Building an attribute schema for Pumps & Fluid Power that buyers and AI can actually use
What attribute schema pumps and fluid power distributors need so filtered search, GTIN feeds, and AI answer engines can actually surface a SKU.
A pump listing that says "centrifugal pump, cast iron, 5 HP" reads fine to a human skimming a page. It is close to useless to a filtered search facet or an AI answer engine trying to match a duty point. In Pumps & Fluid Power, the attributes that matter are the ones tied to an engineering spec sheet, not the ones that read well in a paragraph. Get the schema right and a SKU shows up in every relevant filter and every relevant AI query. Get it wrong and the part is technically in the catalog and functionally invisible.
Why "pump" is not an attribute
Most supplier feeds for pumps arrive as a title, a category, a price, and a block of marketing prose. That works for a homepage banner. It does not work for a buyer or a bot trying to answer "what pump handles 250 GPM at 120 ft of head in 316 stainless." Filtered search on a distributor site runs on discrete, comparable fields: flow rate, head, materials, connection size, seal type. If those fields are blank or buried in a PDF, the facet has nothing to filter on, and the product drops out of every query that uses that facet — even though the pump would have been a correct match.
The same failure shows up one layer up, at the feed level. Google's own product data specification is explicit that incomplete or inconsistent attribute values (wrong category, missing variant attributes, conflicting data between feed and site) cause disapprovals or limited eligibility — not just a slightly worse ranking. Missing attributes are not a cosmetic gap. They are a removal mechanism.
The attribute set that actually matters
For Pumps & Fluid Power, the attributes that drive both filtered search and AI matching map closely to what a mechanical engineer would pull off a real datasheet, not a marketing description:
| Attribute | Example value | Why it matters |
|---|---|---|
| Pump type / configuration | End suction, horizontal, single stage, centerline discharge | Determines mounting and application fit |
| Flow rate (duty point) | 250 GPM (57 m³/h) | Primary filter facet; buyers search by flow first |
| Total dynamic head | 120 ft (36.6 m) | Paired with flow rate to define the duty point |
| NPSHr | 8 ft | Prevents cavitation misapplication |
| Casing / wetted materials | 316 stainless steel | Chemical compatibility filter |
| Impeller type | Closed, semi-open | Solids handling / efficiency filter |
| Seal type | Mechanical seal, single, John Crane-style | Maintenance and fluid compatibility |
| Suction / discharge size | 3 in x 2 in, ANSI 150 flange | Pipe fit-up, non-negotiable filter |
| Motor power | 10 HP (7.5 kW) | Electrical sizing |
| Speed | 3,550 RPM | Duty and NPSH relationship |
| Efficiency at duty point | 78% | Energy cost comparison |
| Max operating temp / pressure | 250°F / 150 psi | Application safety limit |
| Standard / certification | ASME B73.1, API 610 | Dimensional interchangeability, spec compliance |
Note that this list is close to the ASME B73.1 dimensional-interchangeability standard for horizontal end-suction pumps — that standard exists precisely so a pump from one manufacturer can be swapped for another at the same duty point. If your attribute schema doesn't capture the fields the standard governs, you can't make that swap claim searchable, even when it's true.
Before / after: an end-suction centrifugal pump
Here's a typical raw supplier feed description for a mid-size end-suction centrifugal pump:
Raw feed description: "Heavy-duty centrifugal pump for industrial applications. Reliable performance, durable construction, easy maintenance. 5HP motor. Cast iron."
That's a real string pulled from a real class of supplier feed — and it fails almost every filter on a distributor site. No flow rate, no head, no connection size, no seal type, no NPSHr. A buyer filtering for "250 GPM, 316SS, mechanical seal" never sees it, even if the physical pump qualifies.
Enriched, the same SKU looks like this:
| Attribute | Value |
|---|---|
| Type | End suction, horizontal, single stage, centerline discharge |
| Flow rate | 250 GPM at duty point |
| Total head | 120 ft |
| NPSHr | 8 ft |
| Casing material | Cast iron (316SS option) |
| Impeller | Closed, bronze |
| Seal | Mechanical seal, single |
| Suction x Discharge | 3 in x 2 in, ANSI 150 |
| Motor | 5 HP, 3,550 RPM, TEFC |
| Max temp / pressure | 250°F / 150 psi |
| Standard | ASME B73.1 dimensional class |
Same physical pump. One version is invisible to a facet search. The other is matchable on eleven independent filters, and it's the version that shows up when someone asks an answer engine "recommend an ASME B73.1 end-suction pump rated for 250 GPM at 120 ft head in cast iron with a mechanical seal." That phrasing — flow, head, material, seal type, standard — is exactly how a distributor's own sales engineers already talk. Structured data just makes it legible to a machine.
Structuring it so it holds up
The fix isn't a bigger free-text field. It's a schema where every attribute above is its own discrete field with a controlled unit (GPM vs m³/h, ft vs m, both if you serve mixed markets), pulled from the actual supplier datasheet or spec PDF rather than typed from memory. Values need a source and a confidence score, because a flow rate guessed from a title is a liability the moment a buyer specs against it.
This is the part that's tedious at scale and easy to get wrong manually — mapping a hundred-line PDF spec sheet into eleven clean fields per SKU, repeated across thousands of pumps, valves, actuators, and fittings, typically runs 30-45 minutes of manual work per SKU. Anglera plugs into whatever PIM a distributor already runs — Akeneo, Salsify, inriver, or none at all — and does that extraction and gap-filling continuously, scoring each attribute against the source document rather than inventing a number. Your PIM still stores the data; Anglera does the work of making sure every pump has the eleven fields a buyer, a facet, and an AI answer engine all need to find it.