Why QED invested in Ntropy

Sometimes the world puts two big ideas on a collision course.  

For the past decade, the promise of “big data” and “machine learning” has been everywhere, promising to outpace human judgment in accuracy and provide infinite scalability. Yet the power of this promise rested on a valuable and scarce input, raw data. Futurists, entrepreneurs and VCs dutifully parroted the phrase that “data is the new oil”  and every startup software company’s pitch deck ended with the idea that “the data is so valuable we might give away the software for free.”

In a seemingly parallel universe, the European Union passed the General Data Protection Regulation (GDPR); California followed shortly after with the California Consumer Privacy Act (CCPA). Everyone acknowledged that privacy was important; privacy compliance startups were founded, cookies were “banned”, and ad tech was doomed. Global implications loomed. Data privacy was coming, but it certainly wasn’t here yet.  

I still don’t know what will happen to the rise of machine learning in the age of privacy. But it seems to me that we need some real innovation to allow these two big ideas to co-exist.

When we met the Ntropy team in the Summer of 2020, we were immediately struck by the power of their vision. They had developed mass customized labeling algorithms that were naturally privacy protective.  

Let’s decompose this sentence.  

Most machine learning labeling algorithms, whether it be for a cat GIF or anti-money laundering, are designed for precision. Given the same input, they will deliver the same output.

This emphasis is architectural. First, these labeling tasks are deceptively complicated. Moreover, Google and others that have developed these algorithms are serving them to the mass market; the stability of their predictions is an important feature.

But what would you do if you wanted to find your lost cat? Could the algorithms trained on ImageNet help? They’re a necessary but not sufficient condition. To find YOUR cat, you need a model that is both robust (informed by huge data sets) but also flexible (can be trained on a small data set for a much narrower predictive purpose).

It turns out that the Cambrian explosion in fintech that we’re living through requires just this sort of balance between robustness and flexibility. Everyday, we see dozens of new fintechs, affinity-oriented neobanks, banks that want to reward health and wellness, apps to auto-categorize business expenses, apps to buy carbon offsets based on your spending patterns. These fintechs are trying to escape the ruthless logic of a well-funded feature parity war.  

For these companies, the heart of their business model is to build new, differentiated experiences for their customers’ financial lives. And these differentiated experiences require dynamic labeling.  

Let’s use me as an example and compare two trips to my favorite sweetgreen in Old Town Alexandria.

One day, I celebrate the return to open restaurants by going to sweetgreen and buying a salad. I pay with my personal card and my healthy living app picks up the transaction and gives me 10 reward points for eating healthy. One week later, I go to the same sweetgreen and buy two salads. Now I’m celebrating the return to in-office meetings by taking an entrepreneur out to lunch. This time I use my business card, and my automated tax app picks up the transaction and auto-categorizes it as a business lunch.  

Today, without paying (or developing) a transaction categorizer, both of these fintechs would be stuck with an only partially intelligible alpha-numeric string as the data that they would get from the card networks. Ntropy parses that and returns not only the name of the merchant, but also logo, website URL and Yelp reviews. But most importantly it allows for custom categories, so the same merchant and the same end consumer is categorized as a tax deduction by one fintech app and as healthy eating by another.  

This is a magical experience for the product designers and engineers – with Ntropy, they design their unique features and customer delight, they don’t get bogged down in data labeling and the difficulty of hiring world-class data scientists. What Ntropy offers are analytical primitives: scale, customized, accurate category labels. As one of our angel investors told me, “I had been looking for this type of an API for years. I just had an Ntropy-shaped hole in my head.”

Building on top of real-time transaction data with accurate, dynamic labels is not just easier; it will enable experiences that weren’t possible before. Imagine competitive and real-time credit scoring, hyper-personalized recommendations and behavioral prompts in the moment.  

Like a magic trick, it does so without storing the customer data on its servers and without custom building and custom tuning over weeks and months.

Putting these features together has not been easy, but now that it’s here, all fintech product people should be rejoicing.  

Of course, early stage fintechs that are just looking for the shortest route to market are lining up to use their API. One interesting feature of the business case for Ntropy is that the API is attractive for fintechs that are already at scale – with hundreds of millions or even billions of transactions per year.  Ordinarily, these scale fintechs wouldn’t be willing to take the risk of sharing data with a Seed-stage startup. Customer data is a crown jewel and these teams already have large teams of data scientists. But the naturally privacy-protective architecture means that many of the largest fintechs are already testing in Ntropy’s sandbox.

At scale, even fintechs with high-quality data science teams parse their transaction data with a sophisticated rules library, running their data through a waterfall of little tricks. In some of our companies, these libraries have grown to thousands of rules, with no end in sight, and they are sagging under the weight of management and maintenance. Ntropy offers a different way. A scalable, flexible engine that is always improving, informed by the whole world of fintech data.  

We’re excited to be investing in Ntropy to build these analytical primitives and to accelerate customer delight for our own portfolio and the fintech community as a whole.