Why do AI pilots stall?
It depends on the pilot, but the pattern we see repeats. The pilot has no owner whose week gets better if it works. It runs on sample data, so nobody trusts it with real data. And it has no date, so it never has to be finished. A pilot with all three problems can impress a room and still be gone in a quarter. A job with an owner, real inputs and a deadline has to work, which is the point.
What a real job looks like
A real job has four parts. Use them as a test before you spend money:
- One owner. A named person whose work gets easier, or whose customers get a better service, when it works.
- Real inputs. The actual enquiries, reports or bookings, under rules you agree before the build.
- A date. An event, a cohort, a season: something that will happen whether the AI is ready or not.
- A way to tell. A number you can count afterwards: reports sent, enquiries answered, hours given back.
Three jobs we have built
A luxury hotel brand had one event: 21 senior executives, five days together. The companion app had to work on the first morning, so it went from the scope workshop to launch in six weeks.
An education assessment platform had one cohort: 271 candidates on 11 June 2026. The platform had to produce their feedback reports after that day, so we planned the build back from it.
Rayna Local has one front desk per client: enquiries answered, quotes followed up and bookings run, with a named person to call. It is a set package, from £150 a month + £500 setup and VAT, because a local firm's front desk is the same job from one firm to the next.
None of these started as a strategy for AI across the business. Each one started as a job with a date.
"But we need an AI strategy"
You do. A good one is a short list of jobs, in order, with an owner for each. The first job teaches you what your data is like, what your people trust and where the rules need to be tighter. Those lessons are the strategy. A document written before any of that is a guess.
If your team is not yet at ease with the tools, start there. Rainshift, our brand for people and teams, runs a two-hour workshop on one real problem for £500 + VAT. Your team leaves with that problem worked on, not a slide deck.
How to pick the first job
- List the five tasks that take your team the longest each week.
- Cross out any where a mistake would reach a customer before a person sees it.
- Of what is left, pick the one with an owner who wants it fixed and a date coming up.
That is your first job. It will not be the biggest one. It should be the one you can finish.
Once the job is chosen, draw the line between AI and people before you build. To talk it through with us, tell us what you have in mind.
