---
title: "When data is ready for AI and for agents"
description: "Source, definition, owner, quality rule and permission: the checks on data before entrusting it to a person or an agent."
url: https://www.joinx.it/en/resources/ai-ready-data
---

# When data is ready for AI

Source, definition, owner, quality rule and permission: the checks on data before entrusting it to a person or an agent.

JoinX · Last updated 18 November 2026

In many committees you see the same indicator with two values, one brought by finance and administration and one by the sales department. People notice, discuss it and in the end choose which one to use. An agent, by contrast, uses the number it finds, without asking whether another exists, and puts it into a reply or a proposal with the same confidence with which it would use the right one.

That is why, before entrusting data to an agent, we ask of it the same questions we would ask before putting it in a report for management, with a little extra care. These are the questions [Data Foundation](https://www.joinx.it/en/technology/data-foundation) starts from, and they concern the source, the meaning, the person who owns it, the quality and the permissions.

## Where it comes from and what it means

Data that is ready has a recognised source: it is known which system it comes from, who updates it and how often. It sounds obvious, yet one file exported by hand and reloaded elsewhere is enough for nobody to know any longer which version is the good one.

Next you need a shared definition. “Active customer” can mean someone who has made a purchase in the last year for sales, and someone with a contract in force for administration, and neither reading is wrong. The problem arises when the two coexist under the same name, because each function carries on using its own, convinced it is talking about the same thing. We fix the definitions in a glossary that holds for all functions, and where two different concepts are really needed we give them two different names.

## Who owns it, and how its quality is measured

Every data item needs a person who owns it and decides how its definition changes. We usually choose that person from among those who use it to make decisions, more than from those who manage the system it sits in, so we assign margin per branch to finance even when the numbers come from the branch network's management system.

Alongside the owner there is a quality rule. Completeness, accuracy and timeliness are measured, and when a check fails the data is flagged before it reaches a dashboard or an agent. For an agent this matters more than for a person, because a missing value in a file can cause a check to be skipped without anyone noticing in time.

## Who can read it, and for what purpose

The last question mainly concerns agents. For every data item we establish who can read it, person or agent, and for what purpose. An agent that classifies complaints needs the contracts and not the health data, and with [Security](https://www.joinx.it/en/technology/security) we make sure it cannot read it. Confidential data is masked before it is sent to a model, and every access stays in a log that those who carry out controls can consult.

## Ready does not mean perfect

It should be said clearly that a company's data will never all be in order, and waiting until it is is a good way never to start. That is why we work process by process: we prepare the data each process needs, with definitions, owners and checks, and make it available as a documented data product. We deal with the rest when another process needs it, and in the meantime we know where the weak points are because Mapping has assessed its state.

## An example

Take an illustrative case, constructed to show the method. In a company with a branch network, margin per branch comes out in two versions, one from finance and one from the sales network, and at the end of every quarter someone spends time reconciling them before the committee meets. We fix the definition with finance, which becomes its owner, connect the two sources and put a check on the differences. From then on the management dashboard and the agents read the same number, and we measure the discrepancies and the reconciliation times to see whether the work has succeeded.

Then, when a second project arrives, for example a model that forecasts branch performance built with [Data Intelligence](https://www.joinx.it/en/technology/data-intelligence), the definition is already there and nobody has to redo it. It is one of the most concrete ways in which one project prepares the next, as we explain on the [Compounding Intelligence](https://www.joinx.it/en/compounding-intelligence) page.

[All articles](https://www.joinx.it/en/resources#articoli)

## Start from the data of one process

If two systems give you two answers, we can look together at where the difference comes from.

[Contact us](https://www.joinx.it/en/contact?tema=dati)
