The first one has to succeed
The first automation in a business is not really about the hours saved. It is about proving that this kind of project works here, which decides whether there is a second one.
So we choose the first project for likelihood of visible success rather than for size of prize. The biggest opportunity is usually the wrong place to start, because it is big for the same reasons that make it risky.
The scoring method
Score each candidate from 1 to 5 on four axes and multiply. Anything over 100 is a strong first project.
- Frequency: daily scores 5, monthly scores 1
- Time per instance: over an hour scores 5, under five minutes scores 1
- Cost of error: a mistake reaching a customer scores 5, a cosmetic slip scores 1
- Containment: one team, one system scores 5; three departments and an external partner scores 1
Containment is the axis people ignore and the one that most often decides the outcome. A perfectly-scoped automation that depends on another company's cooperation is not contained, whatever the other three scores say.
What a good first candidate looks like
Boring, frequent, owned by one team, and independent of anything you do not control. The classic example: orders arrive by email in a consistent format, someone types them into the order system, and that person is not the only one who understands the rules.
The best first automation is one nobody will write a case study about, delivered in a few weeks, that quietly removes ninety minutes a day.
Three tempting candidates that usually go badly first
- Anything customer-facing. Higher cost, higher stakes, and failure is visible to people who did not sign up for your improvement programme.
- Month-end reporting. Low frequency means slow feedback and a long wait to find out whether it worked. It also tends to expose data quality problems that turn a four-week project into a four-month one.
- The judgement-heavy step. Quoting, prioritising, approving. These are exactly where AI demos look impressive and production systems disappoint, because the judgement encodes context nobody wrote down.
A worked example
| Candidate | Freq | Time | Error | Contained | Score |
|---|---|---|---|---|---|
| Retype orders from email | 5 | 4 | 4 | 5 | 400 |
| Chase overdue invoices | 3 | 3 | 3 | 5 | 135 |
| Month-end management report | 1 | 5 | 4 | 2 | 40 |
| Customer quote generation | 4 | 4 | 5 | 2 | 160 |
| Onboard a new supplier | 1 | 4 | 3 | 2 | 24 |
The order retyping wins clearly, and it is also the least interesting item on the list. That is the usual pattern: the highest-scoring first project is rarely the one anyone was excited about, and it is almost always the one that gets phase two funded.
What to do with the exciting one
Keep it, and put it second. The quote generation in that table is a genuinely valuable project; it just needs the credibility, the data access and the working relationship that the first project establishes.
Doing it in that order also improves the second project technically. By then the integrations exist, the data is cleaner and everyone has learned how the other side works.