| Sector | Water & fluid treatment technology |
| Products | Pumps · filtration · membranes · valves |
| Grown by | Acquiring specialist brands |
| Estate | 28 plants · 11 brands · 8 countries |
| Reference duty | 1 brand · 3 sites · 6 machining lines |
| Run time | 16 months, pilot to rollout |
| Edge | Read-only · ~72 h buffer · on-premise |
The customer class is an international water and fluid-treatment technology group. It makes centrifugal and submersible pumps, filtration housings, membrane modules and process valves for municipal, agricultural, food-and-beverage and industrial buyers.
It did not grow plant by plant. It grew brand by brand, acquiring specialist manufacturers that each arrived with their own factories, their own machine controls, their own maintenance habits and their own spreadsheet for how well a line ran.
The heterogeneity is structural: the group bought it that way, and no directive will standardise it. Every earlier attempt began as an estate-wide programme and stalled at the first plant that counted differently.
Leaves changeovers out of planned time. Its largest loss never enters its own figure.
Measures speed against a nameplate rate set fifteen years ago, before the line was rebuilt.
Books short stops as running time, so its micro-losses disappear into availability.
The reconciliation tax. Every month, controllers rebuild the group view by hand from reports that disagree on what a lost hour is. The arguments about capital start from those reports, and the plant with the most generous definition wins them.
| Term | Plant A · Brand 1 | Plant B · Brand 2 | Plant C · Brand 3 | Group dictionary (one definition) |
|---|---|---|---|---|
| Planned time | Shift time minus changeovers and planned maintenance | Shift time minus planned maintenance | Shift time minus planned maintenance | Shift time minus agreed planned stops only |
| Speed reference | Design cycle per part | Nameplate rate from before the rebuild | Best cycle measured last year | Validated ideal cycle per part number |
| Short stops | Availability loss | Availability loss | Counted as running | Availability loss, logged with a reason |
| Good output | First-pass plus reworked parts | First-pass parts | First-pass plus reworked parts | First-pass parts only |
The three plant managers and group controlling agree one written definition of planned time, availability, speed and good output for the pilot line type. Every machine control is surveyed for what it exposes over OPC UA or MTConnect, and machines without a confirmed read path stay out.
Read-only edge gateways go in on six pump-casing machining and assembly lines, with roughly 72 hours of local buffering. Three brands of machine control map into one namespace, and the three sites are measured identically for ninety days.
Pulse builds per-machine baselines for cycle time and energy per unit. Orbis rules route the largest shared loss on each line to its named owner as a recommendation, with the approval kept on record.
The method scales across 28 plants and 11 brands in 8 countries, one brand at a time. The dictionary, the survey and the edge become the standard onboarding package, with any future acquisition reporting comparably within six weeks.
A group built by acquisition knows its plants are not comparable, and has probably been sold an estate-wide standardisation programme to fix it. This one is deliberately narrower: one brand, three sites, one line type, ninety days. If the numbers are comparable and the plant managers sign them, the group has a method. If not, it has found out for very little.
One acquired brand. Three of its sites. Two pump-casing lines at each, one machining and one assembly. Ninety days, measured to one dictionary.
The manifold. Each plant keeps its own brand of machine control. Nothing is replaced, and Pragmatos 3.0 has no write path into any of it. Read-only edge gateways feed one namespace, where a single metric dictionary turns three local dialects into one comparable figure. Orbis names the largest shared loss on each line as a rule-based recommendation, a named line owner approves or rejects it, and both are kept on record.
No write path into any machine control exists anywhere in the platform. A plant IT team in each of eight countries reviews one property instead of a feature list, and a network outage costs continuity, not data: the edge buffers roughly 72 hours locally.
The metric dictionary is the product here, more than the dashboard. Planned time, availability, speed and good output are defined once, in writing, and applied identically to every machine in the namespace, whatever brand of control it carries and whichever company built it.
Pragmatos 3.0 runs on the group's own servers. Each country's plant data stays on infrastructure the group already controls, which takes data residency out of the conversation before it starts.
Reported vs. comparable, ninety days on pump-casing lines. As reported, the three plants spread across 13 points and Plant A leads. Under one written definition the spread is 7 points and Plant B leads. Plant A had been leaving changeovers out of planned time, so its biggest loss never appeared in its own figure. That loss is now the first Orbis recommendation on its lines, and the 7-point gap is the capacity the group can actually recover.
| Before | With Pragmatos 3.0 (modelled) |
|---|---|
| Each plant reports OEE by the definition it inherited. | One written definition, applied identically at every site. |
| Controllers reconcile plant reports by hand, 40 hours per plant each month. | Figures arrive comparable. Reconciliation becomes review, 80% less time. |
| The most generous definition wins the capital argument. | Capital requests start from one shared loss breakdown. |
| The biggest loss hides inside a local definition. | The largest shared loss is named per line. +3 OEE points on the pilot line type. |
| Energy is a site bill, read once a month. | Energy per unit produced, per line and per shift. 5% less. |
| A newly acquired plant reports its own way for many months. | Comparable within six weeks of the interface survey. |
| 40 h / month | ASSUMED | Plant and group controller time spent reconciling one plant's performance report each month |
| −80% | ASSUMED | Share of that time removed when figures arrive in one definition |
| €80,640 | DERIVED | 3 plants × 40 h × 12 = 1,440 h/yr; 80% = 1,152 h × €70/h |
| +3 pts | ASSUMED | OEE gain on the pilot line type once the largest shared loss is visible and owned |
| €198,000 | DERIVED | 6 lines × 5,000 planned h = 30,000 h; 3% = 900 h × €220/h contribution |
| −5% | ASSUMED | Energy per unit saved once idle running and slow cycles are visible per line and shift |
| €36,000 | DERIVED | 3 sites × 1,200,000 kWh × €0.20/kWh = €720,000; 5% = €36,000 |
| €314,640 | DERIVED | €80,640 + €198,000 + €36,000 per year on the 3-site group, before any rollout |
| 6 wks | ASSUMED | Design target: survey, edge and dictionary applied to a newly acquired plant |
| 81/74/68 → 62/66/59 | ASSUMED | Outlet chart values: a typical pattern of how inherited definitions diverge |
Assumed figures are rates chosen for the model and can be changed. Derived figures are arithmetic on them and move with them.
A group that grows by acquisition buys a new set of definitions with every deal. A dashboard shows the problem once. The metric dictionary, the interface survey and the read-only edge fix it again for every plant the group buys next, so each future acquisition is comparable from week six rather than year two. That repeatable method is what a system built around one brand of control is not designed to give a multi-brand group.