Productivity recovered across multiple APUs at a large-scale European API plant
How a leading European Active Pharmaceutical Ingredient provider replaced annual sales-projection plans with event-driven, AI-generated schedules across its autonomous production units — measurable productivity gains within 8–12 weeks of go-live.
To productivity impact
Weeks to productivity impact
Time from kickoff to documented gains in capacity utilisation and batch cycle time across the plant. Based on Neewee deployment observations in regulated process manufacturing.
Capacity utilisation across APUs
Directional indicator. Capacity recovery across the autonomous production units moved into the signed-off +10–15% range observed across deployments. Specific figure withheld at customer request.
Batch cycle time
Directional indicator. Hidden buffer time between batches compressed once schedules began rebuilding on live shop-floor signals rather than aging assumptions in the annual plan.
WIP and buffer stock
Directional indicator. Inventory tied up in work-in-progress and safety buffers decreased as forecast-driven recommendations replaced precautionary build-ahead. Within the signed-off −10–20% WIP range observed across deployments.
The objective
Replace a once-a-year sales-projection plan with an event-driven scheduling layer that surfaces unused capacity across the plant's autonomous production units and re-plans in minutes when the floor moves.
The challenge
Where the previous approach fell short
01
Annual plan rebuilt by hand
The plan was cut once a year from sales projections and re-cut manually for every demand shift or batch slip. By the time the rebuild had been validated, the plant had moved on.
02
APUs running semi-independently
Each autonomous production unit held its own constraints and bottlenecks. Capacity wasn't optimised across them, and the master plan couldn't see what one APU could do for another this week.
03
Resource pools sitting idle
Resources one APU couldn't reach in a given week sat unused while another absorbed the shortfall through safety buffers, overtime, and precautionary build-ahead.
04
Annual plan, not a live model
The operating document was a snapshot rather than a live model of the plant. Long-term commitments drifted out of sync with the events surfacing daily from the shop floor.
The full constraint universe — the operational dimensions out-of-the-box schedulers can’t hold
Annual plan re-cut by hand
The plan was cut once a year from sales projections and re-cut manually for every demand shift or batch slip — slow and error-prone, and out of date by the time it was validated.
APUs running semi-independently
Each autonomous production unit held its own constraints, schedules, and bottlenecks, and the master plan couldn't see what one APU could do for another this week.
Capacity not optimised across APUs
Capacity wasn't optimised across the autonomous production units, so the plant couldn't act on slack that sat in one APU while another ran short.
Resource pools sitting idle
Resources one APU couldn't reach in a given week sat unused while another absorbed the shortfall through safety buffers, overtime, and precautionary build-ahead.
A snapshot, not a live model
The annual plan was a snapshot rather than a live model of the plant, so long-term commitments drifted out of sync with the events surfacing daily from the shop floor.
No live signals from the floor
The plan waited for the next manual rebuild rather than taking live signals from across all APUs as asset, batch, and quality events arrived.
The plant ran multiple autonomous production units (APUs) — each with its own buildings, equipment lines, and final-processing areas serving different products — on a single annual production plan. The plan was built from sales projections at the start of the year and re-cut by hand whenever a demand shift, an asset issue, or a batch slip forced the change. The hand-cut rebuilds were slow and error-prone; by the time the new plan had been validated, it was already out of date.
The deeper problem was that the annual plan was a snapshot, not a live model of the plant. Long-term commitments couldn't be kept aligned with what was actually happening on the shop floor. Each APU ran semi-independently, with its own constraints, schedules, and bottlenecks — and capacity wasn't being optimised across them. Resource pools that one APU couldn't reach this week sat idle while another absorbed the shortfall through buffers and overtime.
The team needed a scheduling layer that could:
- Take live signals from across all APUs, not wait for the next manual rebuild
- Hold long-term commitments and short-term events in one model
- Surface unused capacity in the existing resource pool, rather than absorb every shift through inventory and overtime
What Bodhee did
How Bodhee rebuilt the plan
Bodhee Production Scheduling deployed as an event-driven layer on top of the existing annual planning process — not a replacement. Alongside the plant's ERP environment, Bodhee took live signals from across the autonomous production units and produced schedules that respected every constraint encoded in the long-term plan and every event surfacing from the shop floor.
