What Metric Can This Workflow Move?
AI is valuable when a specific workflow moves a metric the business already cares about. Sort the idea list by the metric each workflow can move, not by what AI could touch.
Most AI conversations start too wide.
Sales has ideas. Support has ideas. Finance has ideas. Operations has a backlog. IT has governance concerns. Someone already built a prototype. Someone else saw a vendor demo. The list gets long fast.
That usually feels productive.
But a longer list of AI ideas does not make the business case clearer.
At some point, the question has to change. Not “Where can we use AI?” but “What metric should move if this workflow gets better?”
That question is uncomfortable in a useful way. It forces the conversation out of general possibility and into operating reality.
Because AI is not valuable in the abstract. AI is valuable when a specific workflow changes a metric the business already cares about.
The AI tool is not the value case. The workflow is the value case.
”AI can help here” is too low a bar
AI can probably help in a lot of places. It can summarize calls, draft emails, clean up notes, search documents, rewrite copy, generate reports, build first-pass analysis, answer policy questions, and suggest next steps.
That does not mean every use case is worth funding.
A workflow becomes worth funding when you can connect it to a metric:
- revenue
- cost
- speed
- quality
- risk
- recovered capacity
That is the first filter. If the workflow gets better, what changes?
Does revenue move because the sales team responds faster, qualifies better, sends stronger proposals, or recovers stalled deals?
Does cost move because manual work drops, outside spend comes down, tool spend consolidates, or the team avoids another hire?
Does speed move because approvals happen faster, onboarding shortens, reports close sooner, or customer answers go out the same day?
Does quality move because rework drops, first-pass approval improves, or fewer things have to be corrected downstream?
Does risk move because the system catches exceptions earlier, flags policy issues, or reduces the chance that something important gets missed?
Does capacity come back to the team, and does that capacity go somewhere useful?
If the answer is unclear, the idea may still be interesting. It may even be useful. But it is not ready to be an investment case.
The first workflow should have economic shape
The best first AI workflow is rarely the flashiest one. It is usually the one where the pain is already visible and the metric is close enough to measure.
A proposal workflow is a good example.
If every proposal takes two weeks, pulls in three to five people, and often goes out after the deal has cooled, the value case is not “AI can write a proposal.” The value case is:
- shorten proposal cycle time
- reduce internal effort
- improve consistency
- preserve deal momentum
- increase the number of opportunities the team can support
That is a very different conversation.
Same with support. The value case is not “AI can summarize tickets.” It is reducing time to answer, improving first-contact resolution, lowering escalation volume, or giving junior team members enough context to handle more work safely.
Same with compliance. The value case is not “AI can answer policy questions.” It is catching exceptions earlier, reducing manual review, producing better audit packets, and lowering the cost of missed issues.
The metric tells you what kind of system you are building. Without the metric, you are just adding AI to work.
A simple way to sort the list
When the AI idea list gets too long, sort it by the metric each workflow can move.
| Workflow type | Metric that might move |
|---|---|
| Sales proposal | Cycle time, win rate, effort per proposal |
| Lead qualification | Conversion rate, speed to handoff, sales time wasted |
| Support triage | Time to answer, escalation rate, first-contact resolution |
| Finance reporting | Manual hours, review time, rework |
| Compliance review | Missed exceptions, audit findings, review cycle time |
| Contract review | Legal spend, approval time, risk exposure |
| Renewal triage | Churn risk caught, save rate, expansion opportunities |
| Customer onboarding | Time to launch, handoff errors, support burden |
This does not need to be complicated. The point is to stop treating all AI ideas as equal.
Some ideas are productivity helpers. Some are workflow improvements. A few are operating leverage. The first funded workflow should be in that last category.
Time saved is only one kind of value
A lot of teams default to time savings because it is easy to understand. That is fine, but it is incomplete.
If an analyst saves five hours a week, the model still needs to explain what happens next. Does the analyst cover more accounts? Review more exceptions? Reduce backlog? Spend less time preparing and more time deciding? Help the company avoid another hire?
Time savings without a destination becomes a soft claim.
The same is true for speed. Faster only matters if faster changes something. A faster proposal matters because the deal is still warm. A faster compliance review matters because the business can act sooner. A faster support answer matters because the customer experience improves and fewer tickets escalate.
The metric has to connect back to the business. That is what keeps the value case honest.
“Save time” is not a complete economic model. It is a starting point.
Not every good idea is ready
Sometimes the metric is real, but the workflow is not ready.
The data is messy. The source of truth is unclear. The rules are undocumented. The approval path changes by person. The examples are scattered. Nobody owns the process. The team cannot agree on what good output looks like.
That does not mean AI cannot help. It means the first step is not a build. The first step is understanding the work.
This is where a lot of AI pilots stall. They start with the output they want from AI, then discover too late that the workflow around the output is not stable enough to support it.
A better first workflow has a few things going for it:
- repeated work
- clear owner
- visible pain
- enough examples
- identifiable source of truth
- known review path
- measurable business lever
You do not need perfection. You need enough structure to build around.
The better first question
The question “Where can we use AI?” creates a brainstorm. The question “What metric can this workflow move?” creates a decision.
It makes the team define the unit of work. It forces a baseline. It reveals whether the workflow is tied to revenue, cost, speed, quality, risk, or capacity. It also exposes whether the idea is actually ready to build.
That is the shift. Do not start with the tool. Start with the workflow and the metric.
If the metric matters, and the workflow is ready enough to change, you have something worth funding. If the metric is vague, keep looking.
The first AI workflow should not prove that AI is interesting. It should prove that the company can change how work gets done.
Once you have picked the metric, the next step is to measure where the workflow stands today. That is why the baseline is the strategy.
More in this series, Proving AI Value:
- AI Usage Is Not AI Value
- What Metric Can This Workflow Move? (you are here)
- The Baseline Is the Strategy