Which use case,
and what's it worth?
Workshops produce forty ideas. That's the easy part and it isn't the problem. The problem is that nobody scores them against real numbers, so the loudest idea wins and the cheapest win in the building never gets found. This engagement ends in a ranked portfolio with a business case behind every line.
Ideas are free. Sequencing is the work.
An AI opportunity assessment finds the places in your business where AI creates measurable value, then ranks them so you know what to build first. Discovery is genuinely the cheap half. Any decent AI use case discovery workshop will fill a wall with sticky notes in an afternoon. The value comes afterward, when each candidate is priced, sized and scored against the ones next to it, and most of them get killed on paper.
The scoring is where I differ from a typical assessment. Every candidate carries a build cost, a run cost at your actual volume, a value estimate the finance team can audit, and a feasibility score grounded in whether your data can support it. That last column kills more ideas than any other, and it is much cheaper to be killed in a spreadsheet than in month five of a build.
The output is a portfolio, not a project. Two or three things worth starting now, a set worth revisiting once specific gaps close, and an explicit list of what you should not build, with the reason written down so the idea does not come back every quarter. I run this for companies across the United States and Canada, usually as the second half of an AI Diagnostic.
Also known as: AI opportunity assessment framework, AI use case assessment, AI use case discovery workshop, AI value assessment, AI business case development.
The four failures of
use-case selection.
Every one of these produces activity. None of them produce a return.
The loudest idea wins
The use case with the most senior sponsor gets built first, regardless of whether it was the best one on the list.
Value estimated by adjective
"Significant time savings." Nobody counted the hours, so nobody can tell afterward whether it worked.
Build cost only
The business case covers the project and stops. Inference, monitoring, evaluation and the humans supervising it never appear.
No kill list
Rejected ideas were never written down with a reason, so they return at the next offsite and get debated again.
Six columns behind
every ranking.
The model is published in the report, so your team can score the next round of candidates without me.
Value, counted
Hours, error rate, cycle time or revenue, measured against a real baseline rather than estimated in a room.
Build cost
Engineering weeks, integration work, and the data remediation the candidate quietly depends on.
Run cost
Inference at your volume, monitoring, evaluation, and the human supervision the workflow still needs.
Data feasibility
Whether the data exists, is accessible and is good enough. The column that kills the most candidates.
Risk exposure
What a wrong output costs: a customer, a fine, a safety incident, or an apology. It changes the design, not just the score.
Time to first value
Weeks until something real is in production. Early credibility funds the rest of the portfolio, so it counts.
Four weeks to a
defensible portfolio.
Two workshops, a set of interviews, and a lot of counting in between.
Discovery
- Interviews across operations, finance, sales and engineering
- Watch the work rather than reading the process documentation
- A structured discovery workshop with the people who do the job
- First candidate list, deliberately wide
Counting
- Baseline metrics per candidate: volume, handling time, error rate
- Data feasibility checked against real records, not schemas
- Vendor and build-vs-buy options priced for the plausible candidates
- Obvious non-starters cut, with the reason recorded
Business cases
- Build cost, run cost and value modeled per surviving candidate
- Sensitivity on the two or three assumptions that actually move it
- Payback period and first-year net in your CFO's format
- Risk and governance requirements attached to each one
The portfolio
- Ranked, sequenced, and split into now, later and never
- A one-page business case per candidate in the top tier
- The scoring model handed over so your team can run the next round
- Executive readout, with the kill list read out loud
A business case that only counts the build is a sales document. Inference, evaluation and the humans still supervising it are the numbers that decide whether you keep the system.
The ideation workshop
vs. the priced portfolio.
Both end with a list. Only one of them survives a CFO reading it.
Forty ideas, no order
- , Value described in adjectives
- , Feasibility assumed rather than tested against the data
- , Run cost absent from the model entirely
- , Nothing is ever formally rejected
Six candidates, ranked and priced
- →Value counted against a measured baseline
- →Feasibility checked against real records before scoring
- →Build and run cost modeled at your volume
- →An explicit kill list with reasons, so ideas stay dead
Quoted to scope.
Fixed fee, quoted once the scope is clear. What moves the number is how many parts of the business are in play and how much counting the baselines will take: a single operational area is a different engagement from a company-wide portfolio, and pretending otherwise with one published price would mislead most readers.
The commitments do not move. A defined scope agreed up front, a fixed fee against it, and a ranked portfolio with the scoring model handed over so your team can score the next round without me.
Most companies fold this together with the readiness side and run it as one AI Diagnostic, which is priced at a published $20,000 and is usually the better first engagement.
Not sure whether you need constraints or candidates first? Read the readiness assessment, or ask and I'll tell you which one your situation calls for. Email me for a quote ↗
About opportunity and ROI.
What is an AI opportunity assessment?
An AI opportunity assessment identifies where AI could create value in a business and ranks those candidates so leadership knows what to build first. It combines discovery, interviews and a workshop with the people doing the work, with scoring: value counted against a real baseline, build cost, run cost at your volume, data feasibility, risk exposure and time to first value. The deliverable is a ranked portfolio, a business case per top candidate, and an explicit list of what not to build.
How do you build an AI business case?
Start with a measured baseline: the volume of the process, the handling time per unit, the error rate and what an error costs. Model the value as the change in those numbers rather than as a percentage of a salary line. Then price both sides of the cost: engineering weeks and integration to build it, plus inference, monitoring, evaluation and continued human supervision to run it. Finish with a payback period and a sensitivity check on the two assumptions that move the answer most. A business case that survives is one where a skeptical CFO can attack each number individually.
How do you do an AI ROI calculation that holds up?
The AI ROI calculation itself is simple: net annual value divided by total annual cost, where total cost includes running the system rather than only building it. What makes AI ROI for companies hard to compare is that most published figures quietly omit half the denominator. The two mistakes that make AI ROI figures worthless are counting saved hours as cash when nobody's headcount actually changes, and omitting inference and supervision cost from the denominator. Count value the way it will show up: fewer errors, shorter cycle time, more volume through the same team, or revenue you can trace. If the only defensible benefit is time saved by people who stay employed doing other work, say that plainly and let leadership decide whether it's worth it.
Is there an AI business case template?
The one I use is a single page per candidate: the process baseline with three counted metrics, the proposed change, build cost in engineering weeks, run cost per month at current volume, first-year net and payback period, the two assumptions the answer is most sensitive to, and the risk and governance requirements. The report hands that template over along with the scoring model, so your team can build the next case without me.
How is this different from an AI use case discovery workshop?
A discovery workshop is one day and produces a candidate list. It's a component of this engagement, not the whole of it. The other three weeks are the part that determines whether the money is well spent: counting baselines, checking data feasibility against real records, pricing build and run cost, and ranking. Buying only the workshop leaves you with an unranked list, which is roughly where most companies already are.
What if the assessment says nothing is worth building?
It happens, and it's a legitimate result worth paying for. More often the answer is that nothing in the current list is worth building but two things nobody had proposed are, which is why the discovery goes to the people doing the work rather than the people who commissioned the study. Either way you get the reasoning, so the conclusion holds up when someone challenges it at the next board meeting.
Related paths.
Got a list nobody
can rank?
Send me the top five ideas floating around your company. I'll tell you which one I'd price first and what I'd need to see to believe the number.