A familiar situation: the CRM shows one address and one account manager for a customer, the ERP shows another address and another manager, and the sales team's spreadsheet holds a third version. A person who meets this feels that something is wrong and checks. An AI agent does not feel doubt. It takes one of the sources and answers with full confidence.
What a single source of trusted data is
It is an agreement about which system is the master for each type of data: customers in one system, prices in another, contracts in a third. All other systems receive these data from the master and do not change them on their own. When someone asks “which price is correct?”, there is exactly one answer, and it is written in the system, not in someone's memory.
Why this matters for AI in particular
- The agent has no doubt. A colleague who sees two different prices will ask. A model will pick one and continue.
- The agent works fast and at volume. A mistake that a person would make once is repeated across hundreds of documents.
- The agent writes results back. If the result goes into your system, a wrong value becomes a “fact” that other people and processes rely on.
Signs that the single source does not exist yet
- The same data is typed by hand in several places.
- Reports from different departments give different figures for the same period.
- A spreadsheet is treated as the “real” data, while the system holds an outdated copy.
- Nobody can say quickly which version is correct.
- There are duplicates among counterparties, products or contracts.
How to create it: practical steps
- List the types of data the process uses: customers, products, prices, contracts, statuses.
- Choose the master system and an owner for each type. The owner is a person, not a department.
- Find and correct duplicates and contradictions. Start with the data the process actually uses.
- Agree directories and formats: units, currencies, statuses, naming of counterparties.
- Add rules that prevent new contradictions: required fields, checks at the moment of entry, a ban on editing copies.
- Set up automatic data exchange, so that a change in the master system reaches the others without retyping. This is the subject of the next article.
How long it takes
It depends on the number of systems and the state of the data. A realistic estimate appears only after the systems are studied, which is why we start with an audit. The scale of the work is also one of the questions of our solution configurator.
Can we start with AI and tidy the data later?
A demonstration can be built that way, and it is a reasonable way to see what AI could do. For a working system the risk is different: the agent may fix existing errors in place and spread them. That is why we recommend building the foundation first. It takes more work at the beginning, but the result is reliable.
Where to start
Look at one process that you want to automate and ask: which data does it use, and which system is the master for each of them? If the answer is unclear, the first step is an audit of systems and processes. A short readiness checklist will also help you go through the questions with your team.