AI Strategy & Portfolio
Choose a few AI bets tied to business outcomes, fund them in stages, and stop the ones that do not prove their value.
AI strategy & portfolio
An AI strategy is not a list of tools or a mandate to "use more AI". It is a small number of bets, each tied to a business outcome, with an owner, a budget and a way to tell whether it is working.
Most organisations need a portfolio across three kinds of bet:
For each bet, decide whether to buy (use a vendor's product), build on a platform (your engineers on hosted or open models), or partner. Buy where the problem is common, build where the workflow or data is your advantage.
Why it matters
Without a portfolio, AI spreads as dozens of disconnected pilots. Each looks promising in a demo, few have a measurable outcome, and most never reach production. This is often called pilot purgatory, and it drains budget and credibility at the same time.
A portfolio forces the hard choices early: which three to five bets matter most, who owns each, what evidence would justify more funding, and what would make you stop. It also lets you invest in shared foundations, such as a platform, a gateway, evaluation practice and governance, once, instead of every team rebuilding them.
Finally, it gives the board and executives an honest picture: what is being tried, what it costs, and what has been proven.
Building and running the portfolio
Gather demand: collect ideas from business leaders and the people who do the work, written as jobs with measurable outcomes.
Score them on value, feasibility and risk (see the use-case scoring chapter), and check the top candidates' data before committing.
Choose three to five bets, each with a business owner, a technical lead, a budget and success metrics with a baseline.
Fund in stages, with a decision at each gate:
Invest in shared foundations alongside the bets: a platform team, an LLM gateway, evaluation practice and a governance process.
Review the portfolio every quarter: scale what is proven, change what is close, and stop what is not.
A portfolio register
# Reviewed quarterly by the AI steering group
- bet: Claims intake automation
type: workflow # productivity | workflow | product
approach: build # buy | build | partner
business_owner: GM Claims
technical_lead: Platform team
stage: pilot # poc | pilot | scale | stopped
metric: Claims triaged per handler per day
baseline: 42
target: 60
latest: 55 (week 6 of 10)
spend_to_date_aud: 180000
next_gate: 2026-12-04
stop_if: Under 50 at the gate, or accuracy on fraud flags below 99%
- bet: Coding assistant rollout
type: productivity
approach: buy
business_owner: Head of Engineering
stage: scale
metric: Weekly active developers; change in time to merge (control team)
baseline: n/a; 21 h
target: 70% weekly active; no quality regression
latest: 48% weekly active
next_gate: 2027-01-15
stop_if: Weekly active below 40% after enablement programmeWorked example: from 23 pilots to four bets
A mid-sized insurer finds 23 AI pilots running across the business, with no shared measurement and three separate contracts for similar tools.
The steering group scores them in one workshop. Four become funded bets: claims intake (workflow, build), a coding assistant (productivity, buy), broker email triage (workflow, build) and a customer-facing policy assistant (product, held at proof of concept until governance is ready).
Eleven pilots stop, with a short note on why. The rest fold into the bets or wait for the next quarter. The saving funds a two-person platform team and an LLM gateway, which the four bets then share.
Six months later, two bets have moved metrics, one has been stopped at its gate, and the board sees one page instead of 23 slide decks.
In a real organisation
Every bet needs a business owner who will be measured on the outcome. A bet owned only by IT or a central AI team tends to produce a working system that nobody adopts.
Write stop criteria before starting, and use them. Stopping a bet at a gate is the portfolio working, not failing.
Budget for adoption, not just technology. Training, process change and support often cost as much as the build, and they decide whether the value appears.
Beware tool-led strategy. "We bought licences, now find uses" reverses the logic; start from outcomes and choose tools to serve them.
Challenge
Your CEO asks for an AI strategy by next month and has heard competitors are "using AI everywhere". You have 12 candidate ideas, a modest budget and no platform. Write the one-page proposal: which bets, which stage each starts at, what you will build once to share, and what the board will see in six months.
Checklist
Production checklist
0 of 8 checked
Knowledge check
What is the main purpose of stage-gated funding for AI bets?
Complete
You can now build an AI portfolio with owners, gates and stop criteria. Next: govern it with controls proportionate to risk.
Mark this chapter as complete
Track your progress and unlock the next chapter.