Why last-click ROAS pays the wrong channels
In 1902, Hanoi paid one cent for every dead rat. The city ended the year with more rats than it started with.
It was a plague year, and Hanoi was a French colony. The bounty paid out on a tail, not a whole rat, so people adapted. Inspectors started seeing rats running around with no tails. Then they found the smuggling ring bringing rats in from the countryside.
You pay for tails, you get tails. The rats were never the point.
Goodhart's Law
Economists call this Goodhart's Law: when a measure becomes a target, it stops being a good measure.
A last-click ROAS report is a bounty on tails.
How last-click pays for sales that were already coming
Under last-click attribution, budget flows to whoever owns the last click before a conversion. Many of those channels are not causing the sale. They are clipping the tail off a conversion that was already on its way.
- Click injection and click spam do it on purpose. They fire a click just before an install that would have happened anyway, to claim the credit.
- Branded search and retargeting do it by accident. They show up right before a purchase the customer had mostly decided on, after your creators, PR or word of mouth did the convincing.
The sale still happens. You just paid for it.
The tell
The French had a clear signal, and so do you: paid conversions climb while organic conversions drop. Same rats, new tails.
If your attributed paid results keep improving while total revenue stays flat, some of your channels are being paid for organic demand.
I have spent two years defending channels in budget meetings that one geo holdout later showed added almost nothing. The dashboard was not wrong about what it measured. It was measuring the wrong thing.
The move
- Plot paid and organic conversions on one chart. If they move in opposite directions, investigate the paid side first.
- Look at total revenue, not attributed revenue, when you judge a budget change.
- Run a geo holdout on your best-looking channel. Pause it in matched regions and see whether total revenue moves.
- Check your fraud tooling for click injection and click spam, especially on Android and on networks paid per install.
- Shift budget to what moved total revenue when it was paused. That is the only credit that counts.
Bottom line
Pay for rats, not tails. Measure the lift, not the click. Scale what moved.
Further reading: historian Michael G. Vann, "Of Rats, Rice, and Race: The Great Hanoi Rat Massacre" (French Colonial History, 2003).
FAQ
What is wrong with last-click attribution?
It gives all the credit to the final touchpoint before a conversion. Channels that appear right before a decision, such as branded search and retargeting, get rewarded even when the customer was already going to buy.
What is Goodhart's Law in marketing?
When a measure becomes a target, it stops being a good measure. If budget follows last-click ROAS, channels get better at owning the last click, not at creating new customers.
How can I tell if a channel is taking credit for organic sales?
A common sign is paid conversions rising while organic conversions fall and total revenue stays flat. A geo holdout test, pausing the channel in some regions, gives a direct answer.
What should replace last-click ROAS for budget decisions?
Incrementality: the change in total revenue caused by the spend. Geo holdouts and other lift tests measure it directly.