To get AI approved by an FMCG board, anchor the case to one P&L number, not a technology roadmap: name the commercial leak, size it in pounds, propose one use case with a 90-day proof point, and show the downside is capped below the procurement threshold. Boards do not reject AI because they doubt the technology. They reject it because the case is framed as activity, this needs changing to: they reject it because the case is framed as activity: models, dashboards, pilots, instead of margin, working capital, and decisions improved.
84% of UK FMCG leaders say they need to move faster on AI. Only 3% have reached full deployment. That gap is rarely caused by a board saying no to AI. It is caused by AI cases that ask the board to fund a capability rather than a commercial outcome, and boards, correctly, decline to fund capabilities they cannot measure.
What does an FMCG board actually want to see in an AI proposal?
A board wants three things, in this order: the commercial problem in language they already use, the size of the prize in pounds, and the size of the risk if it does not work. Everything else — the model, the data architecture, the vendor — is implementation detail they are trusting you to handle.
The most common failure is leading with the solution. "We want to build a demand forecasting model" invites scrutiny of the model. "We are losing an estimated £280K a year to trade spend that delivers negative incremental ROI, and we cannot currently see which promotions" invites scrutiny of the problem, which is exactly where you want the conversation. Lead with the leak, not the build.
This is the same reason most AI pilots never reach production: the case was built around technology rather than a bounded commercial outcome the business already cared about.
How do you frame AI ROI in terms your CFO will sign off?
A CFO signs off on a number they can audit, not a percentage they cannot trace. "AI maturity uplift" protects nobody and persuades nobody. One concrete metric — margin recovered, working capital freed, write-off reduced, hours redeployed — with a timeline against which to measure it, does both.
Translate the AI outcome into the CFO's own vocabulary before the meeting. A forecasting accuracy gain is not a CFO metric. The reduced write-off and freed working capital it produces are. Do the translation yourself; if you make the CFO do it, they will discount the whole case to zero. State the assumptions in writing so the number can be challenged and survive the challenge.
This is why the £150K AI software trap catches so many mid-market CPG businesses: the purchase is justified in technology terms, and the CFO has no P&L line to hold it against six months later.
What are the five steps to build an AI business case that gets approved?
This is the structure that gets AI funded in mid-market FMCG. Follow it in sequence; skip any step and the case weakens at exactly the point a board director will probe.
Step 1. Name the commercial leak, not the AI solution. One specific, measurable problem the board already recognises: trade spend delivering negative incremental ROI on 30% of promotions. Demand forecasting running at 12% manual correction rate across three markets. Write-off on short-life SKUs averaging £180K per quarter. The problem defines the use case, not the other way around.
Step 2. Size it in pounds against the P&L line it bleeds from. Attach a pound figure the CFO can trace to an existing line in the management accounts. "We estimate £240K in avoidable write-off per year, sitting in the cost of goods line." Not a percentage. Not an efficiency gain. A number they already half-know is leaking somewhere.
Step 3. Estimate the AI-enabled improvement conservatively. State the improvement range, not a guarantee: "Demand forecasting accuracy improvements in comparable mid-market CPG businesses have reduced write-off by 15 to 25%. We are modelling 12% as our conservative case, equivalent to £29K recovered in the first 90 days." Underpromise. Let the result beat the case.
Step 4. Cap the downside below the procurement threshold. A fixed-scope, fixed-price diagnostic below £25K removes the procurement committee and reframes the decision from an open-ended bet to a bounded test. The worst case is a known, small number. Name it explicitly in the proposal: "If we find no viable use case, the cost is £9,000 and we have a 15-page diagnostic we would not have had otherwise."
Step 5. Name the commercial owner and commit to a 30-day deliverable. Boards fund accountability for an outcome. Name the senior sponsor with authority to act on the output, state what they are accountable for, and commit to a hard deliverable at day 30. A working output in beta by day 45 that the commercial team uses weekly. Validation against the P&L by day 75.
The AI FlightCheck™ is built around exactly this structure: a fixed scope, fixed price, below most procurement thresholds, producing a 15-page diagnostic and 90-day action plan the board can read in one sitting.
From the FlightCheck™ Files — what boards actually ask
Across the AI business cases AI Navi has helped take to boards, four questions come up in almost every meeting. Prepare these and you have prepared 80% of the room.
"What happens if it does not work?" Answer with the capped downside. A fixed-scope engagement below the procurement threshold means the worst case is a known, small number, not an open-ended programme. That single move removes the committee and shortens the approval cycle.
"Why now, and not in 12 months?" Answer with the cost of delay, not the excitement of the technology. Every quarter the leak runs is margin gone. A business losing £180K per year to avoidable write-off loses £45K for every quarter the decision is deferred.
