The ROAS Drop That Wasn't
A travel advertiser flagged that conversion value per unit of spend was sliding through September. The instinct in that moment is to start changing bids. I started by checking whether the drop was real.
Step 1: compare like with like
September was only three weeks old. Comparing a partial month to a full one flatters the full one, so I compared the first 22 days of September with the first 22 days of August.
September came out roughly 2% higher, not lower.
Step 2: find where the step-down actually happened
Pulling daily data back to June showed a genuine drop, but between June and July rather than in September. Nothing about efficiency changed at that point. What changed was the conversion mix: an app-download conversion with zero value started counting alongside bookings. Every one of those diluted the blended value-per-spend figure without a single booking being lost.
Step 3: explain the last few days
The final days of the month looked weak. Travel bookings report against the date of the ad click, and they keep arriving for weeks afterwards. Recent days always look worse than they will end up. That’s conversion lag, not performance.
What was actually worth acting on
The diagnosis also surfaced two real issues:
- Two generic-keyword campaigns returning far less value per unit of spend than the account average
- One Demand Gen campaign with unusually high clicks at near-zero cost per click and almost no conversion value. That pattern points to a traffic-quality or tracking problem, not a creative one.
The lesson
Before optimising a falling metric, rule out three things: unequal date ranges, a changed conversion mix, and reporting lag. Here, all three explained the whole “drop”. Changing bids would have fixed nothing and risked a lot.