Where should an owner start with AI? Three decisions, then 90 days
AI is easy to try and hard to use well. The businesses getting real value from it did not start with a tool. They started with three decisions and a short, honest test, and they repeated it.
Most businesses start with AI the wrong way round. A tool is bought, staff are told to use it, and a few months later nobody can say what changed. It is not that the tools do not work. It is that nobody decided what they were for.
The evidence points the same way. The government’s January 2026 assessment found that business adoption of AI has more than doubled since late 2023, yet in firms that use AI, fewer than a third of employees do. Trying AI has become common. Changing the work has not.
Three decisions before any tool
- PurposeWhich one process, if it were faster or more accurate, would matter most this year?
- DataWhat information may AI use, in which tools, and what must stay out?
- PeopleWho owns each use of AI, checks what it produces and answers for it?
1. Purpose: pick the problem first
List the processes that cost the most time or cause the most errors, and pick one. In a small business the best candidates are usually in finance and administration: coding receipts, matching payments, chasing late payers, answering the same customer questions, drafting quotes. They happen often, follow rules and are easy to measure.
2. Data: decide what AI may see
Before anyone pastes anything into a chatbot, decide what is off limits. Personal data about customers and staff, client files and anything commercially sensitive should go only into tools you have checked: where the data is stored, whether it is used to train the model, and who can see it. Write it down in a one-page policy. It takes an afternoon and prevents most of the mistakes we see.
3. People: name who is responsible
Every use of AI needs an owner: the person who checks its output, notices when it is wrong and answers for the result. People also need to know what the change means for them. Be honest about what will change, involve the team in choosing where to start, and show that time saved is going into better work. Trust built early makes every later change easier.
A 90-day test
- Weeks 1 to 2Measure the process as it is today: time taken, cost and errors.
- Weeks 3 to 8Run the AI-assisted version alongside, with a person checking every result.
- Weeks 9 to 12Compare with the baseline and write down the new way of working.
- Day 90Decide: keep it, change it or stop it, and choose the next process.
Most of the saving arrives once the process, not just the tool, has changed.
Illustrative: a small business moving receipt coding to software that reads and suggests codes, with a person approving each one.
Show the data
| Before | 22 |
|---|---|
| Month 1 | 15 |
| Month 2 | 9 |
| Month 3 | 6 |
What good looks like after a year
A year in, the businesses doing this well have three or four processes running differently, a policy everyone knows and a habit of measuring before and after. None of it is dramatic. Together it adds up to a business that closes its books faster, answers customers sooner and spends more of its time on the work only people can do.
Help is available. The government’s October 2026 AI package includes £30 million for local AI partnerships to help businesses, small ones included, adopt AI and train their staff. Your accountant should be able to help with the finance processes, which is where most businesses start.
Sources
More on
This insight is general information, not advice for your circumstances. Talk to us before acting on it.