Why We Invested in Jetlore

A deep dive into AI applied to product/content personalization

Versión en Español

Jetlore, headquartered in Sunnyvale (CA), was founded in 2011 by two Stanford PhD dropouts, Eldar Sadikov and Montse Medina. The company has developed an e-commerce personalization SaaS solution based on machine learning/AI that is already in use by some of the world’s most important e-tailers.

JME Venture Capital led their latest financing round (a Series A extension) in September 2016.


The problem of discovery

It can be said without fear of exaggeration that discovery, the ability to find new products relevant to my interests among an almost infinite online offer, is one of the main problems still to be properly solved on the internet today.

By eliminating the barriers to physical distribution and reducing the costs of production, technology has created an unlimited supply/choice problem: there is so much I can choose from that I feel completely lost. It is very likely that even though there might exist new products that are an excellent fit for my needs or tastes I can neither search for them, nor can these products “find me”.

We were promised the power of the long tail (and flying cars); instead we got “blockbusters” (and 140 characters).

I can’t search for those products, because I don’t even know they exist. And when I do, I go straight to Amazon.

And they can’t “find me” either, because the recommendation algorithms — the tools that e-commerce companies use to suggest products — are somewhat coarse when it comes to learning about my true preferences and, moreover, they don’t work in the upper funnel (promotional email, homepage, product listing pages), where ~70% of the customer interactions take place.

Therefore, in retail, if we combine the existence of an 800-pound gorilla, with which it is (nearly) impossible to compete on price or quality of service, with a continuous increase in customer acquisition costs (launching an e-commerce is easier than ever, but it is perhaps harder than ever to retain customers), e-commerce companies have no choice but to:

  • Sell their own products
  • Use tools to increase the conversion of visits to customers
  • Increase their direct organic traffic (free), attracting customers who regularly visit the web / app to discover new products as a form of entertainment (similar to how we visit Facebook in order to discover content of our interest)

As Benedict Evans pointed out very well:

Traditional recommendation algorithms

The most commonly used e-commerce recommendation algorithms (the so-called collaborative filtering algorithms) are based on a very simple mechanism: they learn which products co-occur more frequently for certain events (“users who buy X also buy Y” or “users who look at Z also look at W”), and use that information to make recommendations whenever those events happen again(“a user just bought X, so I recommend Y” or “a user is browsing the page Product Z, so I’ll also recommend W”).

This approach, while having the advantage of simplicity, is very limited in its effectiveness.

By operating at the product level — rather than at the user level — the algorithm can only base its recommendations on a single action (what product I’m looking at, what product I just bought), rather than learning why I’ve looked or bought a series of products in the last days (or in my whole history). It is even less able to learn what characteristics of those products — which could also be present in other products that have not been bought or seen by anyone yet — are the ones that are actually attractive to me.

Because of this, traditional recommendation algorithms are not able to make “original” recommendations based on the preferences learned from each user, instead they focus on what others have done before. This lack of originality causes the algorithm to become more ineffective over time, since the best-selling products become the most recommended and then even more sold, in a kind of self-fulfilling prophecy. As a consequence, the products that could be more interesting for each customer are increasingly buried under the biggest sales hits.

Enter Jetlore’s learning-to-rank technology

Jetlore has developed a proprietary technology — learning-to-rank — that, in a simplified way, works like this:

  1. Analyzes the retailer catalog, breaking down each product into a series of attributes that characterize it,
  2. Learns (in real time) the preferences of each customer for each attribute, according to a series of signals generated by their behavior in any channel (email, web, mobile), even offline,
  3. Calculates a score that represents the relevance of each product for each customer (based on the attributes of the product and the learned preferences of the customer),
  4. Shows to each user (in each channel, in real time) the products that are most relevant to him or her.

In addition, Jetlore provides its customers with all the tools (e-mail templates and web and mobile layouts) and support necessary to integrate their solution as easily, fast and successfully as possible.

If you are wondering if this approach works, I prefer to let the numbers speak for themselves: Jetlore has managed to increase its customers’ revenues by 20% to 80%, including at such well-known and established players as eBay, Nordstrom Rack, and Inditex.

In short, Jetlore’s technology allows retailers to offer customers a highly individualized and novel experience at each visit that encourages not only conversion but also retention.

A Facebook for products, anyone?

Our investment decision

If, to a product as incredible as the one described here (and with the clear potential of application in additional businesses and verticals), you add…

  • Two cofounders with extraordinary talent, great work ethic, who we enjoy working with and whom we can fully trust, surrounded by a top management team and an offshore technical team that helps to somehow alleviate the through-the-roof labor costs of Silicon Valley,
  • A very healthy cap table, in which the founders still maintain a very significant participation in the capital, and in which we find several international investors with whom we feel very comfortable working,
  • A huge and fast-growing market (marketing technology), populated by big customers willing to pay big chunks of money to solve a problem of vital importance for them,
  • A value proposition validated by top-tier clients,
  • A proven business model with very attractive metrics, including revenue growth, virtually zero churn and high commercial efficiency,
  • A technology based on solid foundations, perfectly prepared to scale,
  • A very contained cash burn, with the possibility of becoming profitable always at an arms length, and
  • Good chances of exit, due to the company’s proximity to large players, who will be potentially interested in making a strategic acquisition to complete their offering in this space

… the investment decision was frankly easy for us.

Eldar, Montse, thank you for letting us be part of this adventure. We hope to be good fellow travelers.


If you’ve found this article helpful, I’d really appreciate it if you could click the ❤ below so more people can discover it here on Medium. 🙏🏽