Skip to content

How to measure whether AI is actually working

6 min read
Toy robot measuring AI results with stopwatch, checklist, enquiries, chart and guardrail panels

AI should earn its place in your business.

That sounds obvious, but it is easy to forget when a new tool looks impressive. A chatbot that writes quickly, an automation that moves data around, or an agent that can search the web and draft emails might feel useful at first. The harder question is whether it is actually improving the business.

For SMEs and professional firms in Gloucestershire and Cheltenham, that matters. You do not need AI because everyone is talking about it. You need it if it saves time, improves service, reduces missed opportunities, makes work more consistent, or helps your team focus on higher-value tasks.

So before you roll out another AI tool, decide how you will measure whether it is working.

Start with the job, not the technology

The worst way to measure AI is to ask, “How much AI are we using?”

Usage is not value. A team can send hundreds of prompts and still achieve very little. Another business might use one small automation every day and save hours each week.

Start with the job you want AI to help with. For example:

  • responding to common enquiries
  • preparing meeting notes or follow-up emails
  • triaging support requests
  • turning messy notes into CRM updates
  • drafting first versions of routine documents
  • checking local information, feeds or sources
  • chasing missing details before a human gets involved

Once the job is clear, the measurement becomes much easier. You are not measuring “AI”. You are measuring whether that job is now being done faster, better, more consistently, or with less admin.

Measure hours saved, but be honest

Time saved is usually the first metric people think about. It is useful, but only if you measure it sensibly.

Do not rely on vague claims like “this saves loads of time”. Take one recurring task and estimate the current baseline. How long does it normally take? How often does it happen? Who does it?

Then compare the AI-assisted version.

If preparing a weekly client report used to take two hours and now takes 45 minutes with a human review, that is meaningful. If triaging ten enquiries used to take 30 minutes and now takes 10, that is meaningful too.

But include the hidden time: checking, correcting, prompting, moving data, and fixing errors. AI that saves 20 minutes but creates 15 minutes of checking is not a breakthrough. It may still be worth doing, but the numbers need to be real.

A simple measure is:

  • time before
  • time after
  • review time required
  • frequency per week
  • estimated monthly time saved

That is enough for most small businesses to make a sensible decision.

Look at response times

Many SMEs lose opportunities because they respond too slowly. Not because they do not care, but because everyone is busy.

AI can help here by sorting enquiries, drafting replies, summarising what a customer needs, or prompting the right person to act. The measure is not whether the AI wrote a nice email. The measure is whether the customer gets a better response sooner.

Useful response-time metrics include:

  • average time to first reply
  • number of enquiries answered within the same working day
  • time from enquiry to quote or next step
  • number of tasks waiting in someone’s inbox

This is especially relevant for professional services, trades, property businesses, consultants, agencies and local service firms. A quick, accurate, helpful response can be the difference between winning and losing the work.

Track missed enquiries and dropped balls

One of the best uses of AI is not glamorous. It is catching the things that slip through the cracks.

Missed website forms. Unanswered emails. Leads with no follow-up. Support requests that sit in the wrong inbox. Notes from a call that never make it into the CRM.

If AI helps reduce those, it is doing useful work.

Measure:

  • how many enquiries had no follow-up
  • how many leads were followed up late
  • how many tasks were created from calls or emails
  • how many CRM records were updated properly
  • how many customer requests needed chasing internally

This is where AI-first workflows can be more useful than standalone chat tools. In Stuart’s AI-first CRM experiment, the aim was not simply to have a chatbot inside a CRM. It was to explore agents that could work through tasks, split larger jobs into smaller ones, and help staff with day-to-day work. That sort of approach should be judged by whether important work moves forward without constant manual nudging.

Measure consistency, not just speed

Speed is helpful. Consistency is often more valuable.

If five people reply to enquiries in five different ways, the customer experience can vary. If internal notes are written differently every time, handovers become harder. If marketing drafts, proposal sections or support responses all start from scratch, quality depends too much on who is busy that day.

AI can help create a more consistent first draft, checklist, summary or process. That does not mean removing human judgement. It means giving the human a better starting point.

You can measure consistency by checking:

  • whether required information is included every time
  • whether tone and formatting are more uniform
  • whether handovers are easier to understand
  • whether fewer items need rework
  • whether staff are following the same process

For many businesses, this is where AI becomes quietly valuable. It makes the ordinary admin more reliable.

Include quality checks and guardrails

A measurement plan should not only count benefits. It should also track risk.

In the Cheltenham Times AI agent experiment, the AI-assisted workflow could find local leads, draft posts, create images, categorise content and publish into WordPress. It also needed rules, correction and human judgement. Some early outputs needed steering. There were lessons around relevance, image choice, geography and whether a thin story was worth publishing at all.

That is exactly the point for business use. AI can be fast, but fast is not the same as right.

So measure the checks too:

  • how many outputs needed correction
  • what type of errors appeared
  • whether any private or sensitive data was involved
  • whether the AI followed the agreed process
  • whether a human approved high-risk outputs
  • whether costs stayed within limits

A useful AI system should become more dependable over time. If the same mistakes keep appearing, the workflow needs better instructions, data, guardrails or human review.

Use a simple scorecard

You do not need a complicated dashboard to start. For most SMEs, a monthly scorecard is enough.

For each AI-assisted workflow, track:

  1. What job is AI helping with?
  2. How often does it run?
  3. How much time does it save?
  4. Has response time improved?
  5. Are fewer enquiries or tasks being missed?
  6. Is the output more consistent?
  7. How much checking is needed?
  8. What did it cost?
  9. What still needs human judgement?
  10. Should we keep, improve, pause or stop it?

That last question is important. Not every AI idea deserves to stay. Some will be useful. Some will be interesting but not worth the effort. Some will be risky until the process is tighter.

Good AI implementation is not about adopting everything. It is about finding the few places where AI genuinely improves the way the business works.

The practical next step

Choose one recurring admin-heavy task in your business and measure it for a week before adding AI. Then run a small AI-assisted version and compare the results.

If you would like help identifying the right first workflow, setting sensible measures, and building AI into your business without the hype, book a short AI consultancy call with Stuart Cole Consulting. A good starting question is simple: where could AI save time, reduce missed work, or make your service more consistent this month?

In Blog

Want help putting AI to work?

If this article sparked an idea, let's have a short call about what AI could do in your business — no jargon, no obligation.