Workflow walkthrough

What a business process automation audit should deliver

Evaluate an automation audit by its deliverables: a current process map, measured baseline, ranked options, and a pilot brief with clear acceptance criteria.

A useful business process automation audit should leave you able to make a decision: what to improve, whether automation is justified, and what a small first build would need to prove. Ask for a map of the current work, evidence of the problem, a comparison of options, and a pilot brief your team can review.

For an owner or operations leader, those deliverables make a discovery engagement easier to evaluate before committing to implementation. A list of AI tools does not explain who will check the output or what happens when source information is missing.

Here, “audit” means an operational assessment of a workflow. Security testing, compliance certification, and financial assurance require their own scope and expertise. Agree on the intended deliverables with your provider before work starts.

The walkthrough below uses a hypothetical service business preparing a weekly operations report. It is an illustration of how to assess discovery work, not a Codi client result or a promise about a particular engagement.

Define the decision before collecting information

Suppose the operations manager assembles a weekly report from a scheduling system, a sales spreadsheet, and staff updates. The owner wants the report earlier, with fewer unexplained differences between the figures.

The audit question is specific: can the team reduce the effort needed to produce a dependable report while keeping a person responsible for release?

Set boundaries around that question. In this example, the assessment covers gathering the inputs, reconciling them, preparing commentary, reviewing the result, and releasing it internally. Replacing the scheduling system and redesigning the sales process are outside the first assessment.

Before signing off the scope, name the process owner, the people who do the work, the intended reader, and the decision-maker for a pilot. Specify what examples and records the assessor can inspect, with appropriate permission. Use sanitized examples where possible and agree how supplied information will be handled and deleted.

Map what actually happens, including the workarounds

Ask the people who prepare the report to walk through a recent example. Follow the source records through to the version the owner actually received. Capture the unofficial spreadsheet and the clarification message as well as the documented steps.

The map should show where the job starts and ends, which systems supply information, who makes each decision, and where work returns for correction. ASQ’s SIPOC method offers a simple starting structure for defining process boundaries and identifying inputs, outputs, and the people who use them.

For this hypothetical report, a useful finding might be that two systems use different definitions of a completed job. Automating the copying would preserve the disagreement. The assessment should identify who owns that definition before proposing a faster report.

Have the staff who do the job review the map. Record unresolved questions explicitly. A map based only on a manager’s description can miss the steps that make the result usable.

Establish a baseline you can compare with a pilot

The baseline should distinguish hands-on work from elapsed time. A report can take an hour to prepare and still wait a day for clarification. Those are different problems with different remedies.

Ask for the measurement period, the number and types of cases observed, and the source of each figure. Separate recorded observations from staff estimates. If the sample covers only straightforward weeks, record that limitation and collect examples of difficult weeks before estimating the benefit.

For the reporting example, collect preparation time, review and correction time, time spent resolving missing inputs, delivery timing, and the errors reviewers find. Also note changes in workload or report scope that could distort a comparison.

The assessment can provide a range where the evidence is incomplete. It should explain what needs measuring next. Use our guide to measuring AI automation ROI for a worked calculation that includes review, exceptions, ongoing costs, and the difference between freed capacity and cash savings.

Compare a process change, conventional automation, and AI

Ask the assessor to explain which part of the problem each option would solve. In the reporting example, the options might look like this:

Option What it could improve What must be resolved first
Agree on definitions and a shared template Inconsistent figures and repeated clarification Who owns each metric and when inputs are due
Use scheduled exports and fixed calculations Repeated copying and arithmetic Reliable access, matching records, and agreed calculations
Add AI-assisted commentary drafts Turning verified figures and staff notes into a readable draft Source traceability, missing-data handling, and editorial review

These options can be combined. Test the simpler changes first where they address the main cause of the work. AI-assisted commentary will not resolve a disagreement about what the numbers mean.

For any AI proposal, ask what information it needs, what it is allowed to do, and how a person can detect and correct a bad result. NIST’s AI Risk Management Framework connects business context and potential impacts with an initial decision about whether to proceed. It is a voluntary risk framework, not a certification supplied by an automation audit.

If you are comparing several workflows, the First AI Project Kit includes a playbook and editable worksheet for recording the inputs, reviewer, risks, and next decision for each candidate.

Ask for deliverables you can check

Use the table below as a proposed acceptance checklist when agreeing on discovery scope. The right depth depends on the process and the decision you are funding.

Deliverable What you should be able to check
Current process map Staff confirm the steps, systems, handoffs, decisions, and recurring exceptions
Baseline and evidence notes Figures identify their sample, source, measurement period, and limitations
Problem and dependency list Each finding names the evidence, consequence, and person who can resolve it
Options and recommendation The recommendation compares process changes, conventional automation, and AI where relevant
Pilot brief Scope, inputs, output, reviewer, failure paths, and acceptance criteria are explicit
Cost and ownership assumptions Setup, ongoing operation, staff review, support, and unresolved access needs are visible
Decision record The owner can choose to build, change the process first, or wait for more evidence

A diagram alone does not establish that an integration is feasible. If access has not been tested, the report should say so. If estimated effort depends on a vendor feature or data export, identify that dependency and who will verify it.

Turn the recommendation into a small pilot brief

In the reporting example, the proposed pilot could cover one internal weekly report. Staff supply approved exports. Fixed calculations produce the figures. AI drafts commentary from those figures and attributable staff notes. The operations manager checks the complete report and releases it deliberately.

The brief should also describe exceptions. A missing export holds the report for correction. Conflicting totals route to the metric owner. A commentary sentence without supporting information is removed or returned for review. The existing manual process remains available if the pilot cannot produce a usable report.

Agree on acceptance criteria before testing. For example, the owner might require every reported figure to reconcile to its approved source, no report to be released without review, and total preparation plus correction time to improve against the observed baseline. These are proposed criteria for this example, not measured results. Define an agreed test set and observation period rather than accepting a single successful demonstration.

Record the person responsible for monitoring the workflow after launch, resolving failures, and approving changes. Ask how the team will access its documentation and continue the work if it changes implementation partner.

Finish with an explicit decision

Close the audit with a short decision record: the problem, evidence, selected option, open dependencies, owner, and next review point.

For our hypothetical business, “standardize the completed-job definition before building commentary automation” could be the most useful outcome. Another assessment might justify a narrowly scoped pilot immediately. Both decisions should follow from the evidence, with the uncertainty visible.

Before accepting the work, ask: could our process owner explain the recommendation and the first test without the assessor in the room? If the answer is unclear, request the missing evidence or a more specific brief.

Bring a recent example, the people who handle it, and the result you need. Start by agreeing on the decision the assessment should help you make.

Find the first AI project worth testing.

Use the free playbook and editable worksheet to compare three tasks your team does every week. Estimate the time you could save, see where someone still needs to review the work, and decide which task to pilot first.

Preview the worksheet

You can download both files after signing up. You’ll get five emails from Lucas at Codi Tech over about two weeks, including invitations to discuss your project. Unsubscribe anytime. Privacy details.