There is a specific CFO you are writing for, and they have read the McKinsey report. They have seen the headline numbers - generative AI could add trillions in value, knowledge-worker productivity up by some eye-watering percentage - and they have arrived at your design budget with a pre-loaded expectation that AI should be producing a large, visible, near-term return. If your memo tries to match that hype, you lose, because your real numbers are smaller and the CFO will catch the gap. If your memo dismisses the hype, you lose, because you are contradicting a McKinsey report the CFO believes. The only memo that survives is the one that engages the report honestly, names the gap between industry promise and your realized return, and tells a credible story anchored to numbers you can actually defend. This lesson teaches you to write that memo, and the artifact is the memo itself - a published, CFO-grade ROI document with honest numbers, named caveats, and a story that holds up under a skeptic's questions.

The CFO You Are Actually Writing For

Start by understanding your reader, because the memo is a persuasion document aimed at a specific kind of skeptic. This CFO is not anti-design and not naive. They are sophisticated, they have allocated capital across many functions, and they have developed a finely-tuned detector for the gap between what a function claims and what it delivers. The McKinsey report has done two things to them: it has raised their expectation of AI returns, and - more importantly - it has primed them to be skeptical of anyone whose claimed AI return conveniently matches the report's most optimistic figure. They have seen three other function heads walk in this quarter claiming AI gave them exactly the productivity boost the report promised, and they have discounted every one of those claims as wishful pattern-matching.

This is the crucial insight: the CFO's skepticism is your opportunity, not your obstacle. The memo that matches the hype gets filed with the other three discounted claims. The memo that says "the McKinsey report projects X for knowledge work broadly; our measured realized return in design is smaller, here is exactly why, and here is the part that is real and defensible" stands out precisely because it does not match the hype. You are the one function head who engaged the report like an adult instead of using it as a prop. Honesty about the gap is not a weakness to be minimized; it is the single most persuasive move available to you, because it is the move none of your competitors for budget are making.

The Structure of a Memo That Survives

A CFO-grade ROI memo has a specific shape, and deviating from it signals that you do not know the genre. It opens with the bottom line, not the build-up, because a CFO reads the conclusion first and decides whether to read the rest. It then engages the external benchmark (the McKinsey framing) honestly. It presents the realized return with its caveats. It names what did not work. And it closes with a forward ask tied to the credible numbers. Each section is doing a specific job in the persuasion.

The Bottom Line Up Front

Open with the answer to the question the CFO is actually asking: what did the design-AI investment return, in their currency, this period. One or two sentences, honest and specific: "Our $200K design-AI investment this year produced a measurable return concentrated in three areas - faster validated-learning cycles, reduced quality debt, and held accessibility compliance under higher output - with an estimated avoided-cost value of $Y, smaller than industry projections for reasons I detail below." This gives the CFO the conclusion and signals immediately that you are going to be honest about the gap, which earns you the read.

Engaging the Benchmark Honestly

Address the McKinsey report directly rather than pretending the CFO has not read it. Acknowledge what it says, then explain - specifically and without defensiveness - why design's realized return differs from the broad knowledge-work projection. The reasons are real: design's AI gains are partly consumed by the verification tax (the time to make generated output trustworthy), the returns concentrate in quality and risk rather than raw speed, and the largest gains are compounding (reuse, system leverage) and therefore show up over years rather than in a single annual figure. Naming these is not making excuses; it is demonstrating that you understand your own numbers better than a generic report does, which is exactly the expertise a CFO is paying you for.

The CFO has discounted three claims this quarter that matched the report's optimistic figure. The memo that names the gap between industry promise and your realized return is the one that survives, because it is the one your competitors for budget are not writing.

The Honest Numbers

The core of the memo is the realized return, and it has to be built from the metrics from the previous chapter - cycle-time-to-prototype, design-debt burn, accessibility-compliance rate, design-system reuse - translated into the avoided-cost and risk-reduction terms a CFO models. The discipline is that every number is decomposable: the CFO can trace each dollar of claimed return to a specific measured metric and a specific cost the organization already counts. A bundled "design AI delivered $Y in value" with no decomposition reads as a black box and gets discounted; the same $Y built up transparently from named components survives.

