How to use AI in your business: start with the work, not the technology

Most businesses asking “How can we use AI?” are starting with the wrong question.

The useful question is: what is happening repeatedly in this business that is slow, expensive, inconsistent or dependent on someone remembering to do it?

Find that first. Then decide whether AI is actually the right fix.

Where should a business start with AI?

Start with one real workflow. Write down what triggers it, who touches it, what information they need, what decision gets made and what “done” looks like. Then measure how often it happens and how much time it consumes.

This matters because AI is only one tool in the box. A problem might need an AI model, a conventional automation, a better form, a CRM rule, an API connection or simply a clearer process.

What business processes are worth automating?

The best candidates tend to be frequent, rules-heavy and measurable. They often involve moving information between systems, researching the same things repeatedly, producing a first draft, classifying incoming work or chasing the next step.

AI vs automation: what’s the difference?

Automation follows defined rules: when X happens, do Y. AI is useful where the task involves language, classification, extraction, summarisation or judgement that cannot be expressed as a simple rule.

The strongest systems often use both. Conventional automation moves the work reliably; AI handles the part that needs interpretation.

What shouldn’t you automate with AI?

Don’t automate a process you don’t understand. Don’t use AI merely because a task is annoying. And don’t hand a high-consequence decision to a model without the controls, data and human oversight the consequence deserves.

A rare five-minute task is usually a poor automation target. So is a broken process: automating it simply makes the failure happen faster.

How do you calculate whether AI automation is worth it?

Start with a basic labour calculation:

frequency × minutes per task × people involved × loaded hourly cost.

Then add what the spreadsheet misses: errors, delays, missed follow-up, rework and the opportunity cost of skilled people doing low-value work.

That gives you a baseline. Compare it with the build cost, ongoing software cost, maintenance and the realistic percentage of work the system can remove.

A seven-step AI audit you can do yourself

1. 1List the repeated work.Look across sales, marketing, operations, customer service, finance and recruitment.
2. 2Measure it.Estimate frequency, handling time, people involved and failure rate.
3. 3Find the friction.Look for waiting, copying, re-keying, searching, drafting and chasing.
4. 4Rank the opportunities.Prioritise high-frequency work with a clear output and measurable value.
5. 5Choose the simplest fix.Process change first; automation second; AI where interpretation adds value.
6. 6Prototype one workflow.Prove it on a contained process before connecting half the company.
7. 7Measure the result.Compare time, cost, quality and errors against the baseline.

When should you bring in an AI automation consultant?

DIY is perfectly sensible for low-risk personal productivity. External help becomes more useful when the workflow crosses several systems, handles customer or commercially sensitive data, depends on APIs, needs monitoring, or will become business-critical.

The consultant’s first job shouldn’t be selling you an agent. It should be working out whether an agent is needed at all.

The rest of the system matters

AI rarely lives alone. If the problem is poor lead quality, read why Google Ads leads don’t convert. If you cannot trust the measurement, start with why GA4 and Google Ads conversions don’t match. And if traffic arrives but nothing happens, use our guide to website traffic with no enquiries.

Frequently asked questions