Role playbook · Operations & Admin · Free
The operations playbook.
Who this is for
Operations is where AI pays for itself fastest and where it breaks loudest. The recurring grind, reports, intake, meeting follow-up, document chasing, is the right target. The undocumented process is not.
Operations managers, office managers, executive admins, finance ops, and anyone who owns a recurring back-office task that nobody else wants.
Stop first
Three things to stop before you add AI to anything.
- Stop automating a process that is not written down. If the steps live in one person’s head, document them first. Thirty minutes with a notes app is enough.
- Stop copying data between tools by hand before checking whether the two tools already connect. Most do.
- Stop rebuilding the same report. Save the one you send most often as a template, with the data sources named.
First 30 days
Document the three worst recurring tasks and time them.
- 01
Step 1
List every task you do weekly or monthly that follows the same steps. Pick the three that eat the most hours.
- 02
Step 2
Write each one down as numbered steps, with where the inputs come from and what "done" looks like.
- 03
Step 3
Time the manual version for two cycles. You will need the baseline, and the number is usually worse than you think.
- 04
Step 4
Try the AI already in your office suite on one of them: drafting the recurring email, summarising the meeting, cleaning the spreadsheet.
Days 31–60
Ship one automation with a review step.
- 01
Step 1
Take the best-documented task and build the smallest automation: a generated draft, a pre-filled form, a report that assembles itself and waits for you to send it.
- 02
Step 2
Keep a human review step until the output has been right for four cycles in a row.
- 03
Step 3
Write a one-paragraph runbook: what it does, where it runs, what to do when it fails, who to call.
- 04
Step 4
Tell the people downstream what changed and what to expect. Silent automations get blamed for everything.
Days 61–90
Prove it, then take the second task.
- 01
Step 1
Compare hours and error rate against the baseline. Share the number with the owner, good or bad.
- 02
Step 2
Remove the review step only if the error rate earned it. Keep it on anything customer-facing or financial.
- 03
Step 3
Start the second task from your list. Reuse the runbook format.
- 04
Step 4
Put recurring reports on a schedule and stop being the person who remembers to run them.
Data rules
What goes into a model, and what never does.
- Vendor contracts, invoices, and employee records go only into tools covered by your company agreement, never into a personal chat account.
- Anything with a payment instruction (bank details, wire changes) is never automated end to end. A human confirms by a second channel.
- Meeting recordings are fine to summarise. Tell attendees they are being recorded, every time.
What to measure
Four numbers. Record them on day one.
Hours per week on the three tasks
Timed at day 0, day 60, day 90.Error rate
Corrections per cycle, counted during review.Cycle time
How long from "needed" to "done" for intake, reports, and follow-ups.Runbooks written
If it is not written down, it is not an automation. It is a dependency on you.
What we would build first
A recurring report that builds itself. The one you copy and paste every week, generated from the source systems, checked by you, sent on schedule. It is boring, it works in a month, and it pays for the next one.
If you want this plan written for your business instead of your role, that is the AI Audit. If your team needs the skills the plan assumes, that is AI for Operations & Admin.