Optimization, Not Prediction: €4.1M From Constraints Rather Than Models
An e-commerce fulfilment operator with three warehouses and 900 staff saw cost per order rise 22% in two years. The COO expected an AI demand-prediction system; Envion’s assessment found demand was already forecast within 8% — the losses were downstream, in how work was arranged. Slotting, batch-and-route and shift optimization delivered a €4.1M annualized benefit, and only one component used machine learning at all.
The challenge
Labour was the dominant cost and picking the dominant labour activity. Envion instrumented three weeks of picking activity — WMS pick logs, RF scanner timestamps and location data across 41,000 storage positions.
Travel dominated: pickers spent 63% of active time walking. Slotting was historical, not analytical — 340 product pairs co-occurred in over 15% of orders while stored an average of 84 metres apart, and seasonal movers stayed in prime positions eleven months after their season. Batching was naive — orders released in arrival sequence, no grouping by zone or route. Shift design fought the volume curve — two fixed shifts against a demand curve peaking sharply twice a day, with 19% idle capacity in one window and mandatory overtime in another, on the same day, in the same building.
One data quality finding changed the scope: pick times for 6% of locations were implausibly fast — faster than the physical walk. Pickers were pre-scanning items from a staging cart to hit rate targets, distorting every productivity metric management used. Envion reported it as a measurement finding, not a disciplinary one.
Decision path
The right problem statement was: minimize total labour cost subject to order SLAs, physical layout, staff availability rules and union shift constraints. That is a constrained optimization problem — and mostly not a machine learning one.
Envion prototyped three components, each measured in simulation against three weeks of real order history: slotting optimization — a mixed-integer formulation placing products by velocity and pairwise affinity, subject to weight, zone, hazmat and shelf constraints, simulating a 31% travel reduction; batch and route optimization — grouping orders into pick batches with routed sequences subject to cart capacity and SLA cut-offs, adding 14% less travel per order line; and shift optimization — staffing shaped to the intraday curve within union rules and staff preference constraints, simulating a 44% overtime reduction. Only one component used ML at all: a small model predicting pick duration by location and item characteristics, feeding the batching objective.
Envion contribution
Envion also modelled the re-slotting transition cost — moving 41,000 positions is not free — and recommended a phased plan targeting the top 12% of positions, capturing 71% of the available gain at a fraction of the disruption.
The production requirements were defined explicitly: live WMS integration for slotting recommendations (nightly batch — Envion advised against real-time as unnecessary complexity), re-slotting as a continuous background process for idle labour rather than a shutdown event, a supervisor override path with logging since floor staff routinely know things the model doesn't, correction of the pre-scanning measurement issue before go-live — or the optimizer would train on fiction — and an estimated 4-month build with an explicit warning that gains decay within ~9 months without continuous re-slotting.
Delivery
The eight-week engagement covered data assessment, operational analysis, optimization prototyping and production requirements.
Practical rules from the project: ask whether your problem is "what will happen" or "what should we do" — if your forecast is already adequate and you are still losing money, the loss is in the decision layer, not the prediction layer; interrogate the arrangements nobody chose — slotting, shift patterns, routing sequences usually accreted rather than being designed; validate operational data before optimizing on it — any metric people are measured against will be gamed somewhere, and an optimizer trained on gamed data confidently optimizes the gaming; and model the transition cost — a solution 31% better and impossible to migrate to is worth nothing.
Outcome and evidence
Twelve months post-implementation: travel's share of picking time fell from 63% to 41%, lines picked per hour rose from 78 to 121, overtime cost fell from €2.9M to €1.4M a year, cost per order fell 27%, peak-season agency labour fell from 210 to 95 staff, order SLA compliance rose from 91% to 98% — a €4.1M annualized benefit.
No permanent staff were released; the reduction fell on agency and overtime spend — which the client reported made floor adoption dramatically easier, since staff weren't being asked to optimize themselves out of a job.
Client feedback
What the client says about this engagement
“I came in wanting AI and left with a linear program and a nightly job, and €4.1 million. Envion's assessment was that our forecasting was already fine and our problem was that we'd never questioned where things were stored. Nobody had looked at that in nine years — it was just how the warehouse was.
The finding I'd have paid for on its own is the pre-scanning. We'd been managing to productivity numbers that were partly fictional, and setting targets from them. They found it in the timestamp data in the first fortnight and told us straight.”
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 your forecast is already adequate and costs keep rising, discuss the decision layer with Envion — routing, slotting, scheduling and staffing are optimization problems, and they are cheaper than you think.
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.