5.5.22-Efficient-AI-Covariance

MICHAEL FLEDER: Hi. Glad to be here today. My name is Mike Fleder. I spent a decade at MIT in LIDS and [INAUDIBLE]. I'm the founder of Covariance, a rapidly growing machine learning startup with roots in our MIT research that's now powering billions in decisions.

Imagine you had access to your customers' financials. You would use that every day in all of your operations. That is now possible through a new type of efficient learning. Our MIT research is the foundational work in this area.

If your competitors are public, once a quarter maybe you get a few high level metrics like revenue. But how many customers do they have? How valuable are the customers that you don't know? Competitors will never tell you. Maybe you buy high level market reports, but high level market reports don't answer the most valuable questions.

In the last 10 years of machine learning, big data techniques have worked really well for prediction problems with clear measurable outcomes. Does this photo have a dog in it or not? A human can help train a machine to do that. But what about problems where there's no clear dog or not answer? What happened yesterday at your competitor? That's never disclosed. And here traditional techniques in machine learning and time series breakdown.

Beyond our research breakthroughs, turning this into practice takes a lot of care. We're combining petabytes of raw data with tiny amounts, megabytes, of limited financial disclosures. This processing takes hundreds of tightly controlled steps. The result is leading forecasts outperforming the hardest benchmarks in multiple industries. And our work is now powering billions of dollars in decisions.

The way to solve this problem is by combining external data with internal data. This is extremely challenging, since these data sets look nothing alike. They are all noisy, they're biased, they're in different units, they vary wildly in size from petabytes to megabytes. Today we partner with our clients to understand the competitors and KPIs that they care about. And usually these are in these areas on the right. Best of all, everything adds up to total. It's a unified model of your competitor's business that's 100% consistent with anything that's reported if anything is reported at all.

This is an example of some work that we've done on a Fortune 100 retailer. In their latest earnings release, we were significantly more accurate than Wall Street consensus, which is the gold standard, one of the hardest forecasting benchmarks in any industry. We are then able to break down these types of high level metrics and aggregates and make it actionable for our clients.

We delivered forecasts on in store versus digital and measured the impact of remodeling stores and more. A single customer profile is not necessarily what's of interest, but it's the multilevel, multi-resolution zoom in and enhanced detail that makes this actionable. Which segments are you winning and losing? We can even get more granular. Media customers are consuming or competitors customers are consuming, distance to competitor stores, places these folks visit.

Today we're actively working with corporates and financial services. We package a tier one machine learning team at a fraction of the cost. We're looking to engage for proof of concept pilot projects. We still have 100% conversion today from pilots to long term. We have a stellar track record in conversion. And long term, we're a partner in helping organizations leverage external data and state of the art machine learning. I look forward to your questions and discussing next steps. Thank you.