On this page
When an AI project stalls, the model is rarely the problem. The data is: spread across tools, defined three different ways, and owned by no one in particular. Before choosing a model, it's worth knowing whether your data can carry the weight.
Know where it lives
List the places your important data actually sits: the CRM, the accounting tool, the spreadsheets, the shared drive, the inboxes. Most teams find more than they expected, and at least one spreadsheet everyone depends on and nobody owns.
Agree what it means
"Customer", "order" and "revenue" often mean slightly different things in different teams. AI will repeat whichever version it finds first. Write down the definitions that matter, and pick one source of truth for each.
Check it before you trust it
- Complete: are the fields you need actually filled in?
- Fresh: how old is the data by the time it's used?
- Unique: is the same customer in three places, spelled three ways?
- Owned: who fixes it when it's wrong?
Make knowledge searchable
Much of what a team knows isn't in a database at all. It's in documents, tickets and email threads. Knowledge search connects those, so people and agents can find answers with a source attached. The source matters: an answer you can't check is an answer you can't use.
Start small, on real work
You don't need perfect data everywhere. Pick one process that matters, say how fast you can answer "has this customer paid?", and get the data behind it clean, connected and owned. Then build on it.
That's usually where our Data & Intelligence work starts: not with a model, but with the questions a team needs answered every day.