"Who owns this?" Boards fund accountability, not activity. Name the commercial sponsor with authority to act on the output, in the proposal, before the meeting.
"How is this different from the last pilot that went nowhere?" Answer with the structural difference: a P&L target agreed before any build, and a working output inside 30 days they can challenge. The case studies from AI Navi's production deployments show what this looks like in practice: four working systems, none of which started with a pilot.
Notice none of those questions are about the technology. McKinsey's State of AI research confirms this pattern: boards do not lack appetite for AI investment; they lack confidence in the framing and ownership structure of the cases being presented.
What is a Flight Risk Index™ and why does it work as a board metric?
The Flight Risk Index™ is a single 0-to-10 score of how likely an AI initiative is to stall before it reaches production, scored across strategy alignment, data readiness, and execution capability. It works as a board metric because it gives directors a number they can track quarter on quarter, the same way they track any other risk on the risk register.
In one case, a £400M CPG business scored 7.2 out of 10 — high risk — driven primarily by demand forecasting still managed in Excel across three markets. After a single data pipeline project, the score dropped to 4.1 inside 60 days. The board did not approve "an AI programme." They approved reducing a measured risk from 7.2 to a target, which is a decision boards know how to make.
This is also why the AI FlightCheck™ diagnostic produces a Flight Risk Index™ score as its primary output: it translates AI readiness into a metric format the board can act on, not a capability assessment they cannot.
The three AI business cases that get approved and the three that do not
| Gets approved | Gets rejected | |
|---|---|---|
| Framing | One commercial leak, sized in pounds | A capability or platform, with benefits described as "efficiency" or "insight" |
| Cost structure | Fixed scope, fixed price, below procurement threshold | Open-ended, requiring a leap of faith |
| Ownership | Named commercial sponsor accountable for the P&L outcome | Owned by a data team reporting below board level |
| Timeline | Working output in 30 days, measurable result in 90 | Discovery and workshops in month one, then a wait |
| Downside | Explicitly stated, capped, and small | Implied, open-ended, or not mentioned |
The technology is usually identical. The framing is not.
How do you structure the first 90 days so the board sees progress, not promises?
Commit to a hard deliverable at day 30, validation at day 90, and a board update at each. The first 90 days of any AI initiative tell the board what the next year will look like. If the first 30 days are workshops and discovery with nothing to show, expect the same for months two and three, and the board will too.
A structure that holds up: a two-week diagnostic that produces a 15-page action plan the board can read; a first working use case in beta by day 45 that the commercial team uses weekly; validation against the P&L by day 75; a board presentation at day 90 with the ROI measurement framework attached. Each checkpoint is a moment the board sees evidence rather than receives a promise.
For businesses that have already been through a stalled AI programme and are taking a second case to the board, this 90-day structure also answers the hardest implicit question in the room: why will this time be different.
A fractional CAIO is the model most mid-market FMCG businesses use to hold this structure to account: senior commercial and technical accountability in a single role, without the 9-to-12 month hiring cycle or the £270K-plus cost of a full-time hire.
Frequently Asked Questions
How do I justify AI investment to a sceptical board? Lead with the commercial leak, not the technology. Size the problem in pounds, propose one use case with a capped downside below the procurement threshold, and commit to a working output in 30 days. A sceptical board is usually sceptical of open-ended technology spend, not of fixing a measured commercial problem.
What ROI should I promise the board from AI? Promise a measurement framework and a conservative working range, not a guaranteed figure. For trade spend optimisation, 6 to 9% margin recovery on rebalanced spend inside 90 days is a defensible range based on comparable mid-market CPG cases. Underpromise against ranges you can evidence, and let the result beat the promise.
Should the data team or the commercial team own the AI business case? The commercial team should own it, with the data team as a named delivery partner. Boards fund accountability for an outcome. A case owned by a function that cannot be held to a P&L result reads as activity, and is funded accordingly — which is to say, often not at all.
How long should an AI business case take to show results? A well-structured case shows a working output inside 30 days and a measurable commercial result inside 90. If the proposal cannot commit to either, it is structured for activity rather than outcome, and the board is right to question it.
How do I cap the risk on an AI proposal so the board approves it? Use a fixed-scope, fixed-price engagement below your procurement threshold — typically under £25,000 — so the worst case is a known, small number. This removes the procurement committee, shortens the approval cycle, and reframes the decision from an open-ended bet to a bounded test.
Where to Start
If you are about to take an AI case to your board and want it framed around a measured risk rather than a technology wish-list, the AI FlightCheck™ produces exactly that: a Flight Risk Index™ score, a 15-page diagnostic, and a board-ready 90-day action plan, in two weeks, below most procurement thresholds. Or take the free three-minute AI Readiness Scorecard first to see where your case is strongest before you build the slide.