Your departments are fine. Your data isn't.

Every retail department — sales, service, parts, accessories — is running a competent playbook on top of records that disagree with each other. Parts buys catalog data from one place, service builds menus from another, accessories prices off a spreadsheet that left the building two managers ago. Marketing is asked to make all of it perform.

It will not, and the reason is boring. Nothing joins.

The join problem, stated plainly

A parts page, a service menu, a merchant feed, a Google Business Profile and a CRM opportunity record all need to agree about the same three things: what the item is, what it costs, and which store has it. When they disagree, four things break at once and nobody can see the break from inside their own dashboard.

• The catalog disagrees with itself. One part exists as three SKUs, two descriptions and a discontinued alias. Search engines index all three. The counter staff sell the fourth.
• The location record lies. Old hours, a marketing agency's phone number, a service department that closed. Google has been told this four times by four systems.
• The price is contextual. Web says one thing, the feed says another, the menu PDF says a third. Every discrepancy is a trust cost paid at the counter.
• The source code is garbage. CRM source fields that read website, walk-in, unknown, or the vendor's name instead of the channel. Attribution becomes fiction, and budgets get set from the fiction.

None of that is a marketing problem. All of it is a marketing outcome.

What clean actually means

Five tests, in order of cost to fix:

1. One identifier per thing. The OEM part number is the key. Your internal SKU is an alias. Fitment follows ACES and PIES conventions rather than a homegrown matrix nobody else can read.
2. One taxonomy. Every department maps to the same category tree, even if each keeps its own view on top. Accessories hanging off the parts tree is how attach-rate reporting dies.
3. One location record. Name, address, phone, hours and department phone numbers defined once, published everywhere, owned by one person.
4. One price truth. A single source that both the site and the feed read. If a human edits two places, one of them is already wrong.
5. One owner. Named, not a committee. Dirty data is what happens when five people assume someone else checks it.

The backend decides what the front end can promise

If the DMS sync runs nightly, do not publish real-time stock. If fitment data is partial, do not build a year-make-model selector that will confidently show nothing. If service pricing comes from a legacy system without line-item codes, do not launch an online scheduler that quotes a range.

The front end can only be as honest as the record underneath it. Stores that get this right stop treating their website as a brochure and start treating it as a read-only view of a system that already has to be correct for the counter, the call center and the parts van.

Metadata is the part other departments never see

This is where clean records become visible work. Everything below is downstream of a database someone else owns:

• Structured data on the templates that matter. Product and Offer on parts and accessory pages, Service and LocalBusiness on department pages, Vehicle listing markup on inventory detail pages. Google documents each type precisely; guessing at properties produces markup that validates as syntax and lies as content.
• Canonicals and a real URL count. Ten years of platform migrations leave thousands of legacy URLs serving near-duplicate pages. Consolidation is not cosmetic; it is the difference between one page competing and nine arguing.
• A merchant feed that matches the site. Feed disapprovals are usually a data-ownership failure, not a Google problem. Missing GTIN, mismatched availability, sale price without an effective date: every one of those is a backend record someone stopped maintaining.
• Hours and department pages that are current. An AI assistant reading your profile will repeat whatever it finds. Wrong hours are not a small error; they are the answer.

What each department gets when it is clean

Parts. Fitment that resolves means the feed qualifies and the answer engine can cite an actual fit. This is the entire ballgame for catalog visibility.

Service. Coded line items mean a menu that is quotable, a scheduler that is honest, and a page that can be structured.

Accessories. A shared taxonomy means attach rate is measurable by VIN rather than guessed at by invoice.

F&I and sales. Vehicle listing markup and consistent offers make the inventory legible to both shopping surfaces and assistants.

Marketing. Real source codes mean the channel report stops being an opinion.

A thirty-day order of operations

Cheap and unglamorous first. Fix the location record and hours. Deduplicate part numbers and decide which one is canonical. Publish one taxonomy mapping and force the feeds through it. Validate the merchant feed and clear the errors nobody owns. Canonicalize and prune legacy URLs. Add structured data to the four templates that carry the most revenue. And stop accepting unknown as a source value.

Then and only then does buying more traffic make sense. Media on top of dirty records is a rental payment: it buys a quarter of lift and leaves you with the same foundation, plus a bigger invoice.

The short version

Departments do not fail because their tactics are wrong. They fail because they are each optimizing a different copy of the truth. Clean identifiers, one taxonomy, one location record, one price, one owner, and metadata that states all of it in the open is not a project you finish. It is the floor every other effort stands on.

A shared definition is not abstract. The report page states every denominator in writing for exactly this reason, so a sales number and a service number can be argued about without arguing about the math.