Briefing

Published model · Version 1.0

How we score AI use cases

The model behind every AI use-case sprint, published so you can check our math before you hire us, or use it on your own list.

  • Four criteria, scored 0 to 100
  • Weights 35, 25, 20 and 20
  • Four tiers, set by rules
  • A confidence label on every dollar figure

It’s the scoring step of our method: read, score, gate, sign.

Four criteria, one score

Each use case is scored 0 to 100 on four criteria, in workshops with the people who do the work. Every score has a written reason.

Value

35%

What is it worth a year, and does it move a number leadership watches?

85–100
Moves a number leadership tracks, such as win rate, margin, uptime or customer quality, and is worth a lot at your scale.
70–84
Clear dollar value in one function, or strategic weight beyond the dollars.
50–69
Useful time saved, with modest dollars.
Under 50
Nice to have.

Feasibility

25%

Can it be built and adopted within months, with the tools and skills you have?

85–100
Runs on tools you already own, follows a proven pattern, and a pilot fits in a quarter.
70–84
Some integration or change work, still within a quarter or two.
50–69
Needs a new platform, a custom model or skills you’d have to hire.
Under 50
Unproven tools, or a long build.

Data readiness

20%

Is the data there, clean enough, and allowed to be used?

85–100
Digital, in one or two systems, clean, and clearly yours to use.
70–84
In your systems, but it needs extracting, cleaning or labeling.
50–69
Spread across spreadsheets and email, partly missing, or with permissions to confirm.
Under 50
Doesn’t exist yet, or contracts, consent or NDAs keep it off limits.

Risk

20%

What happens when it’s wrong, and how easily does a person catch it? Scored so that higher means safer.

85–100
Internal and low-stakes: mistakes are cheap, visible and easy to undo.
70–84
A person approves before anything reaches a customer, a supplier or the books.
50–69
Touches customers, money or safety, with controls that still need work.
Under 50
Would act on its own on money, safety or customers.

The score

A weighted average, rounded to one decimal before the tier rules apply.

Score = 0.35 × Value + 0.25 × Feasibility + 0.20 × Data readiness + 0.20 × Risk

From the sample sprint

RFQ quoting assistant

Value 92 × 0.3532.2
Feasibility 74 × 0.2518.5
Data readiness 72 × 0.2014.4
Risk 78 × 0.2015.6
Score80.7

Quick win. Both bars cleared, so it goes first.

From the sample sprint

Vision inspection of machined surfaces

Value 88 × 0.3530.8
Feasibility 44 × 0.2511.0
Data readiness 48 × 0.209.6
Risk 66 × 0.2013.2
Score64.6

Strategic bet. The most valuable idea on paper waits until the defect images and the tools exist.

Both examples come from our sample sprint for a fictional manufacturer.

Four tiers, set by rules

The rules run in this order, and the first one that fits decides the tier.

Quick win

Score 72 or more, and feasibility 70 or more

Pilot now. These pay for the rest of the program.

5 of 18 in the sample sprint

Strategic bet

Value 75 or more, but not a quick win

Worth more on paper, but harder or riskier to build. Close the data or platform gap first, then pilot.

5 of 18 in the sample sprint

Next wave

Score 62 or more

Queue them behind the quick wins. Many get cheaper once the first platform is in.

6 of 18 in the sample sprint

Parking lot

Everything else

Revisited each year, or sooner when the data or the tools change.

2 of 18 in the sample sprint

A confidence label on every dollar figure

Value leans on dollar estimates, so every estimate says how far to trust it. The label doesn’t change the score. It tells you how much weight to put on the number.

HighBuilt from your own records, and the improvement is backed by a pilot, a time study or a comparable deployment.
MediumBuilt from your own records, and the improvement is a stated assumption the pilot will test.
LowThe baseline is estimated or the data is incomplete. Treat the figure as a rough guide until the data exists.

Each business case is also stress-tested at three months late and at half the value, so you can see what a miss would cost.

Rules that keep it honest

  • Weights are set before scoring

    Use the model as published, or change the weights at kickoff. If you change them, we show the ranking both ways.

  • Every score has a reason

    Written down in the workshop with the people who do the work, and open to challenge in the readout.

  • Feasibility counts the tools you own

    Not the tools anyone would like to sell you. The platform is picked after a use case earns its place, and vendor funding never changes a score.

  • Under 40 on risk, a person approves every action

    Whatever its tier, a use case scored under 40 on risk doesn’t go ahead until a person approves every action it takes.

  • An architect signs the scores

    On client readouts, Lai Long, founder and principal architect, reviews and approves every finding, scores included.

Try it on one idea

Move the four scores and the model does the rest. Nothing you enter leaves your browser.

Examples from the sample:

Showing the sample’s RFQ quoting assistant.

92

74

72

78

Weighted score

80.7

Quick win

This is the scoring step only. In a sprint, the scores come from workshops and your own data, and every one is reviewed before it reaches the readout.

Version 1.0, published October 16, 2026. When we change the model, we’ll post the new version here with the date and what changed.

Score your own list.

In a two-week AI use-case sprint, we score 15 to 20 of your use cases on this model and build the business cases for the top three. Fixed fee: $12,500. See a sample readout

Book the sprintSee the offer