The best data in the CRM is the part nobody structures

The customer needs a third row for a large dog. Her lease ends after the school year. His parents passed away and there may be a trade once the estate settles.

None of that fits neatly into year, make, model, phone and email. It is also the context that makes the next conversation feel like a conversation.

The lead standard was built for transport

ADF is an open XML-based standard created to move automotive leads between shopping services, manufacturers and dealerships. Its basic structure carries the prospect, vehicle, customer, vendor and service-provider information required to deliver a lead.

That was useful infrastructure. It was not designed to preserve every piece of human context created across calls, texts, emails and showroom visits.

As those conversations accumulate, the CRM note becomes a catch-all. Important facts sit beside copy-pasted templates, call summaries, tasks and stale promises. The answer is not to throw the notes into a model and trust whatever comes back.

Use a narrow context schema

Start with fields an operator can define and verify:

• vehicle use case;
• timing constraint;
• trade status;
• payment or term concern stated by the customer;
• preferred channel and time;
• open promise made by the store;
• next action and owner;
• source note and timestamp.

Every extracted field should point back to the original record. If a manager cannot inspect the sentence that created the field, the context should not drive an automated action.

Separate facts, inferences and instructions

"Customer said the third row must fit a dog crate" is a fact from the conversation. "Likely family buyer" is an inference. "Send Palisade inventory tomorrow" is an instruction. Store them differently.

Facts can be durable. Inferences need confidence and expiration. Instructions need an owner, due time and completion receipt. Mixing the three is how a helpful summary turns into a bad campaign.

Protect the customer while improving the follow-up

CRM notes can contain financial, personal and sensitive information. The FTC requires covered dealers to control access to customer information, encrypt it, monitor systems and oversee providers. A context layer should reduce exposure, not spread the full transcript to every tool.

Give a follow-up workflow the smallest useful view. A salesperson may need "needs third row, prefers text after 5 p.m." The marketing platform does not need the entire conversation. A general reporting tool does not need raw finance notes.

A practical first build

1. Choose one department and one conversation channel.
2. Review fifty records with the people who work them.
3. Define five context fields that would have changed the next action.
4. Extract into a review queue before writing anything back.
5. Measure correction rate and field usefulness.
6. Only then allow approved fields to shape tasks or drafts.

The CRM already contains more context than most stores use. The opportunity is not another blast list. It is a follow-up that remembers what the customer actually said.

Keep the note, improve the handoff

Structured fields should not replace the original note or transcript. They should make the next handoff faster while preserving the source for review. A corrected extraction should improve future drafts, but it should never rewrite what the customer actually said.

This is also an adoption advantage. Salespeople are more likely to trust a suggested action when they can see the sentence behind it, correct the field and move on. The system learns the store's language through review instead of asking the floor to trust a black box.