Get Your Information in Order Before You Add AI
Most businesses looking at AI start in the same place. They look for a tool, they try it on something, it produces an answer that is eighty percent right, and the project quietly stalls because nobody trusts the other twenty percent.
The tool is rarely the problem. The problem is that it was pointed at information the business had never cleaned up, agreed on, or written down.
Old advice, newly relevant
Feed a system bad inputs and you get bad outputs. That has been true of reporting for decades and it has not changed.
What has changed is the confidence of the output. A spreadsheet with a wrong total looks like a spreadsheet. A well written paragraph built on the same wrong total looks authoritative, and people are more likely to act on it without checking.
So the standard for the underlying information is higher now, not lower.
Four things worth sorting first
You do not need a data project. You need four fairly ordinary things to be true.
- One version of each record. If the customer list exists in three places with different spellings, anything built on it inherits the mess.
- Agreed definitions. What counts as an active customer, a completed job, a qualified lead. If two people would answer differently, no tool can resolve it for you.
- Written processes. A tool can help with a process you can describe. It cannot help with one that exists only in somebody's habits.
- Known ownership. Somebody responsible for each source, who can say whether it is current and what is in it.
None of that is glamorous, and all of it pays off whether or not you ever use AI.
A confident answer built on unreliable information is more dangerous than no answer at all.
Start with the boring cleanup
Pick the single data set that matters most, usually customers or jobs, and spend a day on it. Duplicates merged. Fields that should be a fixed list turned into a fixed list. Dead records archived rather than left in place. Obvious gaps filled or flagged.
Then write down the rules that keep it clean. Who can create a record, what is mandatory, what naming convention is used. Without that, it degrades again within a few months.
Be clear about what goes where
Before anyone pastes anything into a tool, the business needs a plain rule about what may leave it.
Customer records, employee information, contracts, anything a client gave you under an agreement, anything covered by a regulation in your industry. Decide what is allowed, what is not, and which tools are approved. Write it in one page of plain language rather than a policy nobody reads.
This is not a reason to avoid the technology. It is the thing that lets you use it without an awkward conversation later.
Pick a first use that you can check
The best early uses have two properties. They save real time, and a human can verify the output quickly.
Drafting a first version of a document you will edit anyway. Summarising a long thread so somebody can decide whether to read it. Turning rough notes into a structured procedure. Answering a question where the source is right there to check against.
Avoid, for now, anything that goes straight to a customer without review, or anything where being wrong is expensive and hard to notice.
Keep a person in the loop where it matters
Decide in advance which outputs get reviewed and by whom. That decision should be based on consequence rather than on how impressive the output looks.
Anything that affects money, a customer relationship, a legal obligation or a person's employment gets a human check. Internal drafts and summaries usually do not need one.
Write that down too, because otherwise the review step quietly disappears as people get comfortable.
Measure whether it actually helped
Before you start, note how long the task takes today and how often it has to be redone. Then check again after a month.
Some uses save a lot of time. Some save less than expected because the review takes as long as the original work did. Both are useful findings, and neither is obvious without a before and after.
The unglamorous conclusion
Businesses that get value from these tools are usually not the ones with the best tool. They are the ones whose information was already in reasonable shape, whose processes were written down, and who were clear about what they were trying to improve.
That work is worth doing regardless. It makes reporting more reliable, onboarding faster and handovers survivable, and it happens to be the thing that determines whether a new tool helps or just produces confident nonsense faster.
Getting the underlying processes and information into that shape is squarely what we do with clients. The interesting part usually comes after.
Thinking about where AI might fit? The groundwork is usually the useful conversation. Book a consultation, or visit www.expertechsolution.com.
