If you run client work — agency or solo — the hours you lose rarely go to the work itself. They go to the admin wrapped around it: onboarding, status updates, chasing an invoice, writing up a meeting into actions someone will actually do.
The instinct is to template all of it. That's the wrong unit. Some of what you do for a client needs judgment every time; some of it is identical in shape on every single account. Only the second kind is safe to hand to a template, let alone an agent.
The filter
Ask one question before templating anything: does this task ever change shape based on who the client is, or does it just change the words?
Five tasks pass that test for almost any client-services business:
- Onboarding — same checklist, different name and login.
- Status reporting — same sections, different numbers.
- Meeting notes → action items — same transform, different transcript.
- Invoice chasing — same escalation, different amount and due date.
- Scope-change requests — same form, different ask.
Everything else — "should we push back on this request," "is this client at risk of churning" — needs a human judgment call layered on top. Template the shape, not the decision.
A template alone doesn't run itself
This is the part that gets skipped, and it's the actual engineering problem. A blank template is a file. Someone still has to remember it exists, know when to use it, and catch the cases where it shouldn't go out unedited.
What makes a template operational is two things wrapped around it:
- A trigger — what condition fires it. Not "whenever you remember," but a date, an event, a status change.
- A pre-send check — something that has to be true before anything client-facing actually goes out. For admin work this usually means: a human approves before send, full stop.
A template plus those two things is a small, boring automation. A template without them is a Google Doc nobody opens on week three.
Worked example: the invoice ladder
Invoice chasing is the clearest case because "send a reminder" alone makes it worse — chase too hard, too early, and you sound anxious; chase too late and you've floated the client an interest-free loan.
The fix is a fixed escalation schedule instead of a feelings-based one:
- Day 1 past due — a plain "just checking this didn't slip through" nudge.
- Day 7 — a slightly firmer note referencing the original invoice and terms.
- Day 14 — a direct note naming the days overdue, cc'ing whoever owns the relationship.
- Day 30 — a final notice, formal tone, next steps stated plainly.
Every rung is scripted ahead of time. The only live decision is whether to actually send what's queued — which is exactly the pre-send check from above, applied to a real template.
Where this goes next
Build this yourself and you need, at minimum, the five templates above plus the trigger/check logic around each — a real but small project, a weekend of work for someone who's hit this problem before. If you'd rather not build it: we sell a pack of seven skills and six templates built around exactly this list (Client Ops Autopilot), with the invoice ladder above as one of them.
Either way, the technique is the point: template the shape that repeats, not the judgment that doesn't, and never skip the human check before anything client-facing fires.
