Where AI actually creates value inside a growing business

Most businesses meet AI the wrong way round. Someone sees a demonstration, the demonstration is impressive, and the question becomes "where can we use this?" That question has no natural end. It produces pilots that nobody asked for, a subscription nobody cancels, and a quiet conclusion six months later that AI was overhyped.

The useful question is the opposite one. Not "where can we use AI?" but "what does this business do over and over, badly, because a person has to do it?" Answer that honestly and the list of places AI pays for itself is short, specific and boring. Boring is the point.

Three shapes worth looking for

Almost everything that works falls into one of three shapes.

Work that repeats with small variations. Writing the same kind of reply to the same kind of enquiry. Turning a week of notes into a summary. Producing the caption, the description and the alt text for twenty new products. A person doing this is not making decisions, they are applying a pattern, and applying a pattern at speed is the one thing these tools are genuinely good at.

Work where the input is messy and the output is structured. A WhatsApp thread that needs to become a record in the CRM. A stack of invoices that needs to become a spreadsheet. Six months of enquiries that need to be sorted into what people actually asked for. This is the work that never quite gets done, because it is tedious and nobody's job.

Work that stalls because a first draft is hard to start. A proposal, a job description, a policy, a follow-up sequence. The value is not that the machine writes it well. The value is that a mediocre draft in ninety seconds unblocks a person who would otherwise have left it until Friday.

Three shapes to leave alone, for now

Equally worth naming, because ignoring them is where the money goes.

Anything where being wrong is expensive and hard to notice. Pricing, eligibility, anything legal or medical, anything that touches somebody's money. Not because the tools cannot help, but because the cost of a confident wrong answer swamps the saving, and confident wrong answers are the failure mode.

Anything where the real problem is that nobody agreed what to do. If two departments disagree about who owns a lead, an AI that drafts follow-ups faster will produce a faster argument. Automation fixes execution, not disagreement.

Anything with no feedback loop. If nobody will ever check the output, do not generate the output. Volume without review is how a business ends up with a hundred pages of content it cannot stand behind.

Start with the week, not the tool

A practical way in, and one that costs nothing: for one week, ask each person to note every task they did more than three times. Not a time-and-motion study — just a list. At the end of the week you will have a page that describes your business more accurately than any strategy document.

Then, against each line, ask three questions:

1. Would a mistake here be noticed the same day?
2. Is there a person who would review the output anyway?
3. Does the task have a clear input and a clear finished state?

Three yeses is a candidate. One no is usually a reason to wait.

What it actually looks like when it works

It rarely looks like a chatbot. It looks like a new enquiry arriving, being read, being summarised into four lines, being categorised, and landing in the right person's list with a suggested reply already drafted — while a person still decides what to send. It looks like the monthly report being assembled on the first of the month instead of being reconstructed on the fifth. It looks like nobody retyping anything that already exists somewhere else in the business.

None of that is exciting to describe. All of it is time back.

The test of a good AI use case is not how impressive the demonstration is. It is whether anybody notices when you switch it off.

The part people skip

The tooling is the easy half. The half that decides whether this lasts is whether the work actually changes shape around it. If the drafted reply still waits three days for someone to look at it, nothing has improved — the delay has simply moved. If nobody is accountable for reviewing what comes out, quality drifts until the team stops trusting it, and a tool the team does not trust is a tool the team routes around.

So before buying anything: pick one task, name the person who owns it, decide what "good" looks like, and agree how you will know in a month whether it helped. That is a smaller commitment than most businesses make on a hunch, and it is the difference between a habit and a subscription.

Where to start this month

If you want one place to begin, begin where information changes hands. The handoffs — enquiry to reply, conversation to record, work done to report written — are where time leaks and where the pattern is clearest. Fix one of those properly and the second one is easier, because you will have learned what your business actually needs rather than what a demonstration suggested.