BaseModel is a private foundation model for behavioral data Foundation models like ChatGPT, GPT-4, Dall-E 2, StableDiffusion have revolutionized Text and Image processing.

A single large model trained on massive datasets can replace thousands of specialized models. For the first time, BaseModel allows to apply the same principle to behavioral data.

• countless handcrafted features had to be created using expert knowledge
• despite best efforts, important behavioural cues were often lost due to human limitations
• the information content of raw data was orders of magnitude richer than the actual input of models.

BaseModel eliminates these problems and supercharges behavioural ML.



Example questions BaseModel can answer:

General - How do daily customer interactions influence their future behaviors?

Ecommerce - What is the customer’s likelihood of using a special offer? Which products/promotions/ offers/categories the customer is interested in?

Home & Furniture How to split the customer population into behaviorally distinctive groups?

Retail - How much will the customer spend in a specific category next week?

Banking - What is the utility of customer for your business and what are the behavioral and sociodemographic factors affecting it? What is the customer’s projected profitability in the next quarter?

Payment Is recent behavior of the customer inconsistent with past habits?

Telco - How much data traffic will the customer use this month?

Automotive - What kind of product/category is the customer interested in and why?

Travel - What is the customer’s expected number of trips this year?

News & Publishing Will the reader subscribe to a premium plan?

It utilizes a mix of Graph ML, differential geometry and Deep Learning.

BaseModel uses proprietary research to represent complex multi-modal, multi-source histories of behavior in the form of sparse vectors, called Universal Behavioral Representations. Technically, these vectors represent probability density estimates over Riemannian product manifolds and can serve as both inputs and targets for neural network training. In simpler words, BaseModel compresses multi-modal event series into very wide fixed-length sparse vectors.

The key property of BaseModel’s representations is that they are approximately reversible - which means, that it is mathematically possible to query a Universal Behavioral Representation about the elements aggregated within, with high accuracy. This allows neural models to „ask” very specific questions about the user’s historic activities, without the need for encoding precise knowledge as manual features. This unique property also allows for fine-grained interpretability of models (down to the lowest level of raw data).