When we visit a company for the first time, they almost never say "our data is a mess." They say "we want to add AI here." But as soon as we start looking underneath, the same signs show up again and again, across completely different industries. This article is that list, with what each sign actually means and what to do about it before you spend a single euro on an AI project.
This isn't a list meant to scare you. It's a list meant to save you money: if you recognize several of these signs, the problem isn't that you need "more AI," it's that you need to sort out what you already have first. We go deeper on why in why data comes before AI.
The 10 signs
1. Two reports, two different numbers, for the same question
You ask "how much did we sell last month?" and the ERP says one thing, management's spreadsheet says another, and accounting says a third. Nobody's surprised, because it's already normal. That "already normal" is the most serious sign of all: it means nobody in the company fully trusts any number.
2. There's a spreadsheet that "only one person understands"
Almost every company has one: the sheet someone maintains by hand, that nobody else knows how to update, and if that person takes two weeks off, the process stalls or gets done wrong. That's not a tool, it's a single point of failure disguised as one.
3. The same customers or suppliers exist multiple times, under different names
"Smith Pharmacy," "Smith Pharmacy Ltd." and "SMITH PHARMACY" as three separate records. It looks like a minor detail until you try to calculate how much that customer actually buys, or build anything automated that depends on identifying them correctly.
4. Nobody knows, for certain, where a key figure comes from
You ask someone "how is the margin shown here calculated?" and the answer starts with "I think..." If the "how it's calculated" depends on one person's memory instead of a documented definition, that figure isn't reliable, it's a habit.
5. Systems don't talk to each other (and someone bridges the gap by hand)
ERP, CRM, accounting and the distributor's system each live on their own island, and someone copies and pastes numbers from one to another. That manual work doesn't just cost hours: it's where most errors sneak in, because nobody audits a copy and paste.
6. Invoices get entered by hand, typing in what was already written on a PDF
If your admin team spends hours a month typing in line items, amounts and taxes that were already printed on the supplier's invoice, that's not a problem of people working too little, it's a process that shouldn't require typing at all. We solved exactly this for a veterinary group: see the case in invoices that book themselves.
7. Returns, promotions or discounts "don't always subtract correctly"
It's a sentence we hear literally worded that way, "it doesn't always subtract right," as if it were just part of the landscape. When a business figure has a known, accepted exception, that figure is wrong, period. Nobody has just dared to quantify by how much.
8. Purchasing or stock decisions get made "by feel" more than by data
If the person in charge of purchasing decides how much to order based on what they remember from last month, instead of a reliable figure on real sales and turnover, the company is running on expensive intuition. Not because intuition is worthless, but because it should complement correct data, not replace it.
9. Important notifications or alerts get read late, or not at all
Tax notifications, critical stock alerts, supplier emails about price changes: if someone has to log into several different places every day "just in case," sooner or later something important gets read three days too late. We saw this with tax office notifications at an advisory firm: that case is here.
10. You've already tried adding AI once, and "it didn't work"
This is the sign we hear about most. They hired a generic chatbot or assistant, connected it to their systems as best they could, and it stopped being used within a few months because it gave odd answers, or flat-out wrong ones. Almost always the cause wasn't the AI itself: it was that it got connected to data that wasn't ready.
Summary table: what to do with each sign
| Sign | What it means | Realistic first step |
|---|---|---|
| Different numbers for the same thing | No single source of truth | Define one source and validate it against reality |
| Spreadsheet only one person understands | Fragile, undocumented process | Move that process out of the spreadsheet into the system |
| Duplicate customers or suppliers | Lack of data normalization | Unify records before automating anything on top of them |
| Nobody knows how a figure is calculated | Missing business definition | Document the formula and who validates it |
| Systems that don't talk to each other | Lack of real integration | Connect by code, not manual exports |
| Invoices typed in by hand | Avoidable manual process | Automate extraction and matching |
| Accepted exceptions in calculations | Structural error disguised as normal | Audit the calculation from scratch, once |
| Decisions made "by feel" | Lack of actionable data | Put the reliable figure in front of the decision maker |
| Alerts that arrive late | Lack of automated monitoring | Centralize and classify before alerting |
| AI that "didn't work" before | The mess got automated, not the process | Start over from the data, not the model |
If you recognize three or more
That's not a grim diagnosis, it's the most honest starting point you can have. Most of the projects we run at AutoBoost start exactly like this: a company that recognizes two or three of these signs and wants to stop living with them. The difference between the ones that move fast and the ones that get stuck isn't budget, it's whether they start in the right place.
Our approach in services always starts by looking at the real data first, even if the initial request is "I want a chatbot" or "I want AI in the ERP." And if you want to see what doing it right leads to, success stories has concrete examples, not promises.
How many of these ten do you recognize in your company? Tell us here and we'll tell you, with no strings attached, which ones matter most in your case and where we'd start.
