---
title: Before running your own private AI, sort the data it would read
author: Giacomo Balli
published: 2026-11-17
url: https://BigBalli.com/blog/before-running-your-own-private-ai-sort-the-data-it-would-read
tags: AI, Engineering, Product, Leadership, Data
---

# Before running your own private AI, sort the data it would read

*By Giacomo Balli* | *Published: November 17, 2026*

"Should we be running our own private AI?"

That question usually arrives before anyone can say which data the AI would read. The data decides it.

The companies doing this well sort their data sources once: ordinary, critical, strategic. Ordinary goes to the big public models. Critical and strategic go to a private one. The software routes by which source a request touches, so nobody on staff makes the call at 4pm on a Friday.

In a 40-person distributor that might mean the product catalogue on one side, and customer contracts and real job costs on the other.

Two things can force the question before any sorting: a customer contract that limits where your data may go, or an owner who decides the cost sheets are the company.

Cost is a separate argument. Revolut said this month it moved its engineers' coding assistants onto open models and that the bill fell about 90%. A bank has its own team to run them. A smaller business pays for that team one way or another.

Whoever sells you private AI is paid more if the answer is yes, and cannot know which of your data needs it. So the first step is yours: a list of every source the AI would read, and which of the three it belongs to. If everything is ordinary, privacy is not your reason to buy it.

[Discuss on LinkedIn](https://www.linkedin.com/feed/update/urn:li:activity:7528594072890945539)

---

**Tags:** AI, Engineering, Product, Leadership, Data

---

*View the [HTML version](https://BigBalli.com/blog/before-running-your-own-private-ai-sort-the-data-it-would-read) of this post.*
