How can I use AI in my business? A first-steps plan
Everyone says you should be using AI, and nobody says where to start. This is a five-step plan for picking one job, testing it properly and measuring whether it saved anything.

Why most first attempts stall
The usual pattern is that someone buys a subscription, tries a few prompts, gets mixed results and quietly stops. Nothing was wrong with the tool. There was no specific job, no measure of success and nobody responsible for checking the output.
The plan below fixes those three gaps. It is deliberately dull, because dull is what works.
Step one: audit your tasks, not your tools
Ask each person to write down what they do in a normal week, in plain language. Do not filter yet. You are looking for tasks that share three traits: they repeat often, they follow a pattern, and they involve reading or writing rather than physical work.
Typical finds are retyping data from emails into a system, answering the same customer questions, writing product or service descriptions, preparing weekly reports and checking documents for missing information. For worked examples of each, see our AI automation examples.
| Task | How often | Minutes each | Easy to check? |
|---|---|---|---|
| Retype supplier invoices | Daily | 5 to 10 | Yes, totals must match |
| Reply to “where is my order” | Daily | 3 to 5 | Yes, tracking data is factual |
| Write service page copy | Monthly | 60+ | Partly, needs an editor |
| Decide who to hire | Rarely | Hours | No, do not automate this |
The last row is the point of the exercise. Some jobs fail the test, and you should say so out loud.
Step two: pick one repetitive job
Score your list on frequency, time per task and how easy the result is to check. Choose the top one, and resist the urge to pick three. A single clear win teaches you more about your own business than a broad pilot does.
Good first picks are internal. A mistake in a draft report costs you nothing. A mistake in a customer refund costs you trust.
Step three: test it side by side
Run the AI version in parallel with the normal process for two to four weeks. The person still does the real work. The AI produces a draft alongside, and you compare.
Keep a simple tally: how many drafts were right, how many needed small edits, how many were wrong. Note why the wrong ones failed, because the reasons usually point to a missing instruction or a missing piece of data rather than a bad tool.
This is also where you decide whether a general chat tool is enough, or whether you need something wired into your systems. Copy and paste is fine for a one-off. For a daily job, the saving only appears when the AI is connected to where the data lives.
Rather have someone find the best first job for you?
Our Half-Day AI Audit is about 4 hours on site and gives you the top 3 to 5 opportunities, from £595.
See the AI audit Book a free call
Step four: measure hours saved
This is the step most people skip, and it is the one that tells you whether to carry on. Measure the old process first: minutes per item multiplied by items per week. Then measure the new one, including the time spent checking.
Be honest about checking time. If it takes six minutes to write a reply by hand and four minutes to review an AI draft, you saved two minutes, not six. Multiply by volume and by your real hourly cost to get a figure you can set against the cost of the tool or the build.
- Hours saved per week, after checking time.
- Error rate compared with the manual process.
- Cost to run, including subscriptions and any build.
- What the freed time was actually used for.
Step five: roll out in stages
If the numbers hold, move from side-by-side to AI-first with a person approving. Only after a few weeks of low change rates would we consider letting the simplest, lowest-risk items go through without approval.
Write down who owns the process, how to spot a problem and who to tell. That sounds like admin, but it stops a flow quietly going wrong for months. To understand where the human check fits, see what intelligent automation is.
What our own 1,300 hours taught us
We use AI heavily for our own work. Between June and November 2026 we estimate more than 1,300 hours of build and design work was done with AI, against roughly 140 hours of human direction and testing. These are internal time-log estimates, not audited figures.
The lesson is not that AI does everything. It is that the person directing it matters more than the tool. We write the brief, we test every result, and we throw away plenty. Projects like ASD OrderHub, which pulls four companies into one dashboard, only work because the structure was designed by people first.
We also learned to start narrow. Our best results came from tight, well-described jobs, not from broad requests like “improve our marketing”.
When you should not bother yet
If your processes live in people’s heads, document them first. If your volumes are small, a checklist might do more than any software. And if your customer data is held in a way that would breach your own policies, fix that before sending it to any outside service.
If you are unsure whether you are at the right stage, a short audit is cheaper than a failed pilot. Our No-Fluff Guarantee on audits means that if we cannot find savings worth at least the audit cost, you do not pay.
A realistic first month
Week one is the task list and the choice of job. Weeks two to four are the side-by-side test, with a simple tally of right, nearly right and wrong. At the end of the month you hold a short report: hours saved, error rate and cost.
That report answers the question for your business, not for businesses in general. It is also easy to show an accountant, a partner or a team who are sceptical. Evidence from your own work is far more convincing than a case study from someone else’s.
If the result is poor, that is still a useful result. You have spent a month and a small amount of money to learn that this job is not the right one, and you can pick the next candidate from your list.
Frequently asked questions
What is the easiest way to start using AI?
Pick one internal writing or sorting task and try an AI assistant on it for a couple of weeks, with a person checking every result. Keep notes on time saved and mistakes.
Do I need to be technical?
No, to start. Plain-language instructions go a long way. You need technical help when the AI has to connect to your accounting, CRM or shop, because that wiring is where the real saving sits.
Which AI tool should I use?
It matters less than the job you choose. Several providers are good enough for most business writing and sorting. We pick per task, based on accuracy, cost and how your data is handled.
Is it safe to put customer information into AI?
Only with care. Check the provider’s terms on how data is stored and used, send only what the task needs, and keep sensitive categories out entirely unless you have a clear reason and approval.
What does an AI project cost?
Audits start at £595 (half day) or £995 (full day, with a money-back guarantee). A 7-Day Discovery Sprint is £3,950, and implementation with six months of support starts from £8,500.
Start with one job, not a strategy
Tell us which task eats your week and we will help you test it properly, and say so if AI is not the answer.
Book a free 30-minute call See our prices