Gather appropriate evidence of value

# Telling Value and ROI Story

# Activators

# Champions

Demonstrate whether a workflow is being used, operating as intended, and making work better.

July 17, 2026

Building an AI workflow is not the end of the work. Once people start using it, capturing evidence helps illuminate whether it is working, improve it, and decide what should happen next.

Evidence can be quantitative or qualitative. It might come from usage data, timestamps, workflow records, quality reviews, structured feedback, or direct observation. What matters is being able to explain where the signal came from and make a claim that the evidence actually supports.

Treat the categories below as a menu, not a checklist. Strong evidence in every category is not required. Start with the signals most relevant to the problem the workflow was intended to solve and its current stage of use.

Adoption

Adoption evidence shows whether the workflow has moved beyond initial interest and become part of real work. Look for signals such as:

Use during real work

Repeat use by the same people

Additional people beginning to use the workflow

Reuse of a shared prompt, template, GPT, skill, or other asset

Use across multiple relevant tasks or situations

An established operating cadence, such as a weekly review or recurring handoff

A team process changing to incorporate the workflow

Attendance, downloads, positive reactions, and stated intent to try something are useful leading indicators. They show exposure or interest, not adoption. A person attending a demonstration is different from that person using the workflow, and a first use is different from repeat use.

When possible, capture who used the workflow, how often, for what work, and over what period.

Efficiency

Efficiency evidence shows whether the workflow changes the effort or resources required to complete the work. Look for changes in:

Time required

Number of steps

Manual handoffs

Waiting time

Rework

Cycle time

Volume or throughput achieved with comparable resources

Capacity required from a particular role or team

Preserve the baseline, unit, period, and measurement method. For example, “saved time” is less useful than:

Average preparation time declined from approximately 90 minutes to 40 minutes across six weekly reports, based on timestamps recorded by the workflow owner.

If a result is estimated, label it as an estimate and state the assumptions. Avoid annualizing a small initial result unless the projection is genuinely useful and clearly identified as a projection.

Quality

Quality evidence shows whether the workflow’s output is accurate, complete, consistent, or useful enough for its intended purpose. Look for signals such as:

Accuracy or error rates

Completeness

Consistency across outputs

Correction or rework rates

Performance against defined test cases

Rubric or review scores

Compliance with required formats or standards

Structured feedback from the people who use or receive the output

Fewer missed requirements or exceptions

Faster work is not automatically better work. If a workflow reduces production time but creates more review or correction work, both signals matter.

Define what “good” means for the workflow before making a quality claim. Where possible, use representative test cases, a consistent review method, or a comparison with the previous process.

Safe operation

Safe-operation evidence shows that the workflow has defined boundaries, controls, and ownership appropriate to what it can access or do. Look for evidence of:

Clearly defined and permitted inputs

Human approval or review points

Representative test cases, including exceptions

Redaction or data-handling practices

Source checking or citation requirements

Defined actions the AI may and may not take

Escalation paths for uncertain or higher-risk cases

A named workflow owner

Documentation for maintenance and updates

Monitoring or periodic review

Be precise about what this evidence means. A review step is evidence of a control, not proof that errors never occur. Passing a set of test cases supports confidence within those tested conditions—not a claim that the workflow will perform correctly in every situation.

Team outcome

Team-outcome evidence shows whether using the workflow contributes to a result that matters to the team or function. Look for outcomes such as:

Capacity shifting to higher-value work

Faster or better-informed decisions

Improved service or response times

More consistent performance across a team

Reduced operational backlog

Better coordination or fewer dropped handoffs

Greater ability to handle demand

Improved stakeholder or customer experience

Progress against another team-relevant measure

Keep the connection between workflow use and the outcome appropriately cautious. A team result may have several causes. Describe whether the relationship was measured, directly observed, reported by stakeholders, or inferred from the available evidence.

For example:

Team members reported that the workflow made weekly reviews more consistent and reduced preparation effort.

This is more credible than claiming that the workflow “transformed team performance” without evidence to support that conclusion.

Describe the status of the evidence

Use simple labels to help readers understand the strength and source of each claim:

Measured: Captured using a defined quantitative or qualitative method

Observed: Directly witnessed or documented, but not formally measured

Reported: Shared by a user, manager, stakeholder, or beneficiary

Estimated: Calculated using stated assumptions

Planned: A signal you intend to capture but do not have yet

Unknown: Important information that is not currently available

Reported or observed evidence can still be useful, especially early in a workflow’s life. Labeling it accurately makes the story more credible.

Make every signal easier to defend

For each important piece of evidence, try to record:

What changed

The baseline or comparison, if one exists

Who or what the evidence covers

The relevant period

The source or collection method

Whether it was measured, observed, reported, or estimated

Any limitation or alternative explanation

A perfect measurement system is not necessary to get started. A simple timestamp, usage count, quality review, or recurring feedback question can create a useful starting point.

Capture what can be supported today. A small credible signal is more useful than an impressive claim that cannot be defended.