AI Transformation Series · 03← Lakshmi Peri

You Can't Fund Everything: Ranking AI Bets

More AI ideas is not progress. A ranked portfolio is. The hard part isn't scoring opportunities. It's refusing to average them.

Most organizations rank AI opportunities the same way: score each one on impact, feasibility, risk, adoption, and strategic value, weight the dimensions, sort by the total. It feels rigorous. It quietly destroys the information you needed.

The weighted-scorecard trap
CandidateImpactFeasibilityRiskAdoptionStrategicAvg
A — Claims triage automation934395.6
B — Meeting summary assistant488725.8

The scorecard says B beats A. But these are not comparable bets. A is a high-value initiative that needs preparation. B is a convenience tool nobody will fight for in next year's budget. Averaging made them look like the same decision. The five dimensions were never meant to be added together.

Five Dimensions, Three Jobs

Each dimension answers a different question. Keep them separate and the portfolio ranks itself.

Worth it?

IMPACT · STRATEGIC VALUE

The value side. How much does this move a number leadership tracks, and does it build capability you'll reuse? This decides whether the bet deserves to exist.

When?

FEASIBILITY · ADOPTION

The deliverability side. Is the workflow ready, the data accessible, the owner willing? This decides sequence, never worth. Low deliverability means later, not lesser.

How?

RISK

The governance side. Risk doesn't rank bets, it shapes them: where the human checkpoint sits, what gets recommended versus automated, who reviews. A high-risk bet isn't a worse bet. It's a bet with rules.

The averaging mistake is treating "when" and "how" as if they were "whether."

The Portfolio Map

Plot every candidate: value on one axis, deliverability on the other. Risk travels as a flag on the bet, not as a position on the map.

Value (Impact + Strategic) →

Prepare

High value · low deliverability

Your biggest bets, not ready yet. Fund the remediation (Framework 02 names it), then promote to Now. This quadrant is next year's wins.

Now

High value · high deliverability

Fund first. Baseline written, owner named, KPI tracked. These carry the program's credibility.

No

Low value · low deliverability

Decline politely and in writing. Revisit only if the business changes, not because someone re-pitches it.

Opportunistic

Low value · high deliverability

Easy but small. Take one or two for momentum and training value. A portfolio made only of these is theater.

Deliverability (Feasibility + Adoption) →
RISK FLAG A flagged bet stays where its value and deliverability put it, but ships with conditions: human review checkpoint, recommend-only mode, staged rollout, named approver. Risk changes the design, not the ranking.

Sequencing the Waves

A ranked portfolio still needs an order. Three waves, each with a job.

WAVE 1 · 0–90 DAYS

Prove

Two or three Now bets plus one Opportunistic quick win. Early proof buys the patience the bigger bets will need.

WAVE 2 · 3–9 MONTHS

Scale

Remaining Now bets, plus the first Prepare bets whose remediation finished. Wave 1 baselines become Wave 2 business cases.

WAVE 3 · 9+ MONTHS

Compound

Prepare bets promoted on evidence. By now the portfolio funds itself: measured Wave 1 value pays for Wave 3 ambition.

Balance rule: never a portfolio of only quick wins, and never one of only moonshots. One buys time, the other spends it.

How to Use This

STEP 1

Score separately

Rate all five dimensions per candidate. Resist the total column.

STEP 2

Map it

Value and deliverability place the bet. Risk gets a flag, not a position.

STEP 3

Assign waves

Now funds first, one quick win for momentum, Prepare gets remediation dates.

STEP 4

Decline in writing

The No quadrant gets a documented reason. Silence invites re-pitching.