What Should You Automate First in Your Business?

The first automation project should create an obvious operating win — not a science fair. Teams get stuck when they automate the most interesting problem instead of the most valuable repetitive one, or when they jump to AI because it is trendy while a CRM workflow would have removed the pain last quarter.

This article is about priority: which project ships first. For how to assess a workflow and choose the right lever (simplify, integrate, automate, AI), use the AI workflow assessment.

Direct answer

Automate first where work is frequent, costly in time, low in necessary judgement, measurable and reversible — and stable enough to describe in steps. Prefer a clear owner, accessible systems and a baseline you can compare against. If those conditions are missing, simplify or integrate before you automate.

Score candidates, then pick one

Rank each candidate 1–5:

Rough bands: ~20+ strong first project · mid band possible after simplification · low band usually not first.

A lower-scoring compliance step may still outrank a high-scoring convenience task. The scorecard is a decision aid, not a law.

What a good first automation looks like

High frequency, low ambiguity, measurable, reversible, and not mission-critical if it fails once during the pilot. Typical patterns: routing inbound requests, creating CRM records from forms without retyping, status updates on stage change, document shells from structured fields, contact sync between systems, nudges for stalled items.

These tend to be rule-friendly. Save open-ended language work for later — or handle it as assisted AI under review after the operating path is stable.

What makes a bad first automation

Highly variable work, politically sensitive decisions, poorly documented steps, frequent expert judgement, bad data, no clear owner, or “AI everywhere” without a baseline. Fully autonomous customer replies with legal risk, automating a process nobody can draw, and scraping fragile UIs when an API exists belong here.

Worked example

A services firm spends ~12 minutes re-entering every website lead into the CRM, tagging the service type and assigning an owner. Volume is ~25 leads/week. Frequency, time, error impact (misrouting), judgement (rules can assign by service), stability, systems and adoption all score high.

Strong first project: form → CRM workflow. No model required.

Compare that to “summarize every meeting with AI and auto-update the CRM” — interesting, higher variance, weaker readiness. Usually second or third.

Frequency is not the only score that matters

A daily task that takes thirty seconds may lose to a weekly task that burns forty minutes and creates customer-facing errors. Equally, a high-scoring idea with no system access or no owner is not a first project — it is a dependency list. Fix access and ownership, then re-score.

Reversible matters: a mis-routed lead is annoying; an irreversible message sent to the wrong client is a different class of risk. First automations should fail safely.

Ship the smallest vertical slice

Publish the scorecard so prioritization is visible. Finish a pilot that removes real retyping or waiting. Document the happy path before encoding every exception. Set a review date. Re-score the backlog after the first win — friction moves.

If the backlog is dominated by lead-handling noise, inspect website intake as well as the backend. Automation cannot fix an unclear offer page; it can only move confusion faster. When the path is ready, workflow automation should feel boring in the best way: fewer manual steps, clearer handoffs, measurable time back.

Map first if you have not (workflow mapping guide). If the candidate fails because judgement, data or failure cost make a model the wrong tool, read when not to use AI in your business before forcing an “AI first” pilot. Then pick the highest-scoring candidate you can finish in a short pilot. Momentum compounds. Ambition without a first win usually does not.