The spec is in a PDF. The territory is in a rep's head.
Building products are bought against a specification and sold through manufacturers, reps, distributors and dealers. What a buyer's AI assistant needs is mostly written down, in ten places that don't all agree. This brief draws on Field Research 001 at D.W. Ross Company, a Cleveland building-products supplier.
CoversManufacturers, suppliers, distributors and dealers of building products and equipment
01 · What the industry knows
What a building-products business knows.
- Its role
- Whether it makes a product, represents it, distributes it, sells it or installs it.
- Products
- Families, materials and applications, and the brands behind them.
- Performance
- Thermal, structural and other values, each tested to a named standard.
- Documents
- Spec sheets, submittals, installation guides, CAD and BIM files.
- Listings and approvals
- NFRC, UL, ICC-ES, Florida Product Approval, Miami-Dade NOA, each with an issue date.
- Channel and territory
- Who sells it in each region, and who quotes it: the manufacturer, a rep, a distributor or a dealer.
Where those facts live today.
| Fact | Where it lives today |
|---|---|
| The company | Website, Google Business Profile, directories |
| Products | Product pages, catalogs, configurators, manufacturer sites |
| Specs | Spec sheets and test reports, often PDF |
| Documents | Download pages, spec platforms, shared drives |
| Listings and approvals | Each issuer's own directory |
| Where to buy | Locators, rep sites, distributor sites, sales knowledge |
| Who to contact | Contact forms, phone trees, a rep's cell phone |
The breaks we look for: the website shows last year's series; the current values are in a PDF with no date; a distributor's page still lists a discontinued model; a directory files a supply-only company under installers. A salesperson works around all of that by phone. An assistant answers from what it can read, says it doesn't know, or names a competitor whose information was easier to find.
03 · What is being decided
What the buyer's agent is trying to decide.
For a product to be among the three, and for the request to reach the company that quotes it, five things have to be true.
- Future scenario
“A thermally broken aluminum window for a school renovation, or who carries a brand of air barrier near a job in Akron.”
- Future scenario
“Find three commercial window systems that meet this spec, are available through a supplier serving Northeast Ohio, and get me what I need for a quote.”
| Stage | The assistant's job | What has to be true |
|---|---|---|
| Discover | Find suppliers of commercial windows in the region | AI knows the company exists, what it sells, and where it sells it |
| Understand | Read what each product is and how it performs | Product pages state families, materials, applications and tested values in text, with the documents linked |
| Qualify | Check the spec: thermal performance, structural class, listings | Each value carries its test standard and date, so it isn't guessed |
| Contact | Find who quotes Northeast Ohio | It's published who quotes that region, and how to reach them |
| Act | Send the project details for a quote | A structured request reaches a person with what they need. A request, not an order |
What we've seen, what we expect, what could come later.
- Observed
At D.W. Ross Company, what the business knows about its products and customers lives mostly in people's experience and on paper. The website, last rebuilt around 2002, carried a small part of it.
Source: Field Research 001, part 01. Lehvel fieldwork since September 2026.
- Observed
Before anything was built, the discovery found the business described differently across its own pages, its Google listing, directories and AI answers.
Source: Field Research 001, appendix.
- Hypothesis
People who use AI assistants for travel, errands and purchases will start using them at work, including for product research. That's our read of how useful software spreads, not a measured trend in construction.
- Hypothesis
When that happens, the company whose products are easier to find, understand and act on has the advantage. We can't put a number on it.
- Future scenario
A buyer describes what the project needs. An agent researches products and suppliers, compares them against the requirements, checks spec, location and timing, and sends a supplier a structured request. The buyer reviews what comes back.
05 · Current Lehvel work
The work behind this brief, with its status.
- Lehvel Labs · Field notesAI agents are coming for B2B product discoveryPlanned
- Field researchField Research 001: D.W. Ross CompanySince September 2026 · status on each system
Not built for this sector: a product record with dates and sources on every fact, read-only product, document and territory answers, or a quote request path.
Company truth and product truth, each on its own clock.
- Two records
- The company side (who sells what, where, and who quotes it) and the product side (what each product is and how it performs). Every fact keeps its source, who confirmed it, and when.
- Values in text, linked to proof
- Important values stated on the page, linked to the spec sheet, submittal or test report. When a value comes from the manufacturer or a listing body, the record says so.
- Dates on what expires
- A listing or approval has an issue date, a spec sheet a version, and lead times change weekly. Out of date is answered with “ask us,” not stated as current.
- Public, on request, internal
- Some manufacturer relationships shouldn't be published. Each fact carries one of three marks, and an assistant gets only what was approved.
- Markup follows facts
- Structured data repeats what the page says. If the page lacks the spec or the PDF is out of date, so is the markup.
Authority boundaryThe one action worth studying is a quote request that reaches a person with the project details. It never sets a price, promises stock, commits a delivery date or accepts a substitution.
For one businessBusiness Truth Discovery · $750