- Key takeaways
- Almost every company’s recoverable hours hide in four patterns: double entry, the chase, the read, and the answer desk.
- Rank candidates by volume × rule-clarity ÷ exception-rate. The winner is usually boring — and that is exactly why it pays.
- Start with one workflow, measured in hours returned. Impressive comes later; measurable comes first.
The wrong question everyone starts with
“What’s the most impressive thing we could do with AI?” is how most automation conversations begin — and it is precisely backwards. Impressive optimizes for the demo. Operations run on the opposite of demos: high-volume, repetitive, rule-bound work that nobody puts on a slide.
After a hundred-plus projects, we can tell you where the money actually is. It is not in the exotic. It is in four patterns so common that once you see them, you will find them in every department you walk through.
The four patterns
1. Double entry. Information arrives at your company already written — an emailed purchase order, a PDF invoice, a portal submission — and a person types it again into your system. The customer did the data entry; you are paying salaries to do it twice. A logistics client of ours had six people doing this all day; 97% of it is now untouched by hands.
2. The chase. Any workflow whose real content is reminders: chasing client documents, following up quotes, collecting overdue invoices, nudging approvals. Nobody is bad at their job here — the work is remembering, at scale, forever, and humans are the wrong species for it. Accounting firms lose 12–20 hours a month to this one pattern alone.
3. The read. Documents come in; decisions come out. Contract review, ticket triage, application screening, order validation. The judgment at the end needs a human. The 240 pages before the judgment do not. Our construction clients recovered 90% of review time without removing a single human decision.
4. The answer desk. The same questions, on repeat, around the clock: booking requests, availability checks, order status, “is breakfast included.” The pattern is a conversation with rules — which is to say, automatable — and the volume arrives mostly when your staff does not: after hours and at peak.
Scoring your candidates
List every workflow that matches a pattern, then score each one: weekly volume (how many times it runs), rule clarity (could you write the instructions for a smart temp?), and exception rate (how often does it need real judgment?). Multiply the first two, divide by the third.
The best first automation is almost never the most interesting one. It is the highest-volume, clearest-rule, lowest-exception workflow you have — the one everyone stopped noticing because it never changes.
That boring winner has three properties that matter: the baseline is easy to measure, the payback is fast, and the team trusts the result — which buys you permission for automation number two.
What to do this week
Pick your top-scoring workflow. Measure its current cost honestly — hours per week, loaded rate, error cost. Then define done as a number: “this workflow, at a tenth of the touch-time, by this date.” That one page is most of what our Execution Blueprint produces — and it is the difference between an AI project and an AI result.