What to automate first in a small firm

List every recurring admin task you can think of. Then score each one from 1 to 3 on four things.

Add the four scores. The highest total is your first candidate. If two tasks tie, pick the one with the higher "how easy to check" score.

Here's a worked example of the scoring method:

The checking score matters most for a first pilot. If you can't tell quickly whether the automation did the right thing, you can't trust it. You'll end up redoing the work by hand.

• A task two people do in two different ways. Agree the steps first.
• Anything that decides something for a client, like advice, pricing or what to file.
• A task built on messy records. If client names or emails are duplicated or out of date, fix those first or the automation will copy the mess.
• A task that only runs a few times a year. Setting it up and keeping it working can cost more time than it saves.

Fill this in for your one chosen task before anything is built. It stops a pilot from drifting into "it seems fine."

Task: {e.g. remind clients about documents they still owe}
Starts when: {the event that kicks it off, e.g. an item is marked "requested" and is 3 working days overdue}
Ends when: {the result, e.g. a reminder naming the missing item goes to the client and is logged}
Stops when: {e.g. the item is marked received or the client replies}

In scope: {e.g. new client onboarding documents only}
Out of scope for now: {e.g. anything for clients in a dispute or flagged sensitive}

Who approves before sending: {name} (or "no approval needed after week 1")
Who handles exceptions: {name}

Test set: {e.g. the last 20 real cases, including the awkward ones}
Pass if:
- every case in scope gets the right message or is sent to {name} to handle
- nothing goes to the wrong client
- nothing keeps sending after the item arrives
- review takes {name}'s time down from {current time} to {target}
Fail if: {e.g. any message goes to the wrong person}

Pilot length: {e.g. 2 weeks of live use}
Decision date: {date}. Keep, change or stop.

The test set size and the two-week length are suggestions. Use what fits your volume. A task that runs twice a week needs longer to show anything.

Measure the "before" time for a week before the pilot starts. Without it you're guessing at what changed.

An automation keeps costing something after it's live. Plan for these:

Someone has to own each of these. Write the owner's name on the pilot sheet.

1. Spend 20 minutes listing every recurring admin task. Ask the person who does the chasing, because they'll know the ones you've forgotten.
2. Score each one with the table.
3. Pick the top task and fill in the pilot sheet.
4. Measure how long it takes by hand for one week.
5. Build the simplest version that passes, on top of the tools you already use. Keep a person approving outbound messages at first.
6. Run the pilot. Log every exception.
7. On the decision date, check it against the pass criteria. Keep it, change it or stop it. Then pick the next task.

Runnrr handles setup for one recurring workflow in the tools your firm already uses. A bot-managed call, WhatsApp, email or Slack conversation starts the work; the agreed actions happen in your connected business tools. If you'd like to talk through which task to start with, see how Runnrr sets up a workflow.