There is a lot of noise about AI right now, and most of it is unhelpful. One week it's going to replace your entire workforce; the next it's a useless party trick that makes things up. Both takes sell headlines. Neither helps you decide what to do on Monday morning. So let's skip the theatre and talk plainly about where AI already earns its keep in a real business, and, just as importantly, where it doesn't.
The honest picture is narrower and more useful than the hype suggests. AI isn't magic and it isn't a fraud. It's a very capable assistant that's brilliant at a handful of specific jobs and mediocre-to-dangerous at others. Knowing the difference is where the money is.
Across the businesses we work with, the same categories keep delivering real, measurable returns. None of them are glamorous. All of them save hours that go straight back into the work that matters.
Where AI reliably earns its keep
1. Drafting and summarising
This is the workhorse. AI is genuinely excellent at producing a solid first draft, emails, proposals, reports, job descriptions, and at compressing long things into short ones. Hand it a sprawling email thread or an hour of meeting notes and you get a clean summary with action items in seconds. The gain is real: writing from a draft is far faster than writing from a blank page. The caveat: it's a first draft. A human still needs to check the facts, fix the tone, and own the final version before it leaves the building.
2. Customer support
AI handles first-line questions well, the repetitive "how do I reset my password" and "what are your hours" queries that eat a support team alive. Wired into your own knowledge base, it can answer accurately and cover the after-hours gap when nobody's at the desk. The gain is faster responses and freed-up staff. The caveat: keep a clean handoff to a human for anything sensitive, angry, or out of the ordinary. The goal is a helpful front desk, not a wall customers can't get past.
3. Data wrangling and cleanup
This one is quietly transformative for SMEs drowning in messy spreadsheets. AI is strong at tidying inconsistent data, categorising long lists, extracting structured information from invoices or emails, and reconciling formats that don't match. Work that used to mean an afternoon of copy-paste becomes minutes. The caveat: spot-check the results, especially with numbers. AI is fast, not infallible, and a wrong figure in a clean-looking table is worse than an obvious mess.
4. First-pass analysis and insight
Point AI at a dataset or a pile of customer feedback and it will surface themes, outliers, and plausible explanations far quicker than a person scrolling rows. It's a superb way to get oriented and decide what's worth a closer look. The caveat: treat its conclusions as hypotheses to verify, not findings to act on. It's the start of the analysis, not the end of it.
5. Software development acceleration
For anyone building or maintaining software, AI meaningfully speeds up writing code, explaining unfamiliar systems, and hunting down bugs. Our own delivery is faster for it. The caveat is the same as everywhere else: a developer reviews, tests, and understands what ships. AI accelerates a skilled person; it doesn't replace the skill.
The pattern is consistent across every one of these. AI does the heavy lifting of the first 80 per cent, fast, and a human owns the last 20 per cent that decides whether the work is actually any good.
Measure it honestly
Before you call any AI rollout a success, count the real numbers: hours saved per week, response times, error rates before and after. If you can't point to a concrete gain, you've bought a demo, not a result. Honest measurement is also what tells you when to expand a use-case, and when to quietly kill one.
Where it's still mostly theatre
Now the other half of the map, because pretending AI does everything is how businesses waste money. Be wary whenever someone promises these:
- Fully autonomous decisions. Letting AI act on its own in anything that touches money, contracts, or customers, with no human in the loop, is asking for an expensive, confident mistake.
- Guaranteed accuracy without checking. If a task demands the answer be exactly right every time and nobody verifies it, AI is the wrong tool. It produces plausible, not certified.
- Replacing judgement and relationships. The strategic calls, the hard conversations, the trust you've built with a client, these are the human core of the business. AI can support them; it can't be them.
- Vanity chatbots. The bot bolted onto a website because competitors have one, that nobody actually uses and that frustrates the few who try. That's theatre with a maintenance bill.
None of this is a reason to hold back. It's a reason to be deliberate. The businesses getting real value aren't the ones chasing the flashiest announcement, they're the ones who matched a capable tool to a job it's genuinely good at.
How to actually start
Don't try to "do AI" across the whole business at once. That's how pilots stall and budgets evaporate. Do this instead:
- Pick one or two real, painful, repetitive jobs from the list above, ones you do often and don't enjoy.
- Put AI to work on them for a few weeks, with a human keeping the final say.
- Measure the difference honestly: hours saved, errors caught, response times.
- If the numbers are real, expand. If they're not, drop it and try the next candidate.
That's the whole method. Pick a real job, prove the value, then widen. It's unglamorous, and that's exactly why it works. The hype will keep cycling. Your returns will come from the boring, well-measured wins, and there are plenty of those waiting in the work you already do.
