How to Identify the Best Business Processes to Automate
For enterprise leaders and operators choosing what to automate. Compare process readiness and potential value, clarify decision authority and define a bounded pilot with measurable outcomes.
On this page
The best business processes to automate have a clear outcome, usable inputs, understood decision boundaries and a worthwhile improvement to test. Repetition helps identify candidates; it does not decide which should go first. Consider two invoice workflows: preparing summaries of missing-reference exceptions and resolving disputed prices. Both may recur frequently, but only one may be ready for a bounded pilot.
To choose, separate readiness from potential value. Establish whether each proposed improvement can be tested responsibly, then compare eligible candidates by expected benefit, effort and uncertainty.
Define the improvement before choosing the process#
Start with a shortlist, available process records and the knowledge of the people responsible for the work. Mark gaps rather than filling them with confident estimates.
For each candidate, describe the specific intervention—not just the department or process name. “Automate invoice processing” is too broad. “Prepare missing-reference exception summaries for an authorised reviewer” gives the team something it can assess.
Write down:
- The outcome the intervention should improve.
- What an acceptable output looks like.
- Where the proposed automation starts and stops.
Then distinguish three kinds of work. Exact actions, such as checking required fields, call for explicit rules and validation. Interpretation, such as summarising conflicting records, may warrant tested AI assistance. Consequential commitments, such as approving a payment, need an established authority boundary. One pilot can combine these approaches without automating every step.
Also ask whether the work should exist in its current form. The US General Services Administration’s Eliminate, Optimize, Automate guidance asks federal agencies to remove unnecessary work, improve remaining processes and automate repetitive tasks. That is practice guidance, not comparative evidence of enterprise outcomes. It provides a useful alternative to assuming automation is always the answer. (gsa.gov)
Compare readiness separately from potential value#
Use the same questions for every candidate. Gather representative cases, including exceptions, rather than relying only on the easiest examples. Distinguish observations supported by records, assumptions to test and unknowns that remain unresolved.
Make the selection in two stages:
- Eligibility: Can this scope enter a bounded pilot? Undefined acceptance criteria, unusable inputs or disputed authority are blockers.
- Priority: Among eligible candidates, which expected improvement justifies the implementation effort, ongoing maintenance and uncertainty?
This eligibility-before-priority approach is a proposed decision aid, not a validated predictor or numerical scoring model. A high potential benefit should not compensate for an unresolved permission to act.
The matrix below is both a worked example and a reusable comparison sheet. Replace the candidate headings and entries with your own evidence. All invoice details here and in the pilot example are hypothetical; they are not customer findings or measured results.
| Assessment question | Candidate A: missing-reference summary preparation | Candidate B: disputed-price resolution |
|---|---|---|
| What outcome and acceptance criteria apply? | Prepare a reviewable summary identifying the case, missing information and source records. Acceptance criteria agreed. | Resolve the dispute correctly. Acceptance criteria incomplete because responsibility for accepting a price variance is disputed. |
| What workload, delays and correction are evident? | Assume recurring manual gathering of invoice and order records. Frequency and total effort remain unmeasured. | Assume recurring disputes and waiting between teams. Delay causes and correction effort remain unknown. |
| What is the proposed scope? | Validate required fields, prepare a summary and route it. Interpretive content is checked; payment decisions are excluded. | Interpret contractual and invoice evidence, recommend a resolution and approve the variance. The commitment boundary is unresolved. |
| Are inputs usable? | Approved invoice and order records are accessible. Missing or conflicting evidence can be flagged. | Evidence is available, but controlling terms and escalation rules need confirmation. |
| Who owns the work and permitted actions? | Accounts payable owner identified; designated reviewer handles unresolved evidence. Preparation and routing are permitted. | Operational owner identified, but authority to approve disputed prices is contested. |
| Could elimination or simplification address the problem? | Check whether better reference capture upstream could prevent these exceptions. | Check whether clearer purchasing terms and escalation rules could reduce disputes. |
| What prevents a bounded pilot? | No purpose, input or authority blocker identified for preparation-only scope. Controls still require validation before launch. | Undefined approval authority and incomplete decision rules. |
| What benefit, effort and uncertainty should be compared? | Hypothesis: less preparation effort. Assess setup, checking and maintenance effort; review burden is uncertain. | Potentially valuable, but benefit and implementation effort cannot yet be assessed reliably. |
| What next step is justified? | Develop a preparation-only pilot, subject to confirming controls and baseline. | Clarify authority and rules before reconsidering automation. |
Candidate A can move towards a pilot because its scope is defined and permitted: it prepares information without committing a payment decision. Candidate B needs clarification. Higher workload would not resolve its missing authority.