The honest-numbers section also requires explicit confidence levels, because a CFO trusts a number that knows its own uncertainty more than one presented as precise. Some of your return is hard (avoided accessibility-remediation cost is close to a real invoice you did not have to pay); some is softer (avoided wrong-direction builds from faster learning cycles is a real but estimated counterfactual). Label each. "This component is high-confidence, traceable to actual remediation costs avoided; this one is a defensible estimate of avoided cost with a stated assumption." The CFO will not punish you for soft numbers honestly labeled; they will punish you for presenting a soft number as hard and getting caught. Calibrated confidence is itself a credibility signal, and a sophisticated CFO reads the presence of stated uncertainty as evidence that the rest of the numbers are trustworthy.

Naming What Did Not Work

The section that most distinguishes a credible memo from a sales pitch is the one that names what did not work, and most design leaders omit it out of fear. This is a mistake. A memo with no failures reads as either dishonest or incurious, and a sophisticated CFO trusts neither. Naming a failure - "we piloted tool X expecting Y and it did not deliver, here is what we learned and what we stopped doing" - does three things. It proves the memo is a genuine accounting rather than a budget defense. It demonstrates that you kill what does not work, which is exactly the discipline a CFO wants to see in someone they are funding. And it makes your wins more believable by contrast, because a leader who reports failures is a leader whose reported successes are not cherry-picked.

The skill is to name the failure with the learning attached, so it reads as evidence of good management rather than evidence of waste. "We invested in a generative-prototyping tool that produced impressive demos but whose output consistently failed our verification gate, so the net time savings were negative; we stopped using it for production work and retained it only for throwaway sales demos" is a failure that makes you look more competent, not less, because it shows you measured, you concluded, and you acted. The CFO is not looking for a function that never fails; that function does not exist and claiming to be it is the fastest way to be disbelieved. They are looking for a function that fails intelligently and learns, and the what-did-not-work section is where you prove you are that function.

The Credible Story That Ties It Together

Numbers without a narrative are a spreadsheet, and a CFO does not fund a spreadsheet; they fund a story they believe, supported by numbers they can check. The credible story for design-AI is not "we got faster." It is a specific, defensible narrative about how AI changed the economics of the design function in a way that compounds. The story has a shape: AI commoditized the production layer, which would have been a threat if we had let it degrade quality, but instead we redirected the freed capacity into the judgment work that the model cannot do and that actually drives value - catching the silent failures, paying down quality debt, and enriching the system that makes every future unit of work cheaper. The investment did not buy speed; it bought a higher-leverage design function whose returns grow over time.

This story survives a skeptic because every claim in it is backed by one of the honest numbers and qualified by one of the named caveats. The "we held quality while output rose" claim is the accessibility-compliance trend. The "returns compound" claim is the reuse trend. The "smaller than the report because of verification" caveat is the cycle-time definition. The story is not rhetoric layered over data; it is the data, narrated. When the CFO asks "but how do I know quality did not quietly degrade?" you point at the debt-burn trend. When they ask "why isn't this the McKinsey number?" you point at the verification tax. The story is bulletproof not because it is impressive but because every load-bearing sentence has a defensible number underneath it, which is the only kind of story a sophisticated CFO funds.

The Forward Ask

The memo closes with the ask, and the ask has to be proportioned to the credibility you just built, not to what you wish you could get. If your realized return is real but modest, ask for a modest, well-scoped continuation, not a doubling - because an ask wildly larger than your demonstrated return resets the CFO's skepticism and undoes the trust the honest numbers earned. The credible ask follows from the credible story: "the compounding metrics (reuse, debt paydown) are the ones trending most positively and have the longest runway, so I am asking to continue the investment at roughly its current level, with the next increment targeted specifically at the design-system substrate where the leverage compounds, and I will report against these same metrics next quarter."

Notice that the ask is tied to the specific high-leverage area the metrics identified, comes with a commitment to report against the same metrics, and is sized to match the demonstrated return. This is what a CFO funds: not a leap of faith, but a continuation of a measured investment that is trending well, with a clear next target and an accountability commitment. The ask should also include the honest downside - what the metrics would have to show for you to recommend pulling back yourself - because a leader who states the conditions under which they would cut their own budget is a leader a CFO trusts with the budget. The memo's final move is to position you not as a supplicant defending a line item but as a steward managing an investment, which is the posture that gets the next year funded.

