A simple number like your customer count turns out to be one of the best thermometers for the state of your business data. Below: the test, the four reasons systems contradict each other and what the result means. Including when you do not need to do anything about it.
Ask three colleagues, separately, how many customers your company has. Add that you also want to know where they got the number. One looks in the CRM, a second in the accounting software, the third in a spreadsheet that has been around for years. Chances are you get three numbers, and nobody made a mistake.
That is the whole experiment. It takes five minutes, needs no tools and says more about the state of your data than most reports.
Three colleagues, each on their own. Not in the group chat, or they will brief each other.
Write down per person which number they give and which system or file it comes from.
Are the numbers equal, close or far apart? And can anyone explain the difference?
For each difference, ask what was counted and what was not. That is where the missing definition sits.
It is tempting to assume there is an error somewhere that someone should track down. In practice there are four causes, and none of them is an error.
The CRM counts everyone you have been in touch with, accounting only those who received an invoice. Both call them customers.
Active in the past twelve months, ever invoiced or currently under contract: three reference dates, three numbers.
A company under two spellings, a holding and its subsidiary, a contact person who changed jobs. Every system counts them.
A spreadsheet nobody synchronises lags behind or runs ahead. The number is only as good as the last person who thought of it.
That is also why a clean-up rarely works. You deduplicate the CRM, and half a year later the numbers have drifted apart again, because the definition was never written down. Cleaning up treats the symptom, not the agreement.
Counting customers is the easiest variant. Run the same test for open quotes, active projects and revenue per customer, and you will see the gap widen as the question touches more systems. Open quotes live in the CRM and sometimes in a mailbox. Active projects mean something different to the planner than to the bookkeeper. Revenue per customer requires invoice lines, customer numbers and projects to be linked, and separate systems cannot do that on their own.
If the first question already gives you three numbers, you do not need to do the rest. You know enough. If the first question checks out but the fourth does not, the problem is not your customer file but the connections between your systems.
Not every outcome is a reason to build something. Honestly, for most companies it is not.
Your data is in good shape. Write the definition down so it stays that way, and spend your time elsewhere.
One agreed source for customers often fixes this, possibly with an integration to the rest. A data layer is too big a tool here.
This is the moment to look seriously at a shared data layer, especially if you want to point AI at your own numbers. A model inherits your contradictions, and presents them with confidence.
Write down on one page, per concept, what it means at your company: customer, quote, project, revenue. Add which system is the source and who may change it. That takes an afternoon and needs no vendor. It is the first half step of what we call a data layer, and for some companies it is also the last.
To see what a data layer adds on top of that, read what is a data layer, an honest answer. If you recognise the pattern of separate systems that do not know each other, the related signs are in five signs you have outgrown your CRM. Why this test matters for AI too is covered in SMEs use AI for text, but rarely for their own numbers. How data is eventually moved and checked is in the data migration plan.
How we approach this, with an open in-house data layer where customer, quote and project exist only once, is described on the AI Native Data Layer page.
That depends on what you call a customer. At least one paid invoice, a signed quote, an active project, or just a contact person in your CRM? Without a written definition every system gives a different number. Ask three colleagues independently how many customers you have and where they got the number: if the numbers differ, what is missing is not the count but the definition.
Four things get mixed up: the definition of a customer, the reference date, duplicate records of the same customer and the source being counted. The CRM often counts prospects too, accounting only counts parties with an invoice, and a spreadsheet holds whatever someone once kept. None of those numbers is wrong, they just answer different questions.
First write down, on one page, what each term means and which system is the source for it. Only then check whether an integration is enough or whether you need a data layer. Differences you can explain are not a problem. Differences nobody can explain are, especially if you want to put AI on top.
The whitepaper on the AI Native Data Layer shows how to hold customer, quote and project in one place, what happens to your existing systems and how to start small.
Download the whitepaper