to us as a flat monthly fee, which of course is pretty transparent. If X dollars is being captured by our company, that can be divided by the duration of the engagement. Should they choose to make it performance-based, a little more performance-based, they can split that between a smaller portion as a monthly fee and a certain percentage as a revenue share, as a revenue share of the total revenue generated by the recommendation engine, or if they want to make this a hundred percent performance-based, our entire compensation can also be as a revenue share based on the revenue we generate for the customer.
|
|
| There are various levers that we offer our customers to work with. The idea is essentially pretty simple. We know what’s the value we can create, and we want to hold ourselves by those results. We understand every business is different, and their way of working with suppliers and vendors is different. We want to be as flexible as possible while being fair on both sides. We don’t see any reason not to be transparent in our pricing.
|
|
| Roy: | Have you got a sense, have you guys sat down and estimated the size of the market opportunity for you guys in dollars?
|
| Arjun: | We constantly try to do that, but being a startup, we are also very adaptable. At some point we can make big statements about how such personalization industry is a multi-billion-dollar industry with more than 30 vendors trying to claim the space, including larger organizations like IBM and smaller organizations like, but I think we try not to get swayed by such numbers. We try to be very, very focused on what our differentiation is and who these customers are that will value our service. The short answer is, yes, there are various public reports which give large numbers, but I don’t think we actually get influenced by those. We take one customer at a time.
|
| Roy: | What do you think the potential could be, dollars-wise?
|
| Arjun: | We are pretty sure the path that we are on right now would help us become a billion-dollar company in a not so long future.
|
| Roy: | Wow.
|
| Arjun: | Let me put it that way. It’s not just the, as we talked about, it’s not just recommendation engines. It’s a lot more than that. It’s the intellectual property and the IP that we are creating in the company which has a lot of various applications that we are not even exploring at the moment. We’re trying to put one step ahead of the other in a very strategic manner, excel at what we do first before we start diversifying, but definitely diverse applications are down the roadmap.
|
| You can think of various ways machine learning can benefit different businesses in the world, whether it’s anything from drug discovery applications to healthcare to financial services. There are various, various applications to the technologies that we’re creating. Right now our focus is, deliver value in the industry we play in to the customers we work with and be the world’s best at that. Once we’ve created that intellectual property, we can create the room to explore diverse applications, but right now we’re extremely focused.
|
|
| Roy: | When were you guys founded?
|
| Arjun: | Two thousand twelve, out of MIT.
|
| Roy: | Had the product officially been released yet, or is it still being tested at this point?
|
| Arjun: | Oh, the product’s been released. It’s live on various sites with various customers. One of our strategic decisions early on in the company was to start with emerging economies. First of all, they are seeing just a booming growth in eCommerce. It’s also relatively easier to experiment in emerging economies with companies that are more open to nascent and just technologies that are still under development.
|
| We saw a lot of traction in emerging economies. Currently we have some of the largest platforms, eCommerce platforms, say, from India, working with us. We have penetration in Brazil. We believe we’ve reached a point where we’ve proven our products and our technologies to be the most superior in the world in various aspects, and we feel we’re ready to enter the US market now. Now is the time where you start seeing the name “Infinite Analytics” be more prevalent in the US market and media. We’ve taken the time to not rush into this market with a half-baked solution. We think we’re ready with a solution that can compete with the best in the world, and we’re optimistic.
|
|
| Roy: | Who founded the company?
|
| Arjun: | Two graduates from MIT, one of them in a technical graduate course, and one of them in the MBA program coming out of the school in 2012. It’s an interesting story how they actually came up with the first version of this product.
|
| Roy: | Could you tell us the names?
|
| Arjun: | Yeah. Akash Bhatia, the CEO.
|
| Roy: | Okay.
|
| Arjun: | Purushotham Botla, the CTO in the company. They met at a class taught by Sir Tim Berners Lee called Linked Data Ventures, and Linked Data Ventures essentially talks about the future of the internet, the future of the worldwide web, the semantic web and linked open data, and the value that can be created from systems that utilize this future vision of the web. They started out with a class project called Schmooze-Butler, which was essentially a networking application on social media, but fundamentally it connected disparate data systems across the web in a very intuitive manner that opened a lot of room for analytics and applications. That was so impressive that they decided to launch a company on the basis of that project, and till today Tim Berners Lee is an advisor for Infinite Analytics, so that speaks to the traction we’ve created and the vision we have.
|
| Yes, that’s the foundation. We’re still looking into the future. We always look at where internet is going to be as a whole ten years down the road and not one or two years down the road. That’s always been a driver for our product strategy.
|
|
| Roy: | How many people do you have working there now?
|
| Arjun: | I’m looking around our office right now. We have about 15 people here in Cambridge. A lot of interns have just joined from MIT. More to come next week. We have about five people in Europe. We have about five people in India right now.
|
| Roy: | You’ve got about 25 now.
|
| Arjun: | Yes.
|
| Roy: | Are you looking to hire people?
|
| Arjun: | Yes. We are certainly hiring. We’re hiring for different roles. On the technical side, we’re looking for system operations managers who can help us scale our infrastructure. We’re looking for performance engineers. Our systems are completely real-time with very, very ambitious response times of sub-20-milliseconds for various applications, so performance engineering is a very important role in the company. We’re looking for data scientists. We have a very high density of data scientists in the company already. That’s essentially the foundation of everything we do. Then we’re looking for UX developers, user experience engineers, who can help us build customer-facing products that are seamless, intuitive and best in class in the world. Then on the business and management side we’re looking for product managers, and we’re looking for business development associates to join our team. In business development we’re looking for people both in the US and in Brazil.
|
| Roy: | It sounds like you’ve got a lot of things going on. It’s about to explode, huh?
|
| Arjun: | Yes, it’s an exciting time, so stay in touch.
|
| Roy: | If I come back in five years, will you still be an independent company? Will you be a public company, or will you have been acquired by some big company?
|
| Arjun: | Of course we don’t know the future, but what I can tell you pretty confidently is an exit has never been a driver in the company. We never do anything to make us more attractive for a certain exit opportunity. We never think of acquisitions. We never think of just exiting in a fashion. We think of just excelling at what we do. We think of how we can build our technologies and our team to create maximum growth in the company. I think being distracted by such drivers could be detrimental at an early stage in the company.
|
| Roy: | Is there anything I haven’t asked you that you wanted to share with the audience?
|
| Arjun: | Just that we’re based out of Cambridge in Massachusetts. That’s our headquarters. If any of your audience are in Cambridge or Boston and would like to touch base with us, we’re a great place to work. We work in a shared open environment with a lot of other startups. It’s a lot of fun working here. There are pretty frequent ping-pong tournaments. There are team hangout lunches. Cambridge is a very intellectual and academic environment with a lot of students, a very vibrant community. It’s a great place to work.
|
| If any of your audience members that’s around wants to meet us, then just shoot us a note at Contact@InfiniteAnalytics.com or go to our website and shoot us a note. We’re always happy to connect with people who are eager to contribute or just learn more about us.
|
|
| Roy: | The website is InfiniteAnalytics.com.
|
