Find out where AI pays back before you spend on it
Plenty of pressure to do something with AI, very little honest guidance on what. We size the few opportunities worth pursuing, and name the ones that are not.
The short version
The hardest question is not how. It is where.
Almost every SME can list five places AI might help. Very few can say which one to start with. That is where the money goes.
- Every vendor’s answer is their product. Ask a platform company where AI helps and the answer is wherever their platform runs. You need a view that is not for sale.
- The obvious project is often the wrong one. The most visible process is rarely the most expensive. The real cost hides in rework, chasing and exceptions.
- A bad first project poisons the well. One failed pilot and everyone decides AI does not work here. The first choice matters more than the first date.
A document you can make a decision from
Not a deck of industry trends. A costed view of your business that a finance director and an operations manager can both act on.
- An opportunity map. Every candidate plotted on impact against feasibility, reasoning shown.
- A sized shortlist. Three to five, with cost, benefit and the assumptions behind both.
- An ROI model you own. The spreadsheet, not just the conclusion. Change our assumptions and see.
- A phased roadmap. What to do first, what it unlocks, and where you can stop.
- A risk and governance view. PDPA obligations, model risk, and the controls each opportunity needs.
- A build-or-buy answer. Where an off-the-shelf tool is enough we say so — including when that means not hiring us.
An assessment that can only say yes is a sales call
Two endings, and we are paid the same for either one.
About half of what gets suggested does not survive the first week
There is no interface to show you here, so here is the scoring sheet instead. Every candidate is put against these, including the ones we would rather build.
What tends to survive
- A named owner. One person whose week visibly changes if it works.
- A cost you can already state. Hours, rework, penalties, lost orders. A figure that predates us.
- Data already being written down. Imperfectly is fine. Not at all is not.
- A decision at the end. Something a person does differently once the answer arrives.
What tends to be cut
- “AI for customer insight.” No owner, no decision. Nothing changes on Monday.
- Anything blocked on a data cleanup. The real project is the cleanup. Price it as one.
- Work a rule already handles. If a filter solves it, we do not quote for it.
- Processes that run four times a year. The build outlives the payback.
Why there are no pages from a client assessment here
We will not show pages from a client assessment, redacted or otherwise, without written consent. The shape of a company’s problems identifies it even with the names removed.
The criteria above are ours, and they are the same ones used on every engagement.
Weeks, not months, because the answer decays
An assessment that takes a quarter is out of date when it lands. We keep it tight and evidence-led.
- 01Discovery interviewsThe people who own the processes, not only the leadership.
- 02Process and data reviewWhat the work involves, and what data already exists to support it.
- 03Opportunity workshopCandidates on the wall, argued through with you.
- 04Written assessmentMap, shortlist, model and roadmap, delivered and walked through.
Talk to us
Start with a conversation, not a commitment
Thirty minutes is usually enough to say whether a full assessment is worth your money — or whether you already know your first project and should get on with it.
Three ways this call ends
- You already know your first project. Skip the assessment. Most common outcome.
- You have candidates but no way to choose. That is what the assessment is for, and we scope it on the call.
- AI is not your bottleneck. Sometimes it is a process fix or a report nobody built. We tell you, and there is no invoice.