Expose Manual Steps in an Automated Workflow Demo

The presenter clicks one button. A polished account brief appears. Between those moments, someone may have cleaned the input, selected the correct account, corrected the draft or pasted the result into another system. None of that necessarily makes the workflow poor. Concealing it makes the demonstration difficult to evaluate.

A manual-step ledger describes the work around the automation boundary. It is not a competition to remove every human action. Some human steps provide the judgment and accountability the buyer specifically wants.

For AI-assisted demonstrations, NIST AI RMF GOVERN 3.2 addresses differentiated human-AI responsibilities and oversight. The ledger below is an original way to expose those responsibilities during a buying evaluation, not a NIST-prescribed template.

Ask for a replay, not a faster performance

Choose one synthetic record and ask the presenter to walk through its entire path, including preparation before the visible demonstration. Pause when responsibility changes. Identify whether the software, presenter, buyer or another service performs the next step, and what starts that step.

Use “automatic” narrowly: the described action proceeds under the stated conditions without a person performing that action. A system can generate a draft automatically while still needing someone to supply a source, resolve an exception and approve the result. Describe those actions separately.

A completed ledger for one account brief

This fictional example evaluates the invented HarborDraft workflow using synthetic account A26. The effort figures are illustrative inputs for explaining the ledger, not measured product performance. Waiting time and active work are separate.

1. Prepare the source packet: Buyer researcher selects two approved source excerpts and removes an obsolete version. Manual. Illustrative active effort: four minutes. This preparation occurred before the original demo and must be disclosed.

2. Identify the account: Presenter chooses A26 from two similarly named records. Manual. Active effort: one minute. The product does not resolve this ambiguity in the proposed configuration.

3. Generate the first draft: Software processes the selected packet after the presenter clicks Generate. Automatic generation, manually initiated. Active effort for the click is included in the previous step; illustrative unattended wait is two minutes. No quality conclusion follows from the short wait.

4. Resolve a conflicting date: Presenter inspects both excerpts and removes an unsupported opening date from the draft. Manual exception handling. Active effort: three minutes. The uncorrected draft is retained for the evaluator.

5. Approve the account brief: Buyer reviewer checks source links and marks the brief approved. Required human control, not a product defect. Active effort: two minutes. Approval does not authorize sending the brief externally.

6. Place the brief in the CRM: Presenter copies the approved text into a test account note. Manual transfer. Active effort: one minute. The demonstration has not established an automatic CRM write.

The disclosed example contains eleven minutes of active human work and two minutes of unattended waiting. Those figures describe only this invented walkthrough. They exclude setup, training, failed attempts and maintenance, so they cannot support a savings claim or a per-account operating forecast.

Put hidden preparation back into view

The ledger changes the buyer's questions. Source preparation needs an owner. Account selection needs an ambiguity rule. Date correction requires someone capable of reading the evidence. Manual CRM transfer may be acceptable at a small review volume, but it must not appear as an integrated handoff in the final proposal.

Ask the presenter to confirm each row and disclose anything delegated to another team. If a supplier's operations staff repair a failed record behind the scenes, label that service-assisted handling. Do not count it as self-operating software merely because the buyer does not see the worker.

Capture the initial output as well as the corrected version. Otherwise, the evaluator may attribute the presenter's judgment to the system. The correction should remain visible even when it takes only a moment.

A counterexample: removing the wrong manual step

Suppose the supplier proposes automatic approval to eliminate the two-minute review. The ledger becomes shorter, but the buyer's source-checking requirement disappears. Reject that change unless the buyer deliberately revises the requirement after assessing the consequences. Fewer touches do not by themselves mean a better workflow.

Alternatively, the buyer may accept manual transfer because only approved weekly summaries need to enter the CRM. Record that choice explicitly, including the responsible role and how missed transfers are detected. The ledger should describe the agreed operating model, not a theoretical maximum of automation.

The net-time measurement guide addresses comparative effort once a repeatable process exists. Start here when the immediate problem is more basic: understanding who actually does the work shown in a demonstration.