AI Can Help With Everything. That Is the Problem.

"The businesses getting value from AI are not doing more. They are choosing fewer things and finishing them."

I spend half my time getting people excited about what AI can do.

I spend the other half slowing them down.

That sounds like a contradiction, and for a long time I treated it as one. Two different problems needing two different conversations. The leader who tried AI a couple of times, got mediocre results and quietly stopped. And the leader who has seen what is possible and now wants to rebuild eight workflows at once.

They look like opposite problems. Underneath, they are the same failure. Neither has made one deliberate, bounded choice and stayed with it long enough for it to pay.

That choice is where most AI value is won or lost. Not prompting. Not tooling. Not which model you are paying for. And almost nobody teaches it, because it is not a technology skill.

Here is why it is harder than it sounds.

Every business tool you have ever bought came with a bounded purpose. Your accounting system does accounts. Your CRM holds customers. The scope of the decision was small, because the tool decided most of it for you.

AI arrives with no boundary at all. Draft the board paper. Analyse the tender. Write the job ad. Rethink the pricing model. Plan your daughter's twenty-first. It says yes to everything, immediately and competently enough to be encouraging.

So the constraint moves. It stops being what the tool can do and becomes what you choose to do with it. No previous business tool has presented that choice this broadly, or made every option feel this immediately achievable.

I sat with a leader recently who had a list. AI was going to rebuild the proposal process, tidy up the monthly reporting pack, handle meeting follow-ups, improve onboarding, and take a look at pricing while it was there. Every item on that list was worth doing.

Six weeks later I asked how it was going. All five had been started. Not one of them was in daily use by anyone other than him.

Nothing on that list was beyond what the tools could do. He just never decided which one to finish first, and that it is the most common failure I see.

The opposite failure is quieter but just as expensive. A leader gives AI something genuinely difficult on the first attempt, usually a task carrying a lot of context that has only ever lived in their head. They give it a fraction of what it needed. The output comes back generic. They conclude the technology is overhyped and stop. They were two or three honest attempts from something useful.

Both leaders spent real effort. Neither built anything that compounds. And both now carry a quiet suspicion that this was oversold.

If you do not define the outcome first, AI will help you do the wrong thing faster.

The discipline that fixes this is not new and it is not glamorous. Define the outcome before you touch the tool. Scope what is in and what is out. Break the work into steps and find the ones that actually hurt. Decide who owns it. Decide how you will know it worked. Finish it before you start the next one.

Good project management has always forced this discipline. If you have someone who is genuinely good at it, use them. But two decisions stay with whoever is accountable for the result: deciding what is worth doing in the first place, and staying close enough to know whether it actually worked. Hand over the execution, not the judgement.

So what do you point it at first?

Look for work that is frequent. Work that repeats often enough to learn from. Weekly beats annually, because repetition creates both return and feedback.

Look for work that is painful. Work that consumes time, creates delay, or repeatedly produces rework. If nobody minds doing it and nothing material changes when it improves, it is probably not the place to start.

Look for work that is bounded. A clear start, a clear finish, and a recognisable answer to "did that work?" If you cannot tell whether the output is any good, you have picked something you cannot yet learn from.

Frequent, painful, bounded. By that test most of the highest-value AI work in a business is unglamorous, which is exactly why it loses out to the exciting thing that never ships.

The same three questions work at every level. They are the same questions whether you are choosing the one process your operations team will rebuild this quarter, or the one task on your own desk you will stop doing by hand. Most people apply some version of this thinking to the organisation and never once apply it to their own week. That is why so much individual AI use stays at the level of interesting afternoons.

Then the part almost everyone skips. Embed it. Run it until it is genuinely how the work gets done, not a clever thing someone demonstrated once. That is where capability starts to compound, and it is the point most people never reach, because by then something more interesting has come along.

So before you approve the next licence, pilot or workflow, ask one question in the room: what are we willing to finish before we start anything else?

If nobody can answer that, you do not have an AI problem. You have a decision you have not made yet.

AI will help you do almost anything. It will not tell you what is worth doing.

Find the one AI opportunity worth finishing

Most businesses do not need more AI ideas. They need one clear priority with a defined outcome, an owner and a way to measure whether it worked.

The 5 Minute AI Readiness Snapshot helps you identify where to focus first, so your next AI initiative has a better chance of being finished, embedded and useful.

Take the 5-Minute AI Readiness Snapshot

--

About the Author 

Mark Lucas is Co-founder at Generation AI, where we help New Zealand business leaders build real AI capability.

Information provided is general in nature and does not constitute legal advice