Most AI investments in Consumer Products don't fail because the technology was wrong.
They fail because nobody ran a basic check before the money moved.
A £400M CPG group we worked with had already spent £80,000 on an AI demand forecasting platform before we were called in. The platform was credible. The vendor was reputable. The problem: four years of sales data sat in a legacy ERP that couldn't be reliably extracted. The AI had nothing clean to learn from.
Eight months. £80,000. Zero production output.
This is not unusual. According to Gartner, 30% of GenAI pilots were abandoned by 2025 — not because AI doesn't work, but because the business wasn't ready for it.
Before you commission a consultant, sign a vendor contract, or brief your board on an AI roadmap, run this check. It takes 20 minutes. It has saved our clients hundreds of thousands of pounds in misdirected spend.
Why a Pre-Investment AI Check Matters
The instinct in most CPG businesses is to start with the solution — a platform, a vendor, a pilot. That instinct is expensive.
The companies that get AI to production fastest — and we have worked inside enough of them to say this with confidence — start with the problem. They ask what is broken, what data exists to fix it, and whether the organisation can actually adopt a solution before they spend a penny on building one.
This checklist maps to the five structural gaps we see most often in mid-market Consumer Products businesses. We call them the SCALE AI™ dimensions: Strategy, Capability, Applied AI, Leadership, and Enhanced Data. Every failed AI programme we have encountered has broken down at one or more of these points.
Work through each check honestly. Score yourself zero (not in place), one (partially in place), or two (fully in place). A score below 12 out of 20 means your AI investment is at material risk before it starts.
The 10-Point AI Check
1. Can you name the exact commercial decision your AI should improve?
Not "improve our forecasting." Not "better supply chain visibility."
The specific decision: Which SKU, which market, which week, which buyer?
AI does not improve vague ambitions. It improves specific, repeated, high-stakes decisions that currently rely on manual judgement or incomplete data. If you cannot write that decision in one sentence, you are not ready to invest in AI. You are ready to invest in clarity.
Score 0 — No clear decision identified. Score 1 — A decision area identified, but not pinned to a specific output, frequency, or owner. Score 2 — One specific, high-value decision named, with a clear owner and measurable outcome attached.
2. Is your AI investment connected to a P&L outcome — not a technology goal?
"Deploy a machine learning pipeline" is a technology goal. "Reduce promotional waste by 8% on a £40M trade spend" is a P&L outcome.
Boards block AI budgets when the business case is written in technology language. They unlock them when the case is written in EBITDA language.
At pladis, we did not present AI to the board as an AI project. We presented it as a £2.3M margin recovery programme. The AI was the method. The margin was the ask. Budget released in one meeting.
If your current AI proposal cannot be expressed as a specific P&L line — revenue uplift, cost reduction, margin recovery, waste elimination — it is not ready to present to anyone who controls budget.
Score 0 — No P&L framing in place. Score 1 — A business case exists but relies on vendor projections rather than internal commercial data. Score 2 — A specific P&L outcome identified, costed, and owned by a commercial director or CFO.
3. Do you have two years of clean, labelled data for the decision you are targeting?
This is the check most businesses fail. Not because they lack data — CPG companies accumulate enormous volumes of it — but because it is fragmented, inconsistent, or trapped in systems that cannot talk to each other.
Five years of sales data in a legacy ERP. Promotional data in spreadsheets. Returns data in a separate WMS. Retailer data in emailed Excel files from Tesco's trading team.
AI cannot learn from data it cannot access. And it cannot learn reliably from data that means different things in different systems.
Before any AI investment, run a basic data extraction test on your target use case. Can you pull two years of clean, consistently labelled records for the decision you are trying to improve? If the answer is no, your first investment is not AI. It is data infrastructure.
Score 0 — Data is fragmented across systems with no reliable extraction path. Score 1 — Data exists but requires significant manual cleaning before use. Score 2 — Clean, centralised, consistently labelled data is accessible for the target decision.
4. Has your organisation run an AI pilot that never reached production?
If yes, this is not a failure to note and move on from. It is a diagnostic signal.
Pilots die at the same points every time: data integration, change management, or loss of the internal champion who drove them. Understanding which of these killed your last pilot tells you exactly where your next investment needs to go first.
We have helped three CPG companies recover from failed pilots in the last 18 months. In every case, the pilot itself was sound. The problem was one of three things: the data was not clean enough to scale, the steering committee lost confidence and pulled the budget, or the one person who understood the system left the business.
A failed pilot is not a reason to distrust AI. It is a reason to fix what broke before spending again.
Score 0 — Failed pilot, no post-mortem conducted, no root cause identified. Score 1 — Pilot stalled or failed, some analysis done but no structural changes made. Score 2 — Pilot outcomes reviewed, root cause identified, structural gap addressed before next investment.
5. Does your C-suite have a named AI sponsor with budget authority?
Not an enthusiast. Not a champion at director level who will present to the board.
A named senior leader — CEO, COO, or CFO — who has publicly committed to the outcome, has the budget authority to release spend, and whose performance review includes AI delivery.
Without this, your AI programme will stall at the first steering committee. We have watched it happen in businesses with genuinely strong data infrastructure and credible use cases. The data was ready. The technology was ready. The organisation was not. There was no one at the top who owned the outcome.
AI transformation is a change management problem as much as a technology problem. Change moves at the speed of the most senior person who is personally accountable for it.
Score 0 — No named C-suite sponsor. Score 1 — A director-level sponsor, but no C-suite accountability. Score 2 — A named C-suite leader with explicit ownership, budget authority, and a delivery milestone in their objectives.
