Every dealer group with a multi-channel budget has the same problem. Google Ads reports one number. Meta reports another. Microsoft reports a third. TikTok reports a fourth. The internal CRM shows something else entirely. And the marketing director is asked to move $40,000 of budget on Monday based on numbers that disagreed by 30% on Friday.
This playbook is the weekly cadence that stops the disagreement from becoming a decision paralysis. It is not a model — it is a discipline. The model can be whatever your team wants (last-touch, time-decay, data-driven, custom). The discipline is what makes the model useful.
Step 1. Freeze the model
Pick one attribution model and stop changing it. Time-decay with a 30-day window is a defensible default. Last-touch is fine if the sales cycle is short. Whatever you pick, freeze it for the quarter. No mid-quarter model changes for any reason short of a catastrophic error.
The frozen model is the reference. Every platform's number gets translated into the frozen model. Google's data-driven attribution result is not the truth — it is Google's opinion. Your frozen model is the group's opinion. Both are useful; they are not the same.
Step 2. Define the weekly rebase window
7-day rolling, ending Sunday midnight local time. All budget decisions get made on Monday morning against last week's rebased data. Never make budget decisions against partial-week data — the model needs the full 7 days to stabilize.
The rebase itself is a single job that runs Sunday at midnight: pull the last 30 days of touchpoint data, run the frozen model, and write the current week's attribution ledger. That ledger is immutable. Late-attributed conversions land in the following week's ledger with a footnote.
Step 3. Instrument the two callbacks that matter
Every multi-channel dealer group has two conversion callbacks that carry the pipeline:
- Lead-form submission — captured on the VDP, on landing pages, in inventory ads. This is the touchpoint every platform can attribute.
- Phone call — captured by call tracking (CallRail, Marchex, or equivalent). This is where 40-60% of dealer pipeline lives and where attribution fights hardest.
Instrument both with a unique correlation_id that flows from the ad platform → landing page → form/call → CRM. If the correlation ID does not survive that entire chain, the frozen model has to guess, and the delta between platforms explodes.
Step 4. Reconcile platform-reported vs internal ledger
Every Monday morning, before the budget meeting, run the reconciliation report. Two numbers per channel:
- Platform-reported conversions (from the platform's UI)
- Internal-ledger conversions (from the frozen model)
The delta is the signal. A ±10% delta is normal — attribution windows differ, view-through conversions differ, deduplication logic differs. A ±25% delta is a diagnostic problem. A ±40% delta is broken instrumentation.
Step 5. Publish the delta report
One page. Four channels. Delta % per channel. Comment on any channel above 20%. Send to the marketing director + the sales director + the GM. Not the agency — the internal team. If the agency needs it, they get a redacted copy the following day.
The delta report is what makes the budget meeting productive. Without it, the meeting is spent arguing about which platform's number is real. With it, the meeting is about which delta needs fixing this week.
Step 6. Move budget with a 24-hour cooldown
The budget decisions from Monday's meeting do not go live until Tuesday. This is the single most-broken rule in dealer groups and the single most-valuable one to enforce.
The 24-hour cooldown catches the mistakes that emotional Monday-morning decisions produce. Someone always wants to pull $10k from a channel that had a bad week; the cooldown gives them one night to look at the 4-week rolling average and reconsider. About one in five Monday decisions gets adjusted overnight — and it is almost always the right adjustment.
Failure modes we have seen
- The model gets re-picked every month. The marketing director changes attribution model to whatever the current platform pitch says. Freeze the model. Change quarterly at most.
- Correlation IDs die at the CRM boundary. The CRM strips URL parameters. Fix the CRM integration or the entire cadence collapses.
- Late-attributed conversions get chased. 45-day-tail conversions land in the current week's ledger and shift the delta. Cap at 30 days; footnote the rest.
What to build first
If you have one week: freeze the model, define the rebase window, and instrument correlation IDs on the two callbacks. Skip the alerts. Skip the reconciliation report. Just get the frozen model and the correlation IDs. Two weeks of that and the delta report writes itself.
How AUTONOMi Runs This Playbook For You
The frozen-model + weekly-delta + 24-hour-cooldown discipline described above is what AUTONOMi's attribution engine enforces automatically. The model is frozen at the account level and locked behind AXIOM — nobody, not the agency, not the platform rep, not the marketing manager, can change it without hash-chained approval. The Monday reconciliation report is generated at 6am, delivered to the GM's inbox at 7am, and every budget shift proposed against it goes through the 24-hour cooldown before it hits any ad account.
Correlation IDs are stamped at the pixel layer and preserved through the BDC layer + the CRM webhook, so the "correlation IDs die at the CRM boundary" failure mode described above is designed out of the system. Late-attributed conversions past the 30-day cap are automatically footnoted, not chased. Every conversion decision — attribution model change, budget move, channel pause — is recorded in a hash-chained audit ledger you can hand a group CFO in five minutes.
An internal team building this from scratch is looking at two quarters and a full-time analyst hire; AUTONOMi turns it on in 14 days and the analyst hire becomes optional. If you want the Monday cadence running against your ad accounts by the end of this month, start now. If you're a group of five or more rooftops and want the delta report at the group level, get in touch and we'll scope the group deployment on the same call.
The model is not the discipline. The cadence is the discipline.
