Which admin job should a small business automate first?

Find out which job is taking the time

Ask the person who does the admin to keep a short record over a representative working period. Note how often each job happens, the time spent doing it and the time spent chasing missing information or correcting it. Include the awkward examples that interrupt the normal sequence.

Write down the starting event and the finished result. For the enquiry example, the start is a submitted form and the finish is an assigned task containing the enquiry reference, contact details and service requested. Identify who checks it and what happens when a required field is missing.

Microsoft recommends prioritising straightforward, procedural tasks and recognises that a workflow can retain human decisions while automating other steps. Its guidance concerns Power Platform operations; the shortlist here is my suggested application for small-business admin.

• How often does this job happen in an ordinary week?
• Are the input and the required result clearly defined?
• Which exceptions require somebody to make a decision?
• Can the result be checked and a mistake put right?
• Who will own the process and stop it if it goes wrong?

Source: Microsoft: recommendations for implementing automation

Compare a few candidates before choosing

Look for a job that is repeated enough to matter, with accessible information and a manageable set of exceptions. A task that takes seconds once a month may not justify a new system. Something frequent can still be a poor first choice if every case needs a different judgement.

These are proposed starting points to compare with your own work. They are not customer results or a universal ranking. Check whether your existing software already handles the step before paying for another connection.

Decide which steps need rules and which need AI

If a form already has a service dropdown, that value can determine the internal queue through a normal rule. Check the options in the form, inbox or task system you use. A simpler setup can be easier for the person maintaining it to understand.

AI may be useful when a message needs interpreting, such as suggesting whether a free-text enquiry concerns a repair or a new installation. Keep that suggestion available for review and test the cases where the wording is unclear. An AI draft about your services also needs current, approved business information.

The government's AI adoption guide starts from an identified business problem and considers whether generative AI suits it compared with other options. For a first admin project, I'd ask what the smallest useful change would be, including a clearer form or a better handover between people.

Source: The People Factor: government guidance on adopting AI tools

Keep the first pilot small enough to inspect

Agree one input, a destination and the result a named person will check. For the enquiry pilot, use made-up details in a test setup first. Confirm what is saved in the task system and how the owner is notified before routing live enquiries through it.

Try a complete record, a missing detail and the same enquiry arriving twice. Arrange an unavailable destination in the test setup and confirm that somebody is told. Keep the enquiry reference with the result so a retry can be checked for duplicates.

Write down who is allowed to change the workflow and which accounts it needs. Keep customer messages, quotes and consequential decisions behind the agreed approval step while the first version is being assessed. A useful pilot can prepare work for a person to finish.

Keep a record of the expected result, what happened and what needed fixing. Agree how to pause the workflow and return to the manual process. Make sure the person doing the daily work can find those instructions.

Count the work left after automation

Compare the same type and volume of work before and during the pilot. Record time spent reviewing outputs, correcting errors and maintaining the workflow. Microsoft's measurement guidance cautions against claiming value from theoretical time savings alone. Its agent-reporting examples are not a benchmark for your business.

For illustration, suppose 20 items previously took four minutes each: 80 minutes a week. During a pilot, reviewing them takes 20 minutes, corrections take 10 and monitoring or upkeep takes another 10. That leaves 40 minutes of weekly work and releases 40 minutes. These are sample figures showing the calculation, not measured savings from an installation.

Keep the build cost, subscriptions and usage charges beside that time record. Include the effort needed to change the workflow when an app or a business rule changes. Ask the person using it whether the time released is useful in practice and whether the quality of the result has improved.

Count the outcomes that matter too: an enquiry assigned correctly, a report ready when needed or fewer records waiting for missing information. Keep missed and incorrect results visible alongside successful runs.

Source: Microsoft: measure the impact of your agents

Use the pilot to decide what happens next

Before switching over, agree what a pass means and which failures must be fixed. Review ordinary cases and exceptions with the person responsible. Confirm the saved result, the notification and the fallback, rather than relying only on a successful-run message.

If checking and repairs consume most of the time you expected to release, simplify the workflow or reconsider the task. You may need a better input form before another automated step will help. Keep the evidence so the decision is understandable to whoever pays for the work.

At Flying Penguin, we map the repeated admin before choosing the connections or AI steps. Bring one process and a few typical examples to the discussion. A first project should leave you with a useful result, a named owner and enough notes to look after it.