A P&L Owner's Playbook for Approving AI Investments.

IN ONE MINUTE

Not every AI investment should be judged by the same test. A quick automation win, a multi-year scaling bet and a strategic transformation carry different risks and time horizons. Applying one hurdle rate or payback rule to all of them tends to approve small, easy wins and starve the bigger bets that change a company's position. A three-tier approach helps: payback for quick wins, economic value added with downside modelling for scaling bets, and strategic conviction with a named capability for transformations. Name the tier before the debate starts.

THE OBJECTIVE

A conversation with a CFO this month has stayed with me. The question was simple to ask and hard to answer: what return should we expect from AI, and should every initiative face the same bar? This issue shares where we landed.

WHAT THE RESEARCH SHOWS

• Grant Thornton CFO Survey (Q1 2026, 233 CFOs): 68% expect IT and digital transformation spending to rise, the highest in the survey's 21-quarter history, while 62% are confident they will achieve their technology objectives.
• BCG (Jul 2026): only 14% of CEOs clearly define the P&L impact for all their AI initiatives. High performers are about five times more likely to do so, and bring finance in from day one.
• McKinsey (Jul 2026): about 60% of an agentic AI task's cost goes into refining answers, and 93% of survey respondents reported exceeding their AI budgets.

Investment is rising faster than the confidence to judge it. The tools finance teams use (WACC, hurdle rates, payback, EVA) are sound. The question is which one fits which decision.

WHY ONE HURDLE RATE ISN'T ENOUGH

A single discount rate treats every initiative as if it had the same risk and time horizon. In practice:

• Low-risk, incremental wins clear the bar easily and get approved.
• Longer-horizon bets look weak in year one against the same yardstick, even when they are the ones that shift competitive position.

Payback period works well for a quick win. For a capability investment, the honest answer is often: "This doesn't pay back on its own. It changes what everything after it can pay back."

THE FINANCE TOOLKIT FOR AI DECISIONS

The tools are familiar. What changes with AI is how they should be applied:

Measure	What it tells you	How to apply it to AI
Payback period	How quickly the cash invested comes back	Best for Tier 1 automation. Include run costs, not just build costs.
NPV and IRR	Value created after the cost of capital	Use a risk-adjusted discount rate: higher for unproven use cases, lower once the workflow is running at scale.
EVA	Whether operating profit exceeds the capital charge	Useful for Tier 2 scaling bets, where AI changes the economics of a whole function.
Total cost of ownership	The full cost over the life of the solution	Add inference and token costs, data work, integration, monitoring, model updates and people. Issue 010 sets this out line by line.
Unit economics	Cost and margin per transaction, case or customer	Track cost per case before and after AI. It is the clearest early signal that value is real.
Sensitivity analysis	Which assumptions the case depends on	Stress-test adoption rates, usage volumes and compute prices. These often move more than expected.
Real options thinking	The value of the right to invest further later	Fund Tier 3 in stages. Each stage buys information and the option to scale, pause or stop.

WHERE AI BUSINESS CASES GO WRONG

In my experience, AI business cases rarely fail on the maths. They fail on the assumptions behind it:

• Soft savings counted as hard savings. Hours freed up are not cash saved unless the capacity is redeployed, hiring is avoided or spend is actually reduced.
• Run costs left out. Usage-based pricing turns AI into a variable operating cost. A case built on one-off build cost alone will overstate the return.
• Double counting. Several initiatives claim the same saving or the same revenue uplift. Keep one benefits register across the portfolio.
• Benefits never banked. Savings are approved in the business case but never removed from the budget, so they are quietly reabsorbed.
• No baseline. Without a clear before-and-after measure, nobody can prove what changed.

THE FRAMEWORK: THE THREE-TIER INVESTMENT TEST

Tier	Horizon	Decision owner	Primary test
1. Quick Win	Under 12 months	Business unit lead	Standard hurdle rate and payback period
2. Scaling Bet	1–3 years	Functional executive with P&L	Economic Value Added, with a modelled downside case
3. Strategic Transformation	3+ years	CEO and executive committee, with board visibility	Strategic conviction, bounded downside, named capability

Tier 1 (Quick Win): low uncertainty and reversible. Payback should gate the decision. If it's taking weeks of senior time, the process needs attention, not the project.

Tier 2 (Scaling Bet): real but boundable uncertainty. Approve on positive EVA across a realistic base case, and always review the downside case, not only the upside.

Tier 3 (Strategic Transformation): three questions matter more than a single number. What capability does this build that we don't have today? What does delay cost if others move first? How reversible is it if we're wrong? Where an investment builds a durable capability, discuss with finance whether it should be capitalised and assessed over its useful life rather than a single year.

The identity test for Tier 3: can you name the specific capability it builds, and where it lands on the target operating model? If not, it's probably a Tier 1 or 2 project and should face that test.

WHAT THIS MEANS FOR CFOS AND FINANCE TEAMS

Set the tier before the debate. Agree which tier an initiative belongs to before the numbers are discussed. It keeps the conversation about the right test, not about whose forecast is more optimistic.

Build a benefits realisation discipline. Link each approved AI initiative to a baseline, an owner and a tracked P&L line. Report actuals against the business case every quarter, and move banked savings out of next year's budget.

Treat compute as a managed cost line. Put AI usage into the monthly management accounts, with budget ceilings, variance tracking and a named owner, as you would for any other material operating cost.

WHAT THIS MEANS FOR P&L OWNERS

Show the EBITDA bridge. Explain in plain terms how the initiative moves revenue, gross margin or operating cost, and when. A clear bridge from today's P&L to the target builds more confidence than a single ROI figure.

Separate the one-off from the run-rate. Be clear which costs and benefits are one-time and which recur. Recurring benefits are what change the value of the business.

Commit to the downside as well as the upside. Present a realistic downside case and the trigger points that would lead you to pause or stop. It builds trust with finance, and it makes approval of the next stage easier.

WHAT THIS MEANS FOR CEOS AND BOARDS

Protect the strategic bets. If every AI proposal faces the same hurdle rate, the portfolio will drift towards small, safe wins. Reserve explicit capacity for Tier 3 investments, judged on strategic criteria.

Ask for the portfolio view. Review AI spend as a portfolio across the three tiers, with the balance of quick wins, scaling bets and transformations visible in one place.

Fund in stages, decide on evidence. Stage-gated funding lets the organisation move quickly while keeping capital at risk proportionate to what has been proven.

THE LAST WORD

Anyone can own a number. Judgement is knowing which number the decision depends on.

Which tier does your organisation tend to favour, and where do your best strategic ideas usually get stuck?

—

Until next week,

Artificial Ian

Tell your network: if a finance partner is setting next year's investment criteria, the three-tier table could be a useful starting point.

→ Found this useful? Forward it to a CFO, strategy leader or finance partner.

→ Subscribe to The AI Executive Briefing, weekly in your LinkedIn inbox.

Sources: Grant Thornton CFO Survey (Q1 2026); BCG, "What Separates CEOs Who Scale AI From Those Who Don't?" (Jul 2026); McKinsey, "Is that AI agent worth it? Agentic economics and the modern operating model" (Jul 2026).