Is AI Saving Your Small Business Money? A Practical Test

An AI tool can produce a draft in seconds and still leave you with more work than expected. The useful test is how much time, money and effort it takes to finish the task to an acceptable standard.

For a small business owner, that distinction matters. Subscriptions are easy to buy. A genuine improvement in customer service, working capacity or profitability takes more thought.

My view is simple: assess the completed work. Count the preparation, checking and corrections as well as the time spent generating the first answer.

Updated 9 October 2026, following the ONS and ICO publications of 8 October.

What the latest UK AI figures actually tell us

The Office for National Statistics business survey published on 8 October 2026 found that 30% of businesses used AI in September, broadly unchanged from June.

Among businesses with at least 10 employees using AI, 38% identified administrative and clerical roles as the most affected, up three percentage points since June. In the same size group, 6% of AI users reported reduced headcount because of adoption, broadly stable over the previous year.

These are adoption and reported impact figures. They do not establish how much money businesses saved. The survey is voluntary and its findings are official statistics in development. The figures about roles and headcount should not be presented as results for sole traders, and an affected role is not necessarily an eliminated role.

For an owner, the practical question remains: what has improved in my business, and can I demonstrate it?

Measure the whole task

Suppose you use AI to draft a response to a customer enquiry. The process includes finding the correct information, preparing instructions, generating a response, checking the details and sending it. If the draft invents an availability slot or misstates a price, correcting it is part of the cost.

Compare the existing method with the AI assisted method using four measures:

• Total working time: include preparation, prompting, review, corrections and completion.
• Quality: record factual errors, rejected drafts, repeated work and any customer problems.
• Cost: include subscriptions, training, integration and supervision.
• Useful outcome: identify whether customers receive quicker answers, more work gets completed or a real expense falls.

Measure similar tasks with comparable complexity. Comparing an unusually difficult manual job with a straightforward AI task will give you an unreliable answer.

A worked example of AI cost savings for small businesses

The following figures are illustrative. They are not results from a Skills 2 Grow client or a claim about a particular product.

A business completes 60 similar administrative tasks each month. Each previously took 20 minutes, giving a total of 20 hours.

With AI, preparation and drafting take eight minutes per task. Checking and corrections take another seven minutes. Total working time is therefore 15 minutes per task, or 15 hours per month.

The business has released five hours. If the owner assigns those hours an assumed value of £20 each, that represents £100 of potential capacity value. Subtracting a £40 monthly subscription leaves an indicative £60 benefit before setup and training costs.

That calculation is useful, but it is not automatically a £60 cash saving. If wages remain unchanged, the subscription creates an additional cash expense. The owner still needs to decide what worthwhile work the five released hours will support.

Keep a separate record of initial setup costs and decide whether the expected benefit justifies them. If the tool needs substantial supervision or only handles a small share of tasks, the result may be less attractive.

Separate cash saved, capacity released and service improved

These are different benefits and should be recorded separately.

Cash saved means a real reduction in spending, such as less paid overtime or an external expense that the business can actually stop paying.

Capacity released means time becomes available for other work. That may be valuable even when payroll stays the same, provided there is a sensible use for the time.

Service improved could mean quicker responses, clearer documents or fewer missed follow ups. Record those outcomes directly instead of forcing every benefit into a financial estimate.

Avoid counting the assumed value of released hours and the additional profit earned from those same hours as two separate financial gains. That would overstate the return.

This distinction also matters when deciding whether to recruit. A few hours saved across unrelated tasks may not remove the need for a person with specific skills or availability.

Run a practical comparison before expanding

1. Choose one recurring task. Select work with a clear output and manageable consequences if the draft is wrong. Preparing wording alternatives or restructuring nonconfidential notes can be easier to assess than delegating decisions.
2. Record the baseline. Measure representative tasks using your current method. Note completion time and the standard the work must meet.
3. Test comparable work. Include every review and correction. Keep a person responsible for the final output, and record when the tool fails to produce usable work.
4. Assign the released capacity. Decide in advance whether the time will reduce overtime, clear a backlog, improve response times or support additional delivery.
5. Set a review date. Decide whether to keep, expand, modify or cancel the tool using your results. Repeat the assessment if task volumes, prices or the product change.

A short trial may miss seasonal demands and unusual errors. Where consequences are greater, use a longer assessment and stronger checks. A useful result on one task does not prove the tool is suitable for every part of the business.

Information handling belongs in the assessment

On 8 October, the ICO announced that ten major foundation model developers had made, or committed to make, data protection improvements. It also opened an agentic AI consultation, closing on 20 November 2026.

Those commitments do not certify every tool as suitable for customer information. Before introducing a tool, establish what information it receives, who can access it and what the provider does with it. Keep confidential or sensitive material out of experiments unless the arrangements have been properly assessed.

Verify claims about law, tax, health, finance and current events against authoritative sources. Include that checking time in your assessment. Seek appropriate specialist advice where the intended use requires it.

My view: make the tool earn its place

In my operations and process improvement work, the starting point is understanding how a task moves through a business: where it waits, what gets repeated and who is responsible for completing it.

Apply that same discipline to AI. Sometimes the best first improvement is a clearer template, better information or fewer approval stages. Sometimes an AI tool will make a useful difference once the process is clear.

A subscription is worth keeping when its benefits justify its full cost and the work meets the required standard. That judgement should come from your own results. As with sales growth and cash flow, more activity alone does not establish a better commercial outcome.

Sources: ONS Business Insights and Conditions Survey, 8 October 2026; ICO announcement and consultation, 8 October 2026. The practical assessment and worked example are Julian Frincu’s analysis. No measured client savings are claimed.