When AI Is the Wrong Answer: The Feasibility Review That Stopped a €280k Build
A European logistics operator had already signed off budget for a machine-learning demand forecasting engine. Envion ran a two-week feasibility and data-readiness review before development started. Backtesting showed the client had 14 months of usable history against a 36-month requirement — and that a statistical baseline outperformed the proposed model. We recommended not building AI. The client kept the outcome and released the budget.
The challenge
Warehouse over-stocking and stock-outs were both rising. Planners were overriding the ERP's forecast manually in a spreadsheet, and leadership concluded — reasonably — that this was an AI problem. A vendor proposal for a custom ML demand forecasting engine had been costed at approximately €280,000 over nine months, and the board had approved it in principle.
Envion was engaged for what the CTO described as a formality: confirm the approach and help write the delivery plan.
Decision path
We ran four checks in sequence. Any one of them failing is a reason to pause; three failed.
Data sufficiency: machine-learning demand forecasting typically needs multiple full seasonal cycles to learn seasonality rather than memorise noise. The client had 14 months of clean, post-migration transactional history — everything before the 2024 ERP migration used a different SKU taxonomy and could not be reconciled without a mapping project of its own.
Signal stability: roughly 38% of SKU-level demand was promotion-driven, and promotions were planned in a system not connected to the forecasting data. A model trained on outcomes without the cause would learn the wrong pattern — confidently.
Baseline comparison: this is the check most feasibility reviews skip. We built a deliberately unglamorous baseline — Croston's method for intermittent demand, a moving average for fast-movers, and eleven explicit business rules for promo weeks and known seasonal events — and backtested it against the last four quarters. The vendor's proposed approach, simulated on the same 14 months, did not beat it.
Cost to serve, not cost to build: the €280k figure excluded retraining, drift monitoring, an MLOps owner and a fallback path. Modelled over three years, total cost of ownership was roughly 2.4× the build quote. The statistical baseline had effectively no recurring cost beyond ordinary maintenance.
Envion contribution
Envion delivered a written feasibility verdict — a "no, not yet", with the specific, testable conditions under which the answer becomes yes; the backtested statistical baseline, handed over as production-ready logic the client's own team could maintain; a promotion-data integration as the first item on the revised roadmap — because that is the real blocker, and it is a plumbing problem, not an AI problem; and an 18-month sequence with a re-evaluation gate: revisit ML forecasting once 30+ months of reconciled history exists and promotion data is joined at source.
Delivery
The engagement ran for two weeks with one AI strategy lead, one data engineer and one domain analyst, and produced the feasibility verdict, the backtest report, a cost-to-serve model and the revised 18-month sequence.
Practical rules from the project: insist on a baseline before a model — if nobody has built the boring version, the business case for the clever version is unproven; count your seasonal cycles, not your rows — millions of rows covering one season is one observation of that season; check whether the cause is in the data — if the thing that drives your outcome lives in another system, no model can learn it; cost the second and third year — build cost is the smallest number in an AI programme; and write the "not yet" conditions down — a deferral with criteria is a decision, a deferral without them is a delay that gets re-litigated every quarter.
Outcome and evidence
The build was stopped before kick-off and the budget re-allocated. The backtested baseline improved forecast error versus the incumbent ERP output (client-validated: mean absolute percentage error improved from 31% to 22%), the improvement arrived in 2 weeks instead of the planned 9 months, the review cost came in under 4% of the avoided spend, and one AI use case was deferred — not cancelled — with defined re-entry criteria. Modelled capital avoidance: ~€280k build, ~€672k three-year TCO.
The figures above are reported with their evidence basis (modelled, backtested, client-validated) rather than as bare claims — that is how they survive procurement scrutiny.
Evidence gate. This page publishes only what Envion's project records and client disclosure permissions support. Outcomes are added once verified against a baseline, a measurement period, and an approved source.
FAQ
Questions about this case
Facing a similar challenge?
If budget is already approved for an AI build and you want a defensible go / no-go before kick-off, discuss a feasibility review with Envion — two weeks is enough to know.
Discuss a Similar ChallengeKeep exploring
Similar case studies
Executive Technology Leadership
Support for high-stakes product and AI decisions
Bring senior technology leadership into the business when the roadmap is unclear, delivery is at risk, an AI initiative needs stronger ownership, or the company needs an experienced technical voice before hiring a permanent CTO.
Discuss Interim CTO SupportCore responsibilities
- Align product and technology priorities with business goals and measurable outcomes.
- Review architecture, delivery risks, data foundations, security needs, and AI readiness.
- Lead internal teams and external partners through a practical execution plan.
- Clarify team structure, ownership, decision rights, and delivery cadence.
- Support investor, board, partner, and due-diligence conversations with credible technical judgment.
New experience
Prompt-to-Page — try it right here
Describe the landing page you want, in your own words. We turn it into a finished page and email you a private link in 5–10 minutes — no briefs, no calls, $0 to see the result.
- Describe what you want to create.
- We structure, write, and compose the page.
- You receive a private link when it is ready.
Start with a sentence — the interactive builder takes it from there.
Generate My PageSafe, respectful content only. No obligation.