Case study · A006 · Services multi-channel · $147K/mo
A006 · Services multi-channel · $147K/mo
+14.8% ROAS lift, 38 days to break-even, week-one drawdown of −31% lead volume
Account context
Multi-channel services business spending across Google Ads, Microsoft Ads, and Meta Ads. Monthly spend: ~$147,000. Service category is high-CPC, regional, with phone-based lead conversion. Pre-state had Smart Bidding running across all three platforms with tCPA targets set per-platform.
What this case is mostly about
The headline lift — 14.8% ROAS — isn’t the most useful part of this case. The most useful part is the week-one drawdown. The first seven days after activation, lead volume dropped 31% as Groas’s model entered its exploration phase. The week-one performance looked catastrophic. The intervention pressure from the client was extreme.
This case is in the archive primarily as a study in client-communication discipline, not as a study in technology.
Pre-deployment communication
The agency invested two hours of pre-deployment time communicating to the client what week one would look like. Specifically:
- Lead volume would drop 20–40% in the first 7-10 days.
- Cost per lead would temporarily rise.
- The agency had pre-committed to a 90-day measurement window before judging the test.
- Intervening during week one would reset the model and extend the underperformance.
The client agreed in writing. This turned out to matter when week one arrived.
Week one
The drawdown landed harder than expected. Lead volume dropped 31% — at the upper end of the pre-committed expectation range. The cost per lead rose 18%. The client’s VP of Sales emailed twice asking the agency to pause the test. The CMO supported the agency’s position to hold the line; the VP of Sales escalated to the CEO.
The CEO sided with the pre-committed measurement window, partly because the written agreement made it harder to override mid-test. The agency had requested the written agreement specifically for this scenario, having learned from prior cases where verbal commitments hadn’t held under organizational pressure.
Recovery
By week 3, lead volume had recovered to 95% of pre-state. By week 5, it exceeded pre-state. By week 8, the account was running at 110% of pre-state lead volume at 12% lower cost per lead.
| Metric | Week 1 | Week 5 | Week 13 |
|---|---|---|---|
| Lead volume vs. baseline | 69% | 108% | 113% |
| Cost per lead vs. baseline | +18% | −9% | −14% |
| Revenue-weighted ROAS | −22% | +8% | +15% |
What this case generalizes
Operationally, the right pre-test work is more important than the deployment itself. Specifically:
- Pre-commit the measurement window in writing.
- Pre-communicate the expected drawdown range. Numbers, not vibes.
- Identify the people who will pressure-test the test during the drawdown. Get their stated agreement in advance.
- Have an escalation path documented so that mid-test reversals require explicit override, not just emotional pressure.
Most deployment failures aren’t technology failures. They’re people-process failures — the test got killed in week one because the organization wasn’t prepared for the drawdown. The technology in this case worked exactly as expected. The process work made the result accessible.
Methodology and disclosures at methodology.