That does not make Candidate A universally superior. Upstream simplification might remove much of its work; clarified rules might make Candidate B attractive later.
For effort-saving candidates, estimate the work that could actually disappear, then account for added checking, exception handling and maintenance over the same period. For delay or quality goals, compare the affected bottleneck or error type instead. Total process volume alone is not an estimate of benefit.
Use clarify for unresolved purpose, rules or authority; repair for inadequate inputs or process conditions. Eliminate, simplify or defer where those choices are better justified. If eligible candidates remain indistinguishable, identify the missing evidence—such as review effort or maintenance demands—instead of manufacturing a winner.
Turn the selected candidate into a bounded pilot#
Carry the selected matrix entry forward into a pilot specification. For Candidate A, the proposed sequence is:
Gather approved invoice evidence → validate required fields → prepare a source-linked exception summary → route to the authorised reviewer → record resolution.
Payment execution and vendor-detail changes remain outside scope.
A proposed summary could contain:
Case: Invoice A-104
Issue: Purchase-order reference missing from the invoice.
Evidence: Invoice record A-104; candidate order record PO-218.
Uncertainty: The records do not establish whether PO-218 belongs to this invoice.
Next action: Designated reviewer to confirm the match or request further information.
Status: Awaiting reviewer confirmation; payment approval is outside this summary’s scope.
The acceptance test is inspectable: the summary identifies the case, distinguishes missing from conflicting information, preserves source references and reaches the designated owner.
Before starting, document the baseline, accountable owner, permissions, failure tests, review date and stop condition. Test unavailable records, ambiguous matches, duplicate cases and interrupted routing—not just the expected path.
For AI-assisted interpretation, NIST’s 2023 AI RMF 1.0 Core supports three relevant controls: defined human–AI responsibilities in GOVERN 3.2, performance evaluation under deployment-like conditions in MEASURE 2.3, and assigned responsibility for disengaging unsuitable systems in MANAGE 2.4. These support pilot controls; they do not validate this selection method. (airc.nist.gov)
A reviewer must be able to inspect the evidence, reject the summary and resolve or escalate uncertainty. An approval button alone does not establish that capability. Set review requirements around the action’s consequences rather than requiring approval for every routine step.
Check whether the whole process improved#
Compare the pilot with its baseline using measures defined for this workflow:
- Cycle time: from exception detection to accepted resolution.
- Total handling effort: preparation, review, correction and exception-handling minutes per resolved case.
- Exception rate: cases requiring handling outside the defined route, divided by cases assessed.
- Output quality: whether summaries preserve evidence, flag uncertainty and avoid unsupported matches.
These are suggested measures, not established benefits. Track setup and maintenance effort separately. Record the comparison period, case mix, exclusions and concurrent changes. Faster summary preparation is insufficient if checking effort rises or work moves elsewhere.
Expand only if the evidence supports the intended outcome and the team can maintain the controls. Revise if the mechanism needs adjustment; stop if failures breach the agreed boundary.
Compare your current candidates in the matrix, then document one bounded pilot—or the specific clarification or repair needed before proceeding.
