Your first AI automation: choose a workflow, not a promise.
A practical way to identify a useful first project, define the boundaries and measure what really changes.
Begin with the work people repeat
Look at a normal week inside the company. Which requests get copied from an inbox into a spreadsheet? Which documents are renamed and filed by hand? Which customer enquiries sit unattended because ownership is unclear? Write down the actual steps, the person responsible and the exceptions. A first automation project should address a recognisable operating problem, with a clear beginning and a clear end.
Compare frequency, effort and consequence
For each candidate, record the weekly volume and the time a person spends handling a typical case. Then consider the consequence of an error. Classifying an internal enquiry is different from approving a payment. A frequent task with accessible data and reversible actions is often a useful pilot. Avoid choosing a process simply because a demonstration looks impressive. A low-volume task with many unusual cases may offer little practical value.
Define a small, reviewable pilot
Take one input channel, one team and one output. For example: receive an enquiry, extract the requested service, create a draft CRM entry and ask the team to confirm the assignment. Specify which fields are mandatory and what happens when they are missing. Keep the original message available so the reviewer can compare it with the proposed record. Start with a representative sample that includes awkward cases, not only ideal examples.
Measure the total effort
Compare the previous manual effort with the combined effort after automation: reviewing, correcting, maintaining and handling exceptions. Track completed cases and errors alongside processing time. A workflow that runs quickly but creates more corrections has not necessarily improved operations. Agree on the success criteria before the pilot starts, so the decision to extend it is based on observed results.
Prepare the handover before expanding
Name the process owner and the person who can pause the workflow. Document the connections, access permissions and escalation route. Train the team using realistic examples and record the most common failure patterns. Once the first use case works consistently, extend one dimension at a time: another channel, another team or another document type. The first deliverable is a dependable process that people understand.