Managing the Use-Case Portfolio
Module 2 · Strategy and Value

Managing the Use-Case Portfolio

Run AI as a portfolio: a single intake, a scoring model that weighs value, feasibility, and risk, stage gates with kill criteria, and value measurement that escapes pilot purgatory.

From a List of Ideas to a Portfolio

Every organization has more AI ideas than capacity. Without a portfolio discipline, the loudest sponsor wins, pilots multiply, and very few reach production. The CAIO’s job is to make the trade-offs visible and deliberate.

A portfolio has four parts: one intake, a scoring model, stage gates, and value tracking.


One Intake

All significant AI proposals come through one front door, whether built internally, bought from a vendor, or embedded in a new software purchase. The intake form should be short (one page) and ask:

  • What business problem, and which goal does it support?
  • Who owns the process and the outcome?
  • What data does it need, and is that data available?
  • Who’s affected, especially customers or employees subject to decisions?
  • How will success be measured, and what’s the baseline today?

The intake isn’t approval. It’s visibility, so you can see the whole portfolio and spot duplicates.


Scoring: Value, Feasibility, Risk

Score each proposal on three dimensions, using simple 1–5 scales that business leaders understand:

  • Value: size of the measurable impact, and its link to a strategic goal.
  • Feasibility: data readiness, technical maturity, process owner commitment, and time to impact.
  • Risk: impact on people’s rights or safety, regulatory exposure, reputational sensitivity, and how hard errors are to catch.
Fix foundations firstDo firstAvoid or parkSmall, quick winsABCDEFeasibility →Value →AClaims triageBRésumé screeningCMeeting summariesDDynamic pricingENovelty chatbotRed ring: high risk.More governance,not an automatic no.

High value and high feasibility with manageable risk go first. High value with low feasibility often means “fix the data first,” which may be a project in itself. High risk doesn’t mean no; it means more governance, covered in Module 4.

⚠️ Common Pitfall: Pilot Purgatory

Common mistake: approving many pilots with no plan or budget for production, so promising results stall and teams move on.

Before a pilot starts, agree what result would justify scaling it, who would fund production, and who would own it afterward. A pilot without a path to production is a demo.


Stage Gates with Kill Criteria

  1. 1 Intake

    Problem, owner, baseline

  2. 2 Discovery

    Data and feasibility check

  3. 3 Pilot

    Limited users, measured

  4. 4 Production

    Integrated, governed, supported

  5. 5 Scale

    More units, users, regions

  6. 6 Review

    Value still there?

Fails a gate → fix and resubmit, or stop and record what was learned

Each gate has explicit criteria set before the stage starts:

  • Discovery → Pilot: the data exists and is accessible; a process owner has committed people and time; risk tier is assigned.
  • Pilot → Production: the success metric beat the baseline by an agreed margin; error rates are acceptable with planned human review; users want it; production cost is known.
  • Production → Scale: value held up over several months; support and monitoring are in place.

Kill criteria are as important as success criteria. Stopping a project that isn’t working frees money and people for one that is. Celebrate well-reasoned stops so teams aren’t afraid to report bad news.


Measuring Value Honestly

Boards increasingly ask what AI spending has actually delivered. Credible measurement has four features:

  1. A baseline. Measure the process before AI. Without it, every improvement is a guess.
  2. A comparison. Where possible, compare AI-assisted and non-assisted groups, or before and after with other factors held steady.
  3. Business metrics, not usage metrics. Logins and prompts tell you about adoption. Cycle time, cost per case, error rates, revenue, and customer satisfaction tell you about value.
  4. Full cost. Include licenses, compute, integration, human review time, governance, training, and ongoing support.

Report value in ranges and be clear about what’s measured versus estimated. A CAIO who overstates early results will struggle to be believed later.

🧪 Try It: Score Five Use Cases

Take five AI ideas from your organization, or invent plausible ones:

  1. Score each 1–5 on value, feasibility, and risk.
  2. Plot them on a simple value-versus-feasibility grid, and mark the high-risk ones.
  3. For your top choice, write the pilot-to-production gate criteria and one kill criterion.
  4. Name the baseline metric you’d need to measure now, before any work starts.
🧠 Knowledge Check

1. What's the main purpose of a single AI intake?

2. What best prevents pilot purgatory?

3. Which metric best shows business value from an AI claims-processing tool?

Up Next

Module 3 covers how the organization is set up to deliver this portfolio: operating model, platform, and data.