The sentence arrives early, usually with a shrug: this sort of thing is for big companies in big places. It is the most consistent assumption I meet, from Nouméa to Suva, and it has the causality backwards. The best-resourced organisations on the planet are failing at AI adoption at a rate that would be a scandal in any other capital programme — for reasons a six-person firm on an island does not have.
Start with a number that deserves to be better known. MIT's NANDA initiative published The GenAI Divide: State of AI in Business 2025 in August 2025, built on 150 interviews with business leaders, a survey of 350 employees and 300 public deployments. Roughly 5 per cent of enterprise AI pilots produce a measurable acceleration in revenue. The other 95 per cent return nothing that reaches the accounts, against a spend the report places in the tens of billions of dollars.
The authors are careful about the cause, and the answer is not the intuitive one. The models are not the problem. The failure is organisational — what the report calls a learning gap between generic tools and the way a company actually works. Pilots die in the distance between the demonstration and the workflow.
Three findings are worth setting side by side. More than half of generative AI budgets went to sales and marketing tools, while the returns researchers could trace sat in back-office work. Solutions bought from specialised vendors succeeded around 67 per cent of the time, roughly three times the rate of internal builds. And uptake of generic chatbots was near-universal while the tasks stayed trivial.
Each of those is a symptom of size. Budgets flow to the department with the loudest voice at the table. Internal builds stall because the people who understand the process and the people who write the code are four floors and two quarters apart. Generic tools stay trivial because no one is accountable for pushing them into the real work. Now count how many of those conditions exist in a company where the person choosing the tool is the person doing the job.
The same report records something more interesting than the failure rate. Around 40 per cent of firms had bought official enterprise AI subscriptions; roughly 90 per cent of employees were using personal AI tools for parts of their work anyway. The unofficial version delivered. The sanctioned one stalled in committee.
That gap is a serious governance problem for a bank and a non-event for a plumbing firm in Koné. There is no shadow usage to reconcile, no procurement cycle measured in quarters. Whoever notices that quoting eats four hours can decide that afternoon to do it differently, and will feel the difference on Friday. That closeness is not a consolation prize for being small. On this technology it is the scarce ingredient, and it cannot be bought.
The argument is not marginal, because here the small firm is not the exception — it is the population. Stats NZ counted 617,330 enterprises in New Zealand at February 2025, and 74 per cent of them had no paid employees at all. In Kanaky (New Caledonia), ISEE recorded 61,367 active enterprises on 1 January 2025, down 2.2 per cent in twelve months — a real contraction, which makes recovered hours matter more rather than less.
Wider still: the Asian Development Bank's Asia SME Monitor 2025 puts micro, small and medium enterprises at 99.8 per cent of all firms across its 26 developing member economies, employing 67.6 per cent of the workforce while producing 38.7 per cent of output. About 72 per cent of them operate in traditional services — trade, accommodation, food. Two thirds of the workers against under two fifths of the output is roughly where administrative drag lives, and that drag is what this technology removes most reliably.
| What kills the enterprise pilot | Does a six-person firm have it? |
|---|---|
| Budget captured by the loudest department | No — one person holds the budget |
| Gap between who knows the process and who builds | No — same person, same room |
| Procurement cycle measured in quarters | No — decided over one afternoon |
| Unofficial usage the company must reconcile | No — nothing to reconcile |
| A supplier bench within reach | Yes, and this one is a real disadvantage |
Making this case without the other side of it would be worthless. Small and remote carries real penalties, and three are structural. There is no local bench: few suppliers within a plane ride who have done the work before, which makes a poor first choice slower to recover from. There is single-person risk: the owner deciding everything is an advantage until the week the owner is unwell. And there is billing friction, since most of these tools price in US dollars, with the exchange rate sitting quietly in the margin.
None of those argue for waiting. They argue for keeping the first project small enough that a bad supplier costs a month instead of a year — which is precisely the discipline the 95 per cent never imposed on themselves.
It means something more specific. The researchers found that generic tools rarely survive contact with an entrenched workflow, and that the gap is organisational rather than technical. Where deployments were narrow, bought from specialists and aimed at back-office work, they did produce returns — that is the 5 per cent. The lesson is not to avoid the technology but to avoid the shape of project that failed.
On the evidence, less of one than assumed. Systems built internally succeeded at roughly a third the rate of solutions bought from specialised vendors, so an in-house engineering team is not what separates success from failure. What matters is that someone owns the process being changed and can tell whether it improved — and in a small business that person already exists.
The ADB puts about 72 per cent of MSMEs across its member economies in traditional services — trade, accommodation, food. Those businesses still produce quotes, chase payments, answer enquiries, order stock and file paperwork, and that is exactly the routine cognitive work where returns have been documented. The trade itself stays where it is; the paperwork around it is what moves.
Because the constraint is distribution, not capability. No vendor is flying to Port Vila to demonstrate anything, there is no procurement department to route a proposal through, and the tools carry no instructions for a firm of six. That is a gap in who brings the technology to the work, not a gap in whether the work would benefit.
Kanaky Tech is an AI automation agency working across New Zealand and the Pacific. Start with a free AI opportunity audit: we map how your business actually runs, rank what is worth automating, and give you a clear scope before anything is built — no obligation.