The integration was thin and direct. Standard Excel templates consolidated historical and current production data in the first phase. ERP connectors then kept master data, order books, and crew rotations flowing once Bodhee was live. Live shop-floor signals — asset downtime, resource availability, batch progress, quality holds — fed Bodhee through the plant's existing data layer. Nothing was rebuilt; existing systems kept their existing roles.
The scheduling engine ran multi-objective optimisation against the plant's actual constraints. For each rebuild, it generated several plan candidates weighted differently against capacity utilisation, cycle time, and on-time commitments. Planners reviewed the candidates in a control-tower view — a live Gantt across every APU, an event stream of shop-floor signals, and alerts when a constraint was at risk — and published the version that best fit the week's priorities.
Rollout was phased over twelve weeks. Weeks zero to two were data harmonisation through standard templates and ERP integration. Weeks two to five were constraint modelling — encoding the goals, hard rules, and influencing parameters into the multi-objective scheduler. Weeks five to eight brought the control tower live for planning, supply chain, and operations. Weeks eight to twelve produced the first measurable gains in capacity utilisation and batch cycle time.
Within four weeks of go-live, schedule rebuilds against live signals were running in minutes rather than days. Within twelve, the annual sales-projection plan had stopped being the plant's operating document. The control tower had taken its place.
01
Layer scheduling on top of planning
Bodhee Production Scheduling deployed as an event-driven layer beside the existing annual plan — not a replacement. Long-term commitments and short-term events held in one model.
02
Connect ERP and shop-floor signals
Standard templates consolidated historical data; ERP connectors fed master data and order books; the plant's existing data layer streamed live signals on assets, batches, and quality holds.
03
Multi-objective optimisation
Each rebuild generated several plan candidates weighted against capacity utilisation, cycle time, and on-time commitments. Planners chose the version that best fit the week's priorities.
04
Control tower for planners and ops
A live Gantt across every APU, an event stream of shop-floor signals, and constraint-risk alerts — shared by planning, supply chain, and operations as one operating view.
The engagement
From discovery to cutover
Weeks 0–2
Data harmonisation
Standard templates consolidated historical and current production data; ERP integration mapped master data and order books.
Weeks 2–5
Constraint modelling
Goals, hard rules, and influencing parameters encoded into the multi-objective scheduler — surfaced from the people who held them in their heads.
Weeks 5–8
Control tower live
Planning, supply chain, and operations moved to the live Gantt and event stream as the shared operating view of the plant.
Weeks 8–12
First measurable gains
Capacity utilisation and batch cycle time moved into the signed-off ranges. The annual plan stopped being the plant's operating document.
The outcome
What changed on the ground
Weeks to productivity impact
Event-driven across every APU
When a demand shift or shop-floor signal arrives, the scheduler generates revised plan candidates within minutes; the planner publishes the best fit and the control tower propagates the change to supply chain and operations. Capacity that previously sat unused is now recovered.
Cycle time and capital freed
Hidden buffer time between batches compressed as schedules rebuilt on live signals rather than aging assumptions. WIP and safety-buffer stock dropped as forecast-driven recommendations replaced precautionary build-ahead — freeing working capital previously tied up in inventory.
The planner schedules forward
The planner now spends most of the week scheduling forward — running scenarios for upcoming demand, validating engine proposals, and resolving constraint exceptions — rather than re-cutting the annual plan to absorb whatever the floor surfaced today.
Twelve weeks ago, our operating view of the plant was last week's spreadsheet. Now it's a Gantt that updates when the floor moves, and a plan that rebuilds in minutes when the demand changes. The capacity was always there. We just couldn't see it.
Why this matters beyond one site
The pattern isn't unique to API manufacturing. Any process plant organised around semi-autonomous production units — multiple lines, blocks, or buildings each with their own constraint sets — is paying a coordination tax whenever the master plan is built once a year and re-cut by hand. The deployment shape (event-driven scheduling layered on top of existing ERP, multi-objective optimisation against the plant's actual constraints, a control tower as the operating view) generalises to specialty chemicals, biologics, and other multi-block process operations with similar plant topologies.
What doesn't generalise is the constraint set. Every plant encodes its own combination of equipment dedication, cleaning rules, crew skills, and quality gates. The Bodhee work is in surfacing those constraints from the people who already hold them in their heads — and encoding them once, alongside the existing systems.