What Questions Should Leaders Ask About AI?
Most of the founders I talk to have already started. They opened ChatGPT or Claude a few months ago and did things they'd never done before: built a website, written their own job listings, turned themselves into a part-time developer overnight. A few weeks later, the same worries surface — where did the time go, the admin's half-automated but the funnel's gone dry, the ads aren't pulling, and now they reckon they need to learn SEO and marketing automation too, with no idea where to start.
We've stretched ourselves into work we couldn't have imagined doing before, got more done, and still nothing feels finished.
There's a lot of noise out there. Every week there's a new tool, a new model, a new thing you're apparently already behind on. So before getting to what leaders should do about AI, I want to answer a smaller question: what should they be asking? The trouble usually isn't the tool. It's that the tool arrived before the need was clear.
Start with the need, not the tool
When someone shows me what they've built with AI, it's rarely the building that's the problem. AI will happily make you capable of almost anything, which is the honey trap. You couldn't code, so it wrote you 600 lines to automate a spreadsheet, and three weeks on you're editing that script for the tenth time to patch yesterday's edge case.
Capability isn't the same as direction.
Going back to the drawing board sounds slow when everyone else seems to be shipping. It isn't. It's the only thing that stops you automating a process you shouldn't have been running in the first place.
It comes down to the product and the customers
Strip a small business back to the studs and two things matter: a product worth buying, and people who want to buy it. Everything else — the CRM, the automations, the clever Zapier chain, the dashboard you built one excitable evening — is pipe. Useful pipe, sometimes. But pipe.
So the first questions aren't about AI at all.
Get better at the thing you already make money from. If you're a baker, that's your recipes. If you run an NGO for underprivileged children, it's the clarity of your mission: knowing exactly what you stand for and saying it loudly. AI won't hand you that, and no competitor can borrow it.
Then get specific about who you're for. Who do you want to serve, and where do they already gather? You can't build a funnel for an audience you haven't named.
It helps to know your why, too. You chose this harder path over a steady 9-to-5 with a clear pay cheque for a reason. The answer tells you which work actually needs you, your real zone of genius, and which you can hand to a person or a system.
The one thing a bigger competitor can't copy
As an SME, you won't have the best product in the world or the biggest marketing budget. What sets you apart is you — a face, a personality, a human behind the work, and the specific problem you're good at fixing.
People help people.
Notice how every post, newsletter, and bit of content sounds the same now? AI converges on the average, and the Claude skill you bought last week for $97 (along with a few thousand other people) makes everyone sound identical. Your personality and the clarity of your message are what cut through.
The question worth sitting with: what's the human part of this business that only I can be, and am I protecting it or quietly automating it away?
Where is your time actually going?
The harder question, and the one most worth being honest about: where do your hours actually go, and your team's? Not where you'd like them to go. Where they go. Stress is a decent signal — the parts of the week that drain you are usually telling you something. So is the work that quietly eats a morning before you've noticed.
Some of it won't be anywhere near your priorities. Some of it will be work you genuinely enjoy that still isn't your best use of time. Enjoying something doesn't make it worth your hours. A founder who loves tinkering with the website can lose a week to it while the product and the leads sit untouched (I'm guilty of this too).
Shameless plug: I run a free time audit built around exactly this question. The line I hear most often once the hours are on the table: I've been doing everything myself, and my real hourly rate is barely above minimum wage.
Decide what's a person's job and what's a machine's job
This is what an AI policy is for, and it's simpler than it sounds: a clear view of what you'll hand to a tool and what stays with a person, decided on purpose rather than by whatever the software can grab.
IKEA is the obvious example. Their main operator put a chatbot, Billie, on the routine queries (order status, deliveries, opening hours), and it soon handled close to half of all customer contacts. Most companies would cut the call-centre staff and pocket the saving. IKEA instead looked at the half Billie couldn't handle, saw customers wanting help designing their rooms, and retrained those workers as remote interior design advisers. By their own numbers, that grew into a channel worth around €1.3 billion a year. The machine took the draining work; the people moved to the work only a person can do. You don't need their scale to ask the same question: what's tying your people up that a tool could handle, and what should only ever be a person?
Four questions to ask before you build anything
Once you're clear on the product, the audience, the human bit that's yours, and where your time actually goes, then you can think about tools. These are the questions I'd want a founder to answer before they build or buy anything.
What did I stop doing that was actually working — and why did I stop?
AI work is a constant back-and-forth, and it's easy to get carried away. So something that was working — email marketing, regular follow-ups, posting consistently — quietly slips, because your attention's gone to the new build. It wasn't failing. It just stopped being the priority.
That's the clearest use for AI or automation: the work that already earned its keep, not something new. The harder part is keeping it running once it's back, which usually means getting the operational backend off your plate so it doesn't depend on you finding a spare hour.
Is this genuinely unique to me, or has someone already solved it?
People reach for a custom build when a tool that costs a few euros a month would have done it better, and done it without them having to maintain it. Build custom when your need is genuinely specific. Otherwise, pay for the software and get on with your day.
If I build this, who maintains it?
The fun part is making the thing. The boring part is that it now needs feeding. A fragile workaround held together with a note-to-self so you remember how to run it next month isn't a solution. It's a problem you've postponed, and the kind that eats a lunchtime when it breaks.
Am I using AI to do more, or to do the right things?
This is the trap behind all the others. AI makes it easy to be busier: more content, more tools, more automations. Busier was never the goal. If a new thing doesn't move the product or the leads, it's just more pipe to look after.
I'm guilty too: I built something that already existed
A while back I wanted to know the real minimum cost of running a working RAG system on serverless infrastructure. (In reality, most of us, SMEs included, are never going to maintain our own servers.) So I built one for my book collection.
It needed rich descriptions to match my questions to the right books, so I had to enrich the data, turning each ISBN into a proper description and indexing it. Everything was set up, and I was still hunting for a reliable book API to do that enrichment when I found a service that did the whole thing. Upload your collection, it enriches every book, lets you query it. Exactly what I'd just spent a night building.
Familiar feeling, and not a comfortable one. But it taught me the thing I now repeat more than anything: your problems are almost never unique. Specific to your context, sure. The problem underneath has been had by someone else, and probably already solved. Find that out before you build, not after.
So, what should leaders ask about AI?
Not "what can AI do?" It can do almost anything, and that's the problem rather than the answer. The better questions are quieter. What am I actually trying to do? Is this the simplest way to get there? Have I protected the human part that makes this mine?
Keep it simple. Not everything needs a clever automation or a custom app. A good, working spreadsheet still does wonders for more businesses than anyone selling you AI would like to admit.
Get the product right and get clear on who it's for. Stay human. Then go looking for the tool — and check whether someone's already built it before you lose a weekend to it.