Blog · AI for small and midsize businesses
How to find the repetitive tasks costing your business time
The task everyone complains about is not always the task consuming the most capacity. A short, consistent inventory makes repeated work visible and helps distinguish process problems, useful automation opportunities and tasks that still need experienced judgment. Start with one ordinary working week.

In this article
The same three-minute task can interrupt a team dozens of times a day without appearing on any project plan. Meanwhile, a monthly report may feel exhausting but consume fewer hours overall. To identify repetitive business tasks worth improving, collect a week of comparable observations before choosing a tool.
The aim is an inventory of work, not a ranking of employees. Explain that distinction before recording anything. People need to report waiting, corrections and exceptions honestly; otherwise, the inventory will show an idealized process that nobody actually follows.
Define the unit you are counting
Choose one team and a normal operating week. Ask each participant to describe tasks with a verb and an object: classify an inquiry, copy an order into the system, locate a policy paragraph or prepare a delivery update. Avoid entries such as administration or email. Those categories combine work with very different causes and improvement options.
For a task that crosses several people, keep one case identifier and separate the individual steps. This prevents counting the entire elapsed time as labor for every participant. A request can wait six hours while requiring only twelve minutes of actual handling.
A worksheet that fits ordinary work
| Field | What to record |
|---|---|
| Task and trigger | The action and event that starts it, such as an order-status email. |
| Frequency | Number of cases actually observed, with a separate note for seasonal peaks. |
| Handling time | Minutes actively spent locating information, preparing the result and checking it. |
| Waiting time | Time awaiting information, permission or another team; keep it separate. |
| Rework | Additional minutes correcting the result and the reason for correction. |
| Exceptions | Cases needing a different route, specialist judgment or missing information. |
| Inputs and systems | Where the facts come from and which system receives the result. |
| Owner and acceptable result | The person who can approve a change and the conditions a completed task must meet. |
Use a lightweight shared sheet or existing ticket log. Record counts daily while details are fresh. If precise timing would disrupt work, use agreed time bands and label them as estimates. Sample several cases directly to check whether those bands are credible.
Read volume, rework and interruptions together
For a hypothetical service team, assume fifty order-status requests take four active minutes each, ten quote corrections take twelve minutes each and four weekly summaries take twenty minutes each. The observed handling totals are 200, 120 and 80 minutes. These are illustrative measurements, not typical business benchmarks.
Now add what the team discovers. Most status requests need a current system lookup. Quote corrections mainly arise from missing quantities on the intake form. Weekly summaries require combining narrative notes and would still be reviewed by a supervisor. Three different responses emerge: improve self-service access, repair the form, and test AI-assisted summarization.
Do not convert all observed minutes into promised savings. Even successful assistance leaves reading, verification and exception handling. Also ask whether interruptions are concentrated on one person. A modest task can create a serious bottleneck when it repeatedly pulls the only specialist away from other work.
Turn the inventory into a shortlist
At the end of the week, the process owner reviews the entries with the people who performed them. Merge duplicate names, challenge unusually high estimates and preserve the difficult cases. Mark tasks as process repair, conventional automation, possible AI assistance or judgment requiring further investigation.
OpenAI's discovery guide uses repetitive work and skill bottlenecks as places to look for opportunities. Your inventory adds the evidence needed to decide which local task deserves attention. It should produce three to five credible candidates.
For each shortlisted task, write one testable question. For example: can a reviewed summary preserve all decisions while reducing total preparation time? Assign an owner to obtain permitted sample inputs and define an acceptable result. Keep confidential material out of informal experiments unless the approved environment and access rules cover its use.
Check that the week was representative
Compare the observation period with month-end, holiday and peak-season work. Record known conditions missing from the sample as gaps in the evidence; avoid confident extrapolation. Repeat a smaller sample when the workload changes. The NIST generative AI profile provides a useful basis for maintaining evaluation as context changes.
Bring the completed inventory, a few representative examples and the unresolved exceptions to an AI opportunity assessment. That evidence makes the next decision more concrete than starting with a software catalog.