A workflow is a stronger candidate for automation when it is frequent, understood, measurable, and stable enough to document. Look for clear inputs, repeatable rules, visible bottlenecks, and staff who can help test a redesigned process.
Five Signs a Workflow May Be Ready
1. The work happens repeatedly
Recurring work can make the effort of redesign worthwhile. First understand its normal volume and variation rather than assuming that frequency alone makes it a good candidate.
2. Staff can explain the current process
If people describe materially different ways of doing the same work, clarify and document the process before automating it.
3. Inputs and outputs are identifiable
Identify the forms, emails, documents, approvals, records, and notifications involved, along with who owns them and where they need to go.
4. Success can be measured
Choose measures that fit the workflow, such as cycle time, backlog, errors, rework, staff effort, response time, or customer experience. Record a baseline before making changes so that later comparisons have context.
5. Exceptions are understood
Account for missing information, urgent cases, accessibility needs, and situations that require judgment. Decide who will handle exceptions and when work should return to a person.
A Practical Readiness Review
Ask workflow owners to assess volume, stability, data availability, risk, expected value, integration complexity, and staff readiness. Treat the review as a way to surface questions, not as an automatic approval score. A potentially valuable workflow with sensitive data or unresolved ownership may need more preparation; a smaller, well-understood task may be easier to test.
What to Do Before Building
Map the process as it actually operates. Confirm the problem with the people doing the work, clarify ownership, remove unnecessary steps where appropriate, and define human checkpoints and a baseline. Then test a small redesign with affected users and data controls appropriate to the information and tools involved.
If the workflow uses AI, include risk review and ongoing oversight in the plan. NIST's AI Risk Management Framework is a voluntary resource for incorporating trustworthiness considerations into the design, development, use, and evaluation of AI systems.
Frequently Asked Questions
What should not be automated first?
Be cautious with poorly understood processes, consequential decisions, unstable policies, and workflows without accountable owners or usable baseline information. These may need additional analysis and safeguards before automation is considered.
Is AI required for automation?
No. Rules-based workflow automation may be sufficient for predictable steps. AI may be considered for work involving language, documents, classification, or drafting, with suitable review and safeguards.
How long should a pilot run?
Set the pilot duration according to the workflow's normal volume and variation. Define checkpoints for measurement, feedback, revision, and a decision to scale, change, or stop.