Picture a small plumbing firm. Five staff, a few hundred customers, a CRM someone set up four years ago. The owner buys an AI tool to write follow-up emails to past customers. Day one, it greets a man called "Dave Smith Plumbing Ltd" as "Hi Smith". It chases a customer who paid in March. It invites a woman who moved to Perth to book a boiler service in Leeds.
The AI did exactly what it was told. The list was the problem.
I'm going to be straight with you, mate. Most small business owners I talk to think AI trouble starts with the model. It usually starts with the spreadsheet.
The big companies are paying for this too
If you reckon this is a small business problem, it isn't. Fivetran released a report on 12 November 2026, based on a survey of 500 senior IT, data and AI leaders. It found that 39% of companies spend at least a quarter of their AI budget on data problems, things like fixing quality issues, rebuilding integrations and reprocessing data. Fivetran calls it the "AI tax". The same release says companies with high data costs are five times as likely to have at least a quarter of their AI projects delayed or stalled.
Do the maths on that. A quarter of the budget, gone before the AI does a single useful thing.
Stibo Systems surveyed 500 US business leaders in 2025. 91% said customer data management matters, but only 31% fully trust their data. So nearly everyone agrees it matters, and about two thirds don't trust what they've got.
Gartner has a nastier number. A summary of its Q3 2024 survey of 248 data management leaders says 57% of companies judged their own data unfit for AI. The same summary reports Gartner's forecast from February 2025 that 60% of AI projects built on data that isn't AI-ready will be scrapped.
Those are big firms with data teams. You've got Karen in the office and a shared drive.
Why it hits a small business differently
A big company has people whose whole job is data. You don't. Your customer data lives in five places: the CRM, an invoicing tool, someone's phone, an old spreadsheet and the head of the person who's been there longest.
That last one matters. In a five-person business, a lot of the "data" is knowledge nobody wrote down. Who's the decision maker. Which customer hates phone calls. Which supplier always ships late. An AI can't read your head. It reads what's in the system, and if the system is thin or wrong, it fills the gaps with confident guesses.
Here's the thing: a small business has an advantage. Your mess is small enough to fix in an afternoon. A company with 400,000 records needs a project team. You need a pot of tea and a free Thursday.
The four mistakes I'd check first
I'm not going to pretend this is glamorous. It's four boring checks.
1. Duplicates. The same customer entered three times, with three spellings. The AI treats them as three people. One gets the offer, one gets the reminder, one gets both. Sort them by email address and merge.
2. Dead contacts. Old numbers, bounced emails, people who left the company they were listed under. If you plan to let AI send anything, this is the one that bites. Every message to a dead address is wasted, and it can hurt your sender reputation.
3. Missing fields. Look at the columns you'd want the AI to use: name, last job, last contact date, status. How many rows are blank? If half the status column is empty, the AI can't tell a live customer from a lost one.
4. Old status. The customer who's marked "quote sent" from 2023. The job marked "open" that finished two years ago. This is the sneaky one, because the data looks fine. It's just out of date.
None of that needs software. It needs a filter, a sort and an hour of patience.
A test before you spend a penny
Before you buy any AI tool, run this on your own data.
Pull 20 records at random. Not your best 20, a random 20. For each one, ask: is the name right, is the contact detail current, is the status true today?
Count how many pass all three. If it's 18 or more, you're in decent shape. If it's 12, you've just saved yourself a nasty month. That's my own rule of thumb, not a study, so treat it as a rough gauge.
The point is you get a real answer in twenty minutes instead of finding out after the AI has emailed your whole list.
What the SMB surveys say
This isn't only an enterprise story. A HubSpot-hosted report on data readiness says 70% of companies named improving data quality as their top operational priority for 2025, and 45% said poor data structure was the leading barrier to AI success. People already know the problem. Plenty haven't done anything about it.
Service Direct's 2025 small business AI report found that businesses with AI budgets under $5,000 mostly hit integration (30%) and keeping up with AI advances (29%), while those spending $5,000 to $25,000 ran into data quality and accessibility (49%). In plain terms: the more you spend on AI, the more likely your data becomes the wall.
Which means cleaning up first is cheaper than finding out later.
My own version of this
I'm not above it. When I build automations for my own businesses, the bugs I find are rarely clever. They're a field that was empty, a tag that meant two things, a list with a contact on it twice. I wrote about one of those when a test on my own list emailed 39 people a fortnight late. The automation worked. The list and the timing around it didn't.
That's the pattern. The shiny part works. The boring part underneath is where you get burned.
A simple order of operations
If you want a plan, here's mine.
- Week one: run the 20 record test. Write down what fails.
- Week two: fix duplicates and dead contacts. Fill the blanks in the three or four fields you'd actually use.
- Week three: pick one small job for AI, like drafting follow-ups for customers who haven't ordered in six months, and run it on a handful of records first. Read every output.
- Week four: if the output is clean, widen it. If it isn't, go back to the data, not the prompt.
Notice what's missing. There's no "buy a platform" step. Ground your thinking in the system you've got. If you want the wider framework for how I think about building a business that runs itself, it's on the Autonomous Business page.
Who owns the data
One last thing, and it's the part that sticks. Pick one person to own the data. Not a committee. One name. They don't have to clean everything, they just have to be the one who notices when something's off and fixes the habit that caused it.
Most dirty data starts at the point of entry. Someone types a phone number into the wrong field, or skips the status update because they're in a hurry. Fix the entry habit and the list stays clean. Skip it and you'll be back doing this same afternoon every year.
The short version
AI doesn't fix messy data. It reads it faster and acts on it with more confidence than you'd like. The studies say big companies lose a chunk of their AI budget to this, and most of them don't fully trust their own data. You can beat them on this one, because your pile is small.
Spend an afternoon on the list. Then spend the money on the tool.
Sources
- Fivetran, The AI tax report release, 12 November 2026
- Stibo Systems, Blind Spots and Broken Promises, 18 November 2025
- Gartner figures as summarised by mybusinessfuture.com
- The AI Data Readiness Report 2025
- Service Direct, 2025 Small Business AI Report
Want to see how this could work in your business? Book a call and let's talk about where you're at and what's possible.
#BrewedByStevenPouredByHefti
About Steven Tann: Steven helps business owners build systems that run themselves using AI. After 10+ years helping 7,000+ businesses and building his own autonomous operations, he's the bloke who actually does it, not just talks about it. He's also the founder of Hefti, the AI operating system for small business. Find out more at steventann.com.