Part 1 — Screen It
The Value, Cost & Risk Canvas
Someone brings you an AI idea. A vendor pitches a solution. A team proposes an initiative. Before anyone builds a spreadsheet or writes a business case, run it through two checks: five minutes with the questions below, then one page with the Canvas. Together they tell you whether an idea deserves real work, or a polite no.
Be the realist here, not the skeptic. The job isn’t to block AI investments — it’s to make sure the ones that go forward are measured properly.
First, five minutes: the screener
Score each answer green, amber, or red. Green means clear and specific. Amber means the answer exists but is vague. Red means there is no answer at all. Any red sends the proposal back for more homework — it doesn’t go into a budget.
| Question | What a good answer looks like | The red flag |
|---|---|---|
| What business KPI does this move? | A named metric: average handling time, conversion rate, error rate, satisfaction score | “Improves efficiency.” If there’s no KPI, it’s a science experiment, not a business case. |
| How will you measure it? | A baseline today, a way to track it afterward, and access to the data | “We’ll figure it out later.” Fix that before you invest, not after. |
| What specifically changes in the process? | Which steps AI does or assists with, and what people do differently afterward — you should be able to draw a before and after | “AI will handle everything.” AI almost never replaces a whole process. A fuzzy answer means nobody understands the solution well enough to evaluate it. |
| What does this really cost? | Implementation including internal time, plus annual operating cost — training, process redesign, maintenance | A vendor quote on its own. Add 30–50% for what hasn’t been thought of. If it still looks good with that buffer, good. |
| What’s the main risk, and how is it mitigated? | The single biggest risk for this specific project, named, with a plan beside it | “We’ll deal with it if it happens.” That isn’t a mitigation. |
All green: move to the Canvas below. Any amber: send it back with the specific gaps named. Any red: not ready for investment.
Then, one page: the Canvas
A screening tool, not a substitute for the four-step ROI calculation that follows it. One canvas per project idea.
1. The value: what kind, and how much of it
Score each type High, Medium, or Low for this specific project — don’t assume all four apply.
- Revenue — more sales, better conversion, higher average ticket
- Cost reduction — time saved, fewer errors and less rework, full automation
- Customer or client experience — shorter response times, personalization, retention
- Risk reduction — fraud avoided, regulatory errors avoided, critical operational failures avoided
Most real projects are strong on one of these and weak-to-absent on the other three. That’s normal. Claiming strength on all four is the tell that nobody has actually thought it through.
2. The cost: direct and indirect
Direct costs appear on an invoice. Indirect costs don’t — and they’re the ones that get underestimated.
Direct:
- Software licenses — platform, model usage, management tooling
- Infrastructure — servers, storage, compute (often bundled into the platform subscription)
- Consultancies or providers — usually the largest single line item up front
Indirect:
- Internal staff time — hours × cost per hour, across the business, IT, data, and legal
- Training — materials, sessions, and the time of the people being trained
- Process redesign — workflows, roles, change management. This is the line most often left out, and the one that turns a good projection into a disappointing outcome.
- Maintenance — monitoring, retraining, adjustments. Recurring, not one-time.
3. The risk: what could eat the value
Three risks show up on nearly every AI project. Name them for yours specifically — a generic risk line is worth nothing.
- Model quality and bias errors — what actually happens when it gets one wrong? For a mid-sized company, this is usually a delay or an annoyance, not a catastrophe — but say which, don’t leave it unstated.
- Low internal adoption — this is the risk that most often turns a strong projected ROI into something close to zero. If the people expected to use the tool haven’t asked for it, this line goes from Low to High.
- Legal and reputational exposure — data privacy, GDPR, and AI-specific regulation now catching up with it. Even a low-stakes internal tool touches this if it processes personal data.
4. The verdict
Three yes/no questions, then a decision:
- Is the value high enough? Y/N
- Is the cost reasonable? Y/N
- Is the risk acceptable? Y/N
- Decision: Go / Rework / Drop
Business-case screening and governance screening are different questions. This canvas answers the first. The seven-point use-case fit note in the Documentation Register answers the second.
The rule at the bottom of every canvas
The best AI project is not the one with the most potential value. It’s the one with the best combination of high value, reasonable cost, and acceptable risk. If any one of the three fails, the other two don’t rescue it. A project with enormous upside and unmanaged adoption risk is not a good bet — it’s a story you’ll be explaining in nine months.