Start with a recurring task that has understandable rules, accessible inputs and a clear owner. A good first automation should be easy to test on a small scale, with mistakes you can detect and recover from.
Find the recurring friction
Ask the people doing the work where they repeatedly copy information, chase an update or recreate the same summary. Collect examples from a normal working week. Note what starts the task, what information it needs and where the finished result belongs.
Possible starting points include routing an inquiry to the right person, preparing an internal weekly summary or creating a reminder when an agreed next step becomes due. Each still needs a defined owner and a way to handle exceptions.
Write the workflow in one sentence: “When an inquiry arrives, create one record, assign an owner and notify that person.” If the sentence keeps expanding, narrow the first version.
Check five things before choosing
- Repetition: Does the task happen often enough to justify setup and maintenance?
- Clarity: Can the team describe the normal steps and recognize a correct result?
- Inputs: Is the required information available, current and permitted for this use?
- Consequence: What happens if the workflow is wrong, late or runs twice?
- Ownership: Who receives exceptions and keeps the process working?
Use these questions to compare a few candidates. Avoid giving a frequent task an automatic advantage if it also involves unclear decisions or expensive mistakes. Microsoft’s automation guidance similarly emphasizes selecting appropriate tasks, assessing value and maintaining automated processes.
Separate rules from judgment
Some steps have a straightforward rule: check whether a required field exists, copy an approved value or route a request according to a known category. Other steps need interpretation, such as summarizing a long message or suggesting a response.
Keep those responsibilities explicit. An AI assistant might prepare a summary while a person confirms the scope or approves an outgoing message. Do not quietly turn “prepare” into “send,” or assume a confident answer is a verified one.
Where AI is involved, define what it may read, what it may change and when it must stop for review. NIST’s AI Risk Management Framework includes clear human-oversight responsibilities and ongoing monitoring. Apply that principle in a way the team can actually operate.
An illustrative inquiry-routing pilot
The first pilot could create a draft CRM record from a valid inquiry, check for an existing record, assign an owner using agreed rules and send an internal notification. A person would review the inquiry and decide how to respond.
Keep unusual cases visible. A message with no usable contact information could enter a review queue. A possible duplicate could be flagged. If the CRM is unavailable, the original inquiry should remain recoverable and someone should receive a failure notice.
That scope gives the team a specific result to test without automating pricing, promises or the customer conversation. Any later expansion should have its own rules and review.
Test the whole path
Follow several permitted test cases from the original trigger to the actual destination. Include a normal request, missing information, a duplicate, an unexpected format and an unavailable connected application.
Check that a retry does not create a second record or repeated notification. Confirm that the workflow reports failure clearly, preserves enough context to investigate and can be paused by its owner.
For a small pilot, reviewing every result may be practical. Agree the pilot’s duration and success criteria before running it. Expand only after the evidence supports the next step.
Your first-automation checklist
- Describe the trigger, inputs, steps and expected destination.
- Name the process owner and the person who handles exceptions.
- Confirm access permissions and which information may enter each tool.
- Document the rules, review points and actions that remain manual.
- Test normal cases, incomplete inputs, duplicates and connection failures.
- Provide a pause control and a workable manual fallback.
- Record the baseline, pilot results and ongoing costs.
- Write brief operating instructions and agree a review date.
Measure the value after review and maintenance
Compare the manual baseline with the pilot, including time spent checking results, fixing exceptions and maintaining the setup. Track completed work, errors, delays and whether the right person received the result.
A workflow that saves handling time but creates confusing exceptions needs more work. Keep the parts that help, simplify the parts that do not and stop a pilot if its risks outweigh its usefulness.
Protocorp’s workflow-automation services start with the business process, its information and its owner. Bring one recurring task to the conversation so the first implementation can have a clear scope and a result worth testing.