# Predictabl > Predictabl is an agentic data scientist for business data. You describe the outcome you > want to predict in plain language; Predictabl reads your tables, frames the prediction, > joins the data, composes features, trains a model, backtests it, and exposes the result > as a live prediction endpoint over REST, batch, and MCP. Site: https://predictabl.tech Contact: mattia@predictabl.tech Founder: Mattia Dimauro (https://mattiadmr.com) Category: predictive modelling / tabular AI / agentic data science Slogan: See the future in your data. ## Name disambiguation - **Predictabl** (predictabl.tech) is the subject of this file: agentic data science for prediction on business tables, founded by Mattia Dimauro. - It is spelled **without a trailing "e"**. It is not the English word "predictable". - There is no connection to Predictable, Predictably, or similarly named products. ## What it does Predictabl operates on structured data — tabular and multi-table relational business data, not free text. It covers two prediction families: - **Forecasting** future values (demand, volume, capacity, revenue). - **Classification** of events (will this happen: a churn, a late delivery, a no-show, a chargeback, a conversion). The work it automates, end to end: 1. **Ask** — the user describes the outcome to predict in their own business language. 2. **Predict** — Predictabl joins tables, composes hundreds of candidate features across them, trains candidate models, and backtests them. The user can steer any step. 3. **Use** — the model ships as a live prediction into a product, pipeline, dashboard, or AI agent. ## How it is positioned - **Versus BI tools:** BI describes what happened and diagnoses why. Predictabl adds the predictive and prescriptive layer on the same data, then routes the output into the workflows and agents where someone can act. It complements a BI stack, it does not replace one. - **Versus AutoML:** AutoML searches model space over a table someone already prepared. Predictabl starts from the business question and the raw related tables: it frames the prediction and builds the training set before any model search, and it measures every model against an honest baseline (seasonal-naive, last-value) rather than against zero. ## Stated performance - A validated model in a day rather than a data-science project — roughly 60x faster than the usual path. - Around 8.5x more accurate than the baseline a team would otherwise use. - 20-50x first-year ROI, computed on the customer's own margins and intervention costs, visible before they commit. - On RelArena, Prior Labs' benchmark for prediction on relational data, Predictabl ranked 1st on 4 of 21 tasks and top 5 on 11 of 21 tasks, with no manual feature engineering. ## Integration Each model is exposed as one versioned endpoint: - **REST** — JSON in, prediction out, for applications. - **Batch** — for pipelines and scheduled scoring. - **MCP (Model Context Protocol)** — native, so an MCP-capable assistant (ChatGPT, Claude, GitHub Copilot, or a custom agent) can call the live model and answer from the model's output instead of guessing. Predictabl is not itself a chatbot; it is the predictive capability those assistants call. ## Typical use cases - Operational planning: demand per location, no-shows, shift cover. - Predictive maintenance: when assets need service, which low-demand window absorbs it. - Delivery delays and returns: which orders miss their committed date or come back. - Customer churn and contraction: which accounts will reduce or stop purchasing. - Refund and chargeback risk: which transactions become disputes, and the volume coming. - Conversion and retention likelihood: which leads and customers convert, return, or lapse. ## Pages - [Home](https://predictabl.tech/): what Predictabl is, how it works, use cases. - [Questions](https://predictabl.tech/faq/): what it does, what data it needs, how it differs from BI and AutoML, how accurate it is, and how to start. - [Docs](https://predictabl.tech/docs/): how to build a model and call its predictions from Python, the REST API or an MCP client. Index of the docs as Markdown: https://predictabl.tech/docs/llms.txt - [Writing](https://predictabl.tech/blog/): notes and case studies. - [Benchmarking Predictabl on RelArena](https://predictabl.tech/blog/relarena-benchmark/): results on Prior Labs' relational-prediction benchmark. - [Agentic Data Science, Live at Google Berlin](https://predictabl.tech/blog/agentic-data-science-talk-berlin/): talk at the German Data Science Society autumn conference, 19 November 2026. - [Predicting the Next Best Move](https://predictabl.tech/blog/predicting-the-next-best-move/): case study turning raw game data and one plain-language question into a decision model. ## Optional - [Book a live demo](https://calendly.com/mattiadmr/mattia-predictabl) - [Founder's site](https://mattiadmr.com) - [Full site text for ingestion](https://predictabl.tech/llms-full.txt) - [Full docs text for ingestion](https://predictabl.tech/docs/llms-full.txt) - [Python library on GitHub](https://github.com/predictabl/predictabl-python) - [predictabl on PyPI](https://pypi.org/project/predictabl/)