A programmatic boiler site is only as good as the spreadsheet under it. Dataset building is the work of turning scattered supplier files, spec PDFs and job tickets into one structured source where every boiler model, part and city has a clean row and verified fields.
This is marketing data work. We are not a heating contractor and we don't sell parts; we organise information you already hold so it can be published accurately.
Why the dataset matters more than the design
The data is the only part of a page set a competitor cannot copy overnight. Anyone can buy a template; nobody else holds your compatibility notes or your job counts by town.
It is also what keeps a large site on the right side of Google's rules. A page that shows fields specific to one model or one part has a reason to exist. A page whose only unique field is its title doesn't, and no amount of copywriting fixes that. Our page data threshold guide sets out the field counts we use as a pass mark.
Where boiler data usually lives
Most of what we need is already in the building, just not in one place.
- Supplier and distributor price files. Part numbers, descriptions, sometimes compatibility.
- Manufacturer installation and service manuals. Output ranges, venting limits, clearances, fault-code tables.
- Warranty and registration records. Which models you install most, by year.
- Job tickets or field-service software. Job type, equipment serviced and municipality.
- Technician knowledge. The faults that recur on certain models, usually written down nowhere. We interview two or three senior technicians and record it.
What's included
- Source inventory. A list of every file and system we draw from, with its owner and how current it is.
- Schema. The columns for each entity: model, part, fault code, city, and the links between them.
- Normalisation. One spelling per brand and model. We often find the same boiler listed several ways across supplier files.
- Compatibility links. Which parts fit which models, with the source for each link recorded.
- Gap report. Rows missing key fields, ranked by how much search demand they likely carry, so you know where an hour of work pays back.
- Refresh process. How new supplier files get merged in, and who checks them.
Timeline
| Stage | Typical length |
|---|---|
| Source inventory and schema | 1–2 weeks |
| Cleaning and normalising | 2–4 weeks |
| Compatibility links and technician interviews | 1–2 weeks |
| Gap report and handover | 1 week |
A worked example of scale: a parts seller with 3,000 SKUs where 40% already have compatibility data might end up with 1,200 rows ready to publish in the first batch, and a ranked list of the remaining 1,800 to fill over time. That is a far better launch than 3,000 pages where most are empty.
How results are measured
We measure the dataset by what it lets you publish, and how accurate it stays.
- Rows passing the publish threshold, by entity type
- Field completeness for the fields that matter most, such as compatibility and output range
- Error rate found in spot checks against manufacturer documents
- Freshness: the date each row was last checked
Once pages are live, the indexing and crawl control reports link page performance back to row quality, so you can see whether better-filled rows earn more visits.
Mistakes to avoid
- Trusting a single supplier file. Distributors make transcription errors, and those errors then appear on hundreds of pages.
- Mixing customer data into the publishing set. Keep names, addresses and phone numbers out. Only aggregate job facts belong in the dataset.
- No source column. If you can't say where a compatibility claim came from, you can't defend or fix it.
- Building once. Supersessions, discontinued models and new releases change the data every quarter. Without a refresh process the site decays.
- Inventing specs to fill a template. Leave a field blank and hold the page back; never guess an output rating.
Once the dataset is ready, programmatic boiler page systems turn it into pages. The dataset discipline behind BoilerPart.ca is described on our results page.
Frequently asked questions
Who owns the dataset?
You do. It is delivered as files you keep, in a format any developer can read, and it stays yours whether or not you continue working with us.
Can you scrape competitor sites to fill gaps?
No. Copying other sites' content is one of the patterns Google's spam policies name, and it rarely produces accurate data anyway. Gaps are filled from manufacturer documents, your suppliers and your own records.
Do you need customer names from our job records?
No. We only need the job type, the equipment and a coarse location such as the municipality or the first three characters of the postal code. Personal details should stay in your own systems.
How long does a dataset take?
A clean parts or model library of a few thousand rows usually takes four to eight weeks, depending on how scattered the source files are.