A Worked Build-Up: What the Numbers Actually Look Like

Abstraction is easy to nod along to and hard to defend in the room, so make the honest-numbers section concrete with a worked example you could adapt to your own figures. Suppose the year's design-AI investment was $200K - seats, tokens, training time, and the verification and governance overhead that the metrics lesson insists you count honestly. The memo's value build-up might total an estimated $310K against that $200K, and the discipline is that the $310K is never presented as a single figure; it is assembled in front of the CFO from four labeled components.

The first component is avoided accessibility-remediation cost, say $90K, marked high-confidence, because it traces to specific defects the team caught before ship that would otherwise have required documented remediation at known rates - this is the closest thing to a real invoice you did not have to pay. The second is avoided wrong-direction build cost, say $120K, marked medium-confidence, estimated from two initiatives that faster validated-learning cycles let you kill before engineering committed, with the explicit stated assumption that each would otherwise have consumed roughly a quarter of build. The third is design-and-engineering efficiency from reuse, say $70K, marked medium-high, derived from the measured reuse-rate increase applied to component build cost. The fourth is avoided future rework from the design-debt burn-down, say $30K, marked low-to-medium, a defensible estimate of deferred cost reduced.

What makes this build-up survive a sophisticated reader is the structure of its honesty. You present the $310K only after the components, you note that the high-confidence component alone ($90K) nearly covers half the spend with near-certainty, and you label the softer components so the CFO can apply their own discount rather than feeling you applied a flattering one for them. A CFO can attack a bundled $310K with a single skeptical sentence; they cannot easily dismiss a $90K invoice-grade figure sitting next to three clearly-labeled estimates, because each piece invites scrutiny on its own terms and most of the pieces survive it. The worked build-up is the difference between a number the CFO has to take on faith and a number the CFO can audit, and the auditable one is the only one that earns the next year's budget.

The Traps That Sink the Memo

Four traps recur. The hype-match: claiming the McKinsey number for design, which gets you filed with the discounted claims. The antidote is engaging the gap explicitly. The black box: presenting a single bundled value with no decomposition, which a CFO reads as obfuscation. The antidote is decomposable numbers, each traced to a metric and a cost. The no-failure pitch: a memo with only wins, which reads as a sales document and triggers full skepticism. The antidote is the what-did-not-work section with learning attached. The oversized ask: requesting far more than the demonstrated return justifies, which resets the skepticism the honest numbers earned. The antidote is an ask proportioned to credibility with a stated self-cut condition.

All four antidotes share a logic: in a conversation with a sophisticated skeptic primed by hype, the persuasive move is consistently the honest one, because honesty is the scarce signal. Everyone else is bringing optimism; you bring calibrated, decomposable, self-critical honesty, and that is what cuts through. The memo is not an exercise in making design look as good as possible. It is an exercise in being the most trustworthy voice in a room full of people overclaiming, which is both the right thing to do and, against this particular reader, the most effective. Write the memo a skeptical CFO would write if they were on your side, and you will have written the one that gets funded.

Key Takeaways

• Your reader is a sophisticated CFO primed by the McKinsey report to expect large AI returns and primed to discount any claim that conveniently matches the report's optimistic figure. They have already discounted three hype-matching claims this quarter. Honesty about the gap is your scarcest and most persuasive signal.
• Structure: bottom line up front (the answer first), engage the McKinsey benchmark honestly (name why design's return differs - verification tax, returns in quality/risk not speed, gains compound over years), the honest numbers, what did not work, and a forward ask proportioned to credibility.
• Every number must be decomposable: the CFO can trace each dollar to a specific measured metric (from the previous chapter) and a cost the org already counts. A bundled value with no decomposition reads as a black box and gets discounted.
• Label confidence explicitly. Hard numbers (avoided accessibility-remediation cost) and softer estimates (avoided wrong-direction builds) get different labels. Calibrated uncertainty is a credibility signal; a soft number presented as hard and caught is fatal.
• Name what did not work, with the learning attached. A no-failure memo reads as a sales pitch and triggers full skepticism; a named failure with a kill decision proves the memo is a genuine accounting and makes the wins more believable by contrast.
• The credible story is the data narrated, not rhetoric over data: AI commoditized production, you redirected the freed capacity into judgment work that compounds. Every load-bearing sentence has a defensible number underneath it. Close with an ask sized to the demonstrated return, tied to the highest-leverage metric, with a stated condition under which you would cut your own budget - positioning you as a steward, not a supplicant.