Which workflow should you automate first?

Buying an AI tool is easy. Choosing a useful first project takes more thought. If the process is unclear, the data is scattered, or nobody owns the result, automation can move confusion around faster.

A better starting question is: which piece of work could become easier without introducing an unacceptable cost of being wrong? Use the following scorecard as a discussion tool. It is a practical planning aid, not a validated risk assessment or a promise of results.

1. List the friction before the features

Ask the people doing the work to name the tasks they repeat each week. For each one, record the trigger, inputs, steps, output, reviewer, and destination. Include work that happens between systems, such as copying information, preparing an internal summary, or routing a request.

For example, “improve sales with AI” is too broad to test. “Prepare a draft account brief from approved CRM notes for a salesperson to review” has a clearer boundary. It describes what goes in, what comes out, and who remains responsible.

2. Use five questions to compare candidates

1. Does it repeat? A task needs enough real examples to test. A rare process with many exceptions may be a difficult first pilot.
2. Are the inputs ready? Identify where the information comes from, whether it is accurate enough, and whether your team is allowed to use it in the proposed tool.
3. Can someone judge the output? The reviewer should be able to compare the result with a trusted reference or a clear quality standard.
4. Can mistakes be caught before they matter? Prefer a draft, recommendation, or internal routing step over an irreversible action. Define the human approval point.
5. Does an owner want to use it? Choose someone who can explain the current process, test the new version, and decide whether it helps.

For each question, mark the candidate green, amber, or red. Green means the requirement is clear and ready. Amber means there is a gap to resolve. Red means the pilot should wait or be redesigned. Do not average away a serious privacy, safety, or quality concern.

3. Consider these bounded starting points

Internal brief preparation

A system can prepare a draft from approved business records. A person checks the facts and source references before using it. The pilot should test whether the draft saves preparation time without creating more correction work.

Inquiry triage

A system can suggest a category or responsible team for a new request. Start with human confirmation and an easy way to correct the suggestion. Check unusual requests and cases that fit more than one category.

Recurring report commentary

Use already-validated figures to prepare a first draft of an internal narrative. Keep calculations in dependable systems and require a person to verify the interpretation. A fluent explanation is not evidence that the numbers or conclusions are correct.

These examples are illustrative. They are not DotBranded client results. The right choice depends on your actual process, data, and obligations.

4. Define the pilot on one page

Before implementation, write down these eight items:

• The exact workflow and the person responsible for it
• The approved inputs and information that must stay out
• The output and the quality standard
• The tools, permissions, and access limits
• The human review and approval points
• The ordinary cases and edge cases you will test
• The baseline and the measurements you will compare
• The conditions for pausing, changing, or stopping the pilot

NIST’s voluntary AI Risk Management Framework emphasizes defined responsibilities, understanding the context of use, measurement, and ongoing management of risk. Its core functions are govern, map, measure, and manage. They are not a fixed sequence or a certification checklist. See the NIST AI RMF Core for the underlying framework.

5. Measure the whole job

A faster first draft can still create a slower process. Include the time needed to prepare inputs, check outputs, correct mistakes, and handle exceptions. Ask whether the result meets the same standard as the current method.

Useful measures include total completion time, correction rate, reviewer effort, cost per completed task, and whether the intended team actually uses the workflow. Pick measures before the test so the decision is not based only on an impressive demonstration.

Compare similar kinds of work and record important differences. If the new system handles easy cases while people still do all the hard ones, the headline average may hide the real workload.

6. Decide what happens next

At the review point, choose one of three paths: continue within the existing scope, revise and test again, or stop. Expand only when the evidence supports it and the next use has been assessed on its own terms.

A pilot that reveals weak data or an unclear process can still be useful. It tells you what needs to be fixed before making a larger commitment.

Common questions

Do all automation projects need AI?

No. If a task follows stable rules, a conventional automation may be easier to test and maintain. Use AI where its capabilities fit the work, rather than adding it to every step.

Should the first pilot use customer data?

Only when the use is authorized and the proposed tools and controls are appropriate. Start with synthetic or suitably de-identified examples where practical, then assess the real-data requirements before expanding.

How long should a pilot run?

Long enough to observe representative work and important exceptions. Define a review point around the business cycle and the evidence needed, rather than assuming a universal number of days.

When is outside help useful?

Outside support can help when the team needs to map the workflow, evaluate tools, define controls, connect systems, or organize a credible test. The goal is a clearer decision and a system the team can own.

Find your first useful change

DotBranded helps businesses explore AI strategy, workflow automation, and team enablement from Arizona and across the United States. Bring a repeated task or a bottleneck, and start with a focused conversation about the outcome you want.

Discuss your first AI pilot · Explore AI and automation resources