The demo worked. At head office, with your best staff, on clean data, the AI thing did exactly what the agency promised. Then you rolled it to branch three, and it quietly fell apart. If you run a multi-outlet group, you've probably lived some version of this.
The reason it broke isn't the model. It's that a single-location pilot and a multi-branch rollout are two different problems — and almost nobody sells for the second one.
The one-line version
A head-office pilot proves the tool can work. It says nothing about whether it works the same way in every branch — which is the only thing that matters for a group. Multi-outlet AI dies in the gap between "worked in the demo" and "works in all eleven locations, every Tuesday, with the staff you actually have."
The four ways it breaks across branches
- Fragmented systems. Head office ran one POS, one booking tool, one clean spreadsheet. Branch three is on a different POS, branch seven still runs on WhatsApp and memory. The pilot assumed one system; the group has several.
- Data that isn't the same per location. The tool was tuned on head-office customers, menu, or patient mix. Another branch has a different clientele, different peak hours, different edge cases. The model that looked sharp at HQ gets vague where the data shifts.
- Staff turnover eats the champion. The pilot worked because one motivated person drove it. Branches have turnover, and the next hire was never trained on it. Six months later it's a login nobody uses.
- No one owns "the same, everywhere." Head office had attention on it. No single person owns whether it's live, correct, and used in every branch — so it degrades location by location until it's shelfware with a subscription.
None of these are model problems. They're operational ones — the exact failure mode MIT's NANDA research points at: generic tools "stall in enterprise use since they don't learn from or adapt to workflows," and the projects that die do so for organisational reasons, not technical ones (MIT NANDA, 2025). For a multi-branch group, that stall just happens in more places at once.
Why generic AI advice misses this
Most AI advice — and most AI vendors — are written for a single business: one location, one owner, one system. A multi-outlet operator doesn't have an AI adoption problem so much as an AI distribution problem: getting one capability to work consistently across branches that are never quite identical. That's a different discipline, and it's the one almost nobody has productised.
It's also why "we ran a successful pilot" is the wrong bar. A pilot that can't survive branch two was never the win it looked like.
What actually works for a group
Not a bigger pilot. An owner whose job is making AI work across every branch the same way — mapping the real systems in each location, deciding what's standard vs. what flexes per branch, and tracking it monthly so it doesn't quietly decay after the launch meeting.
That's the whole idea behind the way Dagaz works: we're built for multi-outlet operators specifically, and the monthly AI Close tracks what's live, parked, killed, costing, and next — across the group, not just at head office. The playbook has to work in every branch, or it doesn't count.
Where to start
Before you green-light another rollout, get an honest read of where AI actually stands across your locations. A Dagaz AI Reality Check — one week, fixed price, money-back — maps what you're already paying for, the biggest data risk hiding in branch habits, and the highest-value things AI could do across the group, in what order.