A small business can run its own AI readiness assessment on one page in an afternoon: list the tasks that eat your week, score each on time, clarity and risk, and pick the top two.
You do not need a consultant, a maturity model or a six-week project for this. You need the people who do the work, a spreadsheet and an honest answer to "what would we stop doing by hand if we could?"
This article gives you the template, the scoring rule, a worked example and the questions to ask before anything gets built.
Why start with tasks and not with tools?
Because the main thing that stops UK firms using AI is not cost or technology, it is not knowing where it would help.
The Office for National Statistics reports that about 35% of UK businesses were using at least one AI technology in June 2026, almost three times the share in late 2023.
In its earlier analysis of barriers, the most common were difficulty identifying use cases, cost and lack of expertise. The ONS analysis of AI in UK businesses has the details.
A list of tasks fixes the first of those, and usually shrinks the other two. Once you know the job, you can see whether a tool you already pay for covers it.
Many "AI readiness" checklists start with data strategy, governance and culture. Those matter in a large organisation. For a firm with five to fifty people they delay the useful part, which is finding the two or three jobs that repay automating.
What goes on the one-page template?
One row per task, across eight columns.
The columns cover the task, who does it, how often, how long it takes, whether the steps are clear, what data it touches, what happens if it goes wrong and whether existing software already offers a fix.
Copy this into a spreadsheet:
| Column | What to write |
|---|---|
| Task | A verb and an object, such as "chase overdue invoices" |
| Owner | The person who does it now |
| Times per week | A count, not a feeling |
| Minutes each | Time it once, do not guess |
| Rules | Clear, partly clear or judgement |
| Data | None, business data or personal data |
| Error cost | Low, medium or high if it is wrong |
| Already covered | Does software we pay for have this feature? |
Fill in 15 to 25 rows. Ask each person to list the tasks that repeat, annoy them or get skipped when it is busy. Skipped work is a better signal than complaints, because it tells you what is already costing money.
Do not list strategic work. Pricing decisions, hiring and client conversations are not candidates. The template is for the repeated, rule-bound tasks that sit around them.
How do you score the tasks?
Turn each row into weekly hours, then discount it for unclear rules, personal data and the cost of an error.
The method has four steps:
- Weekly hours = times per week x minutes each / 60
- Rules: multiply by 1 if the steps are clear, 0.5 if partly clear, 0.1 if it depends on judgement
- Error cost: multiply by 1 if low, 0.7 if medium, 0.4 if high
- Data: multiply by 1 if none or business data, 0.7 if personal data, since it needs extra care
Add a note rather than a multiplier for "already covered". If your accounting or booking software already does the task, switch that feature on first and take the row off the list.
The numbers are a ranking aid, not a business case. The point is that a task that happens daily with clear rules and low risk should beat one that is large but vague.
What does a worked example look like?
Here is what the rows might look like for a ten-person service firm.
| Task | Per week | Minutes | Hours | Rules | Error | Data | Score |
|---|---|---|---|---|---|---|---|
| Acknowledge and log new enquiries | 25 | 6 | 2.5 | 1 | 0.7 | 0.7 | 1.2 |
| Chase overdue invoices | 15 | 8 | 2.0 | 1 | 0.7 | 0.7 | 1.0 |
| Write first draft of proposals | 5 | 45 | 3.8 | 0.5 | 0.4 | 0.7 | 0.5 |
| Update client records after calls | 20 | 5 | 1.7 | 0.5 | 1 | 0.7 | 0.6 |
| Weekly figures report | 1 | 90 | 1.5 | 1 | 0.7 | 1 | 1.1 |
The proposals row has the most raw hours and ends up last. An AI agent could help there, but the rules are partly judgement and the error cost is high, so it comes after the plain ones.
The top two, enquiry handling and the weekly report, are the sort of workflow we cover in five automations that save a small business the most hours.
Who should fill it in, and what goes wrong?
The people who do the tasks should fill it in, with the owner or manager checking the scores afterwards.
If the owner writes the list alone, it is full of tasks the owner dislikes rather than tasks that consume time. Ask each person to keep a tally for one week, and use that, not memory, for the "times per week" column. People routinely underestimate small repeated jobs and overestimate rare big ones.
Three mistakes show up again and again. The first is scoring tasks you would like to do rather than tasks you do. The second is skipping the "already covered" check, then paying for a build that duplicates a setting.
The third is treating the score as a verdict. If a row feels wrong, say so and talk about why, because the people closest to the work often know about an exception the numbers cannot see.
It also helps to look for work that people do around the system rather than in it: retyping from one screen to another, copying into a spreadsheet, forwarding an email for approval. These are the cheapest automations, and they rarely appear on an official process map.
Which questions should you ask before you build anything?
Six, in this order, and a no to any of them means stop and fix that first.
- Is the process written down, step by step, including the exceptions?
- Does someone on the team own it and agree it should change?
- Does it use personal data, and do we know where that data goes if a tool processes it?
- What is the worst thing that happens if it goes wrong, and who finds out?
- How will we know it has worked, in hours or pounds?
- Is there a simpler answer, such as a setting in software we already own?
Question three is where small firms trip up. Before anyone pastes customer details into an AI tool, decide whether the tool should see them at all. For firms handling sensitive client matters, such as law firms, it should be settled before the first build.
The government has also published a tool for this side of things. The Department for Science, Innovation and Technology ran a consultation on its AI Management Essentials self-assessment, which is aimed at helping organisations, especially SMEs and start-ups, put responsible AI management processes in place.
The consultation closed in January 2025 and the government response followed in February 2026, so check the consultation page on GOV.UK for the current status before relying on it.
Should you build it, buy it or switch on what you have?
Check what you already own first, then look for an off-the-shelf tool, and build only when neither fits.
Many CRMs, accounting packages and booking systems now include automation and AI features that cost nothing extra or little. Retailers and shop owners often find this with their ecommerce platform, since ecommerce brands and retailers use systems that already send abandoned-cart and order messages.
When nothing fits, the decision between a subscription and a bespoke build is mostly about volume and fit. We explain how we think about that in build vs buy for AI tools.
Keep the unscored list too. Tasks that score badly today, such as the proposals row above, are often the ones that become worth doing six months later, once the simple automations are running and you trust the process. Revisit them each quarter and see whether the rules have become clearer or the error cost has dropped.
What happens after the audit?
Pick the top two, set a target in hours, and run each as a four to six week trial with one named owner.
Write down the baseline before you build. Check at the end of week two whether the numbers are moving. Keep the audit sheet and refresh it every quarter, because the answers change as the business does and as software adds features.
If you want a second pair of eyes on your list, the sheet is also a good brief for an agency. It lets a supplier quote a real job instead of describing "AI transformation". Our wider view on choosing one is in what people mean when they search for an AI agency.
We run this kind of review as the first step of our AI automation services. If you would like help scoring your own list, get in touch and send us the spreadsheet.