6. Does your team have a single, trusted source of truth for your core commercial KPIs?
Three versions of the same gross margin report. Two demand forecasts — one from finance, one from supply chain — that never agree. A category performance dashboard that nobody trusts because the underlying data pulls from different time windows.
We see this in almost every mid-market CPG business we enter. It is not a data technology problem. It is a data governance problem.
AI amplifies the quality of the data it receives. If your organisation cannot agree on a single version of gross margin, AI will produce four confident, contradictory predictions — and nobody will act on any of them.
Fixing this does not require a data platform. It requires a decision about who owns each KPI definition and one person with the authority to enforce it.
Score 0 — Multiple conflicting KPI sources, no resolution in sight. Score 1 — Governance discussions underway but not resolved. Score 2 — A single agreed source of truth for core commercial KPIs, owned by a named individual.
7. Can your team evaluate an AI vendor's proposal without relying on the vendor to explain it?
This is the check that nobody wants to answer honestly.
If the only person who can explain why your AI investment will work is the person selling it, you have a problem. Not because vendors are dishonest. Because vendors optimise for winning the contract. The job of evaluating whether the contract makes commercial sense belongs to you.
At minimum, someone on your team — or someone you bring in — should be able to answer three questions independently: What data does this system need, and do we have it? What does the implementation timeline look like, and what are the real blockers? What does ROI look like at 90 days, not 18 months?
If the answer to any of those questions currently comes only from the vendor's slide deck, that is a risk worth addressing before signing anything.
Score 0 — No independent evaluation capability in the team. Score 1 — Some internal capability, but significant dependence on vendor guidance. Score 2 — Clear internal or external capability to evaluate proposals independently, with specific questions for each.
8. Is your data function more than one person deep?
A single data analyst or data scientist who holds all the institutional knowledge of your data architecture is one resignation away from a crisis.
We have seen this in CPG businesses of £200M+ revenue. One person who built the dashboards, knows where the data lives, and understands why the ERP extracts are inconsistent. When they leave — and they often leave, because their skills are in high demand — the organisation loses months of capability.
AI compounds this risk. A single-person data function cannot build, maintain, and improve AI models at the same time as keeping the lights on. It cannot document its own work reliably enough to hand anything over. And it certainly cannot drive the change management required to embed AI in the business.
If your data function is one person deep, your AI programme has a structural dependency that no technology investment will fix.
Score 0 — One person owns all data knowledge with no documentation. Score 1 — Two or more people in the team, but knowledge transfer is incomplete. Score 2 — Documented data architecture, knowledge shared across at least two people, succession considered.
9. Have you calculated what doing nothing is costing you?
Most CPG leaders frame AI as an investment with an uncertain return. Few frame inaction as a cost with a measurable impact.
Demand forecasting running at 68% accuracy costs money. Every percentage point of forecast error translates to stockouts, waste, or excess promotional spend. If your promotional trade spend is £30M and your spend allocation is 12–18% suboptimal — a figure we see consistently in CPG businesses at this revenue scale — that is £3.6M to £5.4M of recoverable margin sitting in the status quo.
The question is not whether AI delivers ROI. The question is whether the current manual process has a lower cost than the AI alternative. In almost every case we have examined, the answer is no.
Before you model the return on an AI investment, model the cost of not making it.
Score 0 — No quantification of current-state cost of manual processes. Score 1 — Some awareness of cost but not formally modelled or presented. Score 2 — Specific cost of inaction calculated, expressed as an annual P&L figure, and visible to budget decision-makers.
10. Do you have a change management plan for AI adoption?
Buying AI without a change management plan is like buying a treadmill and leaving it in the garage. The capability exists. The investment was made. Nobody is using it.
We have watched an AI tool go live in a CPG supply chain team with 8% adoption six months after launch. Not because the tool was poor. Because nobody changed how the team worked around it. The old spreadsheet was still there. The old process still felt safer. And no one in leadership was tracking whether adoption was happening.
AI adoption fails at the human layer. The questions that matter are: Who will use this, specifically? What does their current workflow look like, and how does this change it? Who will train them? What happens if they revert to the manual process? What does success look like at 30 days, 60 days, 90 days?
If you cannot answer these before launch, you will be answering them at a post-mortem.
Score 0 — No change management plan, adoption left to happen organically. Score 1 — Some communication plan in place but no structured workflow redesign or adoption tracking. Score 2 — Named change lead, workflow redesign documented, adoption milestones defined, escalation path agreed.
What Your Score Means
| Score | What It Signals |
| 17–20 | Strong foundations. You are ready to invest in AI with confidence. The risk of failure is low. |
| 12–16 | Mixed readiness. Two or three specific gaps need addressing before committing to a full investment. |
| 7–11 | Significant risk. Spending on AI now is likely to repeat the same failure patterns. Fix the fundamentals first. |
| 0–6 | High risk. An AI investment at this stage will almost certainly fail — not because of the technology, but because of what sits beneath it. |
What Comes Next
If your score landed below 14, you are not behind. You are at exactly the right point to make decisions that will save you from the £80,000 mistake we described at the start.
The AI FlightCheck™ is the structured version of this diagnostic. In two weeks, we audit your strategy alignment, data infrastructure, and execution readiness — and deliver a 15-page report that tells you, specifically, where your highest-ROI AI opportunities are and what needs to be in place before you pursue them.
Fixed price. Two weeks. No commitment beyond the diagnostic.
If your score was above 14 and you already know what you want to build, the AI FlightPath™ Sprint takes you from strategic alignment to a production-ready AI output in 10 weeks.
Either way, the next step is a 30-minute AI Navigation Call. No pitch. No slide deck. We will tell you honestly what we think your highest-leverage move is whether that involves AI Navi or not.\