Unifying inventory visibility across a multi-supplier distribution network

UK Valves Direct supplies industrial flow-control equipment into water treatment, HVAC, oil and gas, chemical processing, food and beverage, pharmaceutical, marine, power generation and mining, dispatching next-working-day on stocked items ordered before the afternoon cut-off from its West Midlands depot.

Like every distributor operating this model, it carries far more in its catalogue than on its shelves. Around 780 SKUs are depot-held. The rest sit with manufacturers and stockists across the supply network, to be drop-shipped or transferred in on demand. That is precisely what a customer is buying: one relationship instead of fourteen, and access to a range no single warehouse could hold.

The engineering consequence is that a single availability figure has to be assembled from fourteen sources that report stock in fourteen different ways. Two order paths depended on it. Instant checkout resolved availability to a single in-stock flag. The quote desk handled bulk orders, project bills of materials, competitor part-number cross-references and custom actuation packages, against a published commitment of two to four working hours, with technical sales confirming availability supplier by supplier.

Both paths worked. Neither could express the one thing the underlying data actually varied on, which is how firmly a given figure was known.

Of roughly 1,850 saleable SKUs, around 780 are depot-held. The remainder are supplier-held across fourteen trading relationships. Three of those suppliers offer a machine-readable feed: two an EDI 846 Inventory Inquiry/Advice, one a REST API. Four can produce a scheduled CSV drop to FTP, refreshed nightly. The remaining seven, which between them hold most of the specialist and large-bore lines, publish nothing machine-readable at all. Availability comes by phone, by email, or from a PDF price list that carries prices rather than quantities.

Availability confirmation accounted for a substantial share of elapsed time on standard quotes. On complex quotes it was dwarfed by specification clarification: establishing whether the customer needs PN16 or Class 150, which flange table they are mating to (BS EN 1092-1, ASME B16.5, or a legacy BS10 Table D/E/F that tends not to come up until the valve arrives), whether EPDM or Viton suits the media, whether "stainless" means 316 bar or a CF8M cast body, and what a competitor part number actually cross-references to.

That work is the service. It is the reason customers route a project bill of materials through a technical desk instead of a search box, and no amount of stock automation touches it. Measuring first is what stopped us scoping a tool against the smaller half of the problem and committing to targets the architecture could not reach.

A large share of orders here carry documentation requirements: WRAS approval for potable water, ATEX for equipment in explosive atmospheres, PED and PER pressure-equipment conformity, API 607 or 6FA fire-safe certification on soft-seated quarter-turn valves, and above all EN 10204 3.1 material certificates.

A 3.1 certificate is tied to a specific material heat and batch number. An order carrying that requirement cannot be filled from whichever source shows stock; it has to come from material with matching traceable paperwork, which on most lines is a small subset of what physically exists. Any available-to-promise calculation that treats units as fungible will therefore overstate what can be supplied against the documentation. Carrying certification as a first-class sourcing constraint, rather than checking it at pick time, is what makes the promise reliable.

Each figure is stamped with its source tier, its age and a confidence state. Verified means a depot count or a live feed under two hours old. Indicative means a feed or CSV under twenty-four hours old, with a safety-stock buffer applied and a despatch window shown. Lead-time only means no availability data exists for that line, so the page states a realistic despatch window instead of a quantity.

SKU cross-reference tables handle part-number and unit-of-measure normalisation between supplier schemes and internal SKUs. Certification requirements ride along as sourcing constraints, so an order flagged for 3.1 documentation only ever draws on stock that can produce it.

The interesting engineering is not in consuming an API. It is in an availability model that degrades gracefully: one that can hold "I know this precisely", "I know this approximately as of last night" and "this one is lead-time quoted" in the same schema, and surface each appropriately to both a checkout page and a sales engineer.

The approach has good precedent in the sector. All Valves Online run a public green, amber and red indicator with an explicit statement that a red item is available from suppliers in roughly seven to ten days, and openly note that a green item may sell during the working day.

The actuator is sized against the valve's break-away, running and seating torque under actual differential pressure, seat material and temperature, with a safety factor applied: commonly around 1.3 to 1.5 times for ball valves and roughly double for butterfly valves, where dynamic torque is significant. The result has to stay within the valve's maximum allowable stem torque. Spring-return units are checked at minimum available air supply rather than nominal. Valve and actuator have to share a compatible ISO 5211 mounting interface, or the assembly needs a bracket and coupling to bridge it. The finished unit is then dry-bench function tested before it ships.

That is a manufacturing process with its own queue, and the routing now recognises it as one. Assemblies are identified at the point of order from the SKU's build flag and sent to the build bench with a lead time that matches the work, rather than inheriting the despatch path of a boxed item.

Every order line now classifies on its true stock source, its build flag and any documentation requirement travelling with it, then routes accordingly: depot pick, supplier drop-ship, inter-supplier transfer, build bench, or indent. A depot pick and a duplex indent are different operations, and the customer now sees a date derived from the one that applies.

The quote desk gained the same data. Live availability, lead time, cost and current volume-break pricing surface inside the quotation workflow, keyed to the customer's application and quantity tiers. For the seven suppliers without feeds the tool generates a pre-filled availability request and logs the response back against the SKU, so a check made once for a quote improves the answer for everyone who looks at that product next. The manual work that remains is captured rather than repeated.

Left deliberately untouched: the technical conversation. Specification matching, actuator sizing and cross-referencing stay with the sales engineers. The objective was to give them their week back, not to automate the expertise customers come to them for.

Two results are worth reading closely, because in both the honest number is the less impressive one.

Manual supplier checks fell by roughly three quarters rather than to near zero. Seven suppliers still publish nothing machine-readable, and the thirty-odd checks a week that remain are irreducible until that changes. We reported the ceiling in the final report rather than projecting a number the architecture could not reach.

Complex quote turnaround improved less, proportionally, than standard. That is specification work reasserting itself, exactly as the instrumentation predicted, and it is the shape you should expect when the dominant cost is engineering judgement rather than data retrieval. Setting the targets after the measurement rather than before it is why both landed.

What carried the engagement was not the integration work. It was deciding what the business could evidence, and then building something whose promises are bounded by that.