If you know AI is going to materially change your business but don’t yet know where to invest, you’re in the most common — and most dangerous — position an executive can be in right now.
Dangerous because the market is happy to sell you an answer before you’ve asked the question. Someone will build you a chatbot. Someone else will fine-tune a model. Someone will run a proof of concept that proves the technology works and proves nothing about your business. Six months and a budget line later, nothing has changed in your P&L.
The starting point isn’t a model. It’s a problem.
Be Wary of Any Conversation That Opens With the Technology
If an AI partner opens with “what model do you want us to use?” or “we can build you an agent for that,” you’re being sold to, not advised.
The better opening is the one they should be asking you:
What part of your business is expensive, slow, repetitive, error-prone, or dependent on scarce expertise?
Only once that’s answered does it make sense to ask whether AI can materially improve it. Data quality, governance, integration cost, and organisational capability determine whether an opportunity is real or theoretical — and none of that depends on which model happens to be fashionable this quarter.
Arrive With a Business Problem, Not a Solution
The strongest brief you can bring to an AI conversation looks nothing like a solution spec. It sounds more like: “We spend twenty hours a week doing X. Our people do Y manually. Customers keep asking us Z. I want to understand whether AI can change any of this — and which one is worth doing first.”
That gives a good partner something real to work with. It also protects against the two failure modes we see most often: AI theatre, an impressive demo that never touches operations, and proof-of-concept purgatory — three pilots, none in production, no measurable value.
What a Good AI Engagement Looks Like
Six stages, in order: discover the business and where money and time actually go; identify three to five candidate opportunities across operations, decision-making, customer experience, and knowledge work; prioritise them against value, feasibility, data availability, and risk; prove the top-ranked opportunity with one small working solution; measure whether it saved cost, grew revenue, or improved the experience against a baseline; then scale it properly, or stop and move to the next.
Organisations most often skip the first three stages — and those are exactly the ones that determine whether the last three produce anything worth having. It’s the same logic behind our GreenSpot Framework: evaluate technology maturity, organisational readiness, and commercial opportunity before capital is committed to a build.
Prioritise Ruthlessly, Then Build One
A proper prioritisation exercise should leave you with a short, ranked list of candidate opportunities — each scored against value and difficulty, not against how impressive it would look in a demo. The highest-value, lowest-difficulty opportunity is rarely the one that started the conversation.
Build only that one first. Measure it against a baseline. That’s the difference between “we need to do something with AI” and an actual AI strategy — one opportunity, proven in production, before anything else moves.
Where AI Meets the Physical World
Most AI consultancies stop at the software boundary — a model, an API, a chatbot. That’s fine if the opportunity is entirely digital. It isn’t fine if the business runs field service, manufacturing, logistics, or infrastructure, where value is created by people acting in physical environments.
A conventional AI engagement in field service produces a chatbot that answers questions. The version that actually changes the operation diagnoses the fault, retrieves the right technical documentation, determines the procedure, and guides the technician on site — using AR to identify the equipment and updating the system of record as the job is done. That needs AI, retrieval, spatial computing, and systems integration working as one, not as separate projects. It’s the kind of engagement we’re built for.
Common Questions
Where should a business actually start with AI? With a business problem, not a model. Identify the parts of the operation that are expensive, slow, repetitive, error-prone, or dependent on scarce expertise, then assess whether AI can improve them — ranked by value, feasibility, data availability, and risk. Build the top-ranked opportunity first and measure it against a baseline before scaling.
Should we build an AI chatbot first? Only if the highest-value opportunity is genuinely a customer-conversation problem. Most enterprise AI value sits in operations, logistics, decision-making, and knowledge work — not customer-facing chat. Chatbots are visible and easy to demo, which is why they’re over-represented in first projects and under-represented in projects with measurable ROI.
What’s the difference between a proof of concept and an AI strategy? A proof of concept demonstrates the technology can work. A strategy determines which problem is worth solving, whether the organisation can operationalise the result, and how the investment will be measured. Running proofs of concept without a strategy is how organisations end up with three demos and no production system.
Readiness First, Investment Second
None of this means AI is the wrong technology for your business — it usually means the sequence was wrong. Problem definition, data readiness, and a realistic measurement plan can all be built deliberately, in weeks rather than quarters, at a fraction of the cost of a failed deployment. If AI is the right answer for your highest-value problem, a good partner will help you scope it. If it isn’t yet, they’ll tell you why.