Automation projects in small and mid-sized companies often start in the wrong place: an ambitious AI assistant, a new system, a dashboard nobody asked for. The better starting point is humbler. Find the report someone rebuilds by hand every week, and make it arrive by itself, correct, before the meeting that uses it.
Why start with reports
- The value is visible: hours returned to the team, every week.
- The definition already exists: the manual version shows what people actually need.
- Errors surface: automating a report exposes inconsistent data between systems, which needs fixing before anything more ambitious.
The sequence
1. Inventory the routine
List recurring tasks for two weeks: what is done, by whom, how often, how long it takes, which systems it touches. Most teams find more than they expected.
2. Score each task
| Criterion | Question |
|---|---|
| Hours | How much time does it take per month? |
| Stability | Do the rules stay the same week to week? |
| Data access | Can the inputs be fetched automatically? |
| Cost of error | What happens if the output is wrong for a day? |
Start with tasks that score high on hours, stability and data access, and low on cost of error.
3. Build with monitoring from day one
Every automation should report its own health: when it last ran, how many records it processed, and whether the totals reconcile with the source. A silent failure is worse than a manual task.
4. Keep a person on decisions
Automate the gathering, cleaning and calculating. Keep human approval for actions that spend money, change prices or contact customers, at least until the automation has a track record.
5. Add AI where judgment is repetitive
Classifying return reasons, drafting listing updates, triaging customer messages: tasks where the judgment is repeated many times and a person can review the output quickly. Measure accuracy against human decisions before widening its scope.
A test of readiness. If the team cannot agree how a number is defined when it is calculated by hand, automating it will only produce the disagreement faster. Agree the definition first.
What it costs to keep
Automations need maintenance when marketplaces change their reports, APIs or fees. Budget a small, regular amount of time for upkeep, and keep short operating notes for each automation so that it does not depend on one person.
Weighing a decision like this?
Discuss it with usMore research