The crash that wasn't

The message came in at 8:40 on a Tuesday morning. "Conversions have collapsed. Down 96% since Friday. Pausing all campaigns until we understand."

I have received a version of this message perhaps a hundred times in sixteen years, from clients across time zones. The first few times I panicked with everyone else. Now I ask one question before I do anything else. What else moved?

What the dashboard showed

Conversions had gone from about 150 a day to near zero. Cost per acquisition had gone from a reasonable number to a number with too many digits. The dashboard was red from top to bottom. Every automated bidding system on every platform had noticed the same thing and was starting to do what those systems do when the conversion signal disappears. They panic too. They cut bids, shift budget and go looking for conversions somewhere else.

What else moved

The answer was nothing. Spend was flat. Clicks were flat. Paid sessions on the site were flat. Add-to-cart events, which fired from a different tag, were flat.

A developer had shipped a site update on Friday evening. The conversion tag had been moved to a template that did not load on the thank-you page. The marketing was working. The measurement was broken.

Think of a bathroom scale with a dying battery. It reads 40 kilos one morning. You have not lost 30 kilos overnight. You need a battery. Nobody starts a crash diet over a dead battery, but I have watched companies pause a working campaign over a dead tag.

If this client had paused everything, they would have lost four days of sales in a good week, and the bidding algorithms would have spent the next two weeks re-learning what they already knew.

The bigger version of the same problem

Small accounts are not the only place this happens. The whole industry is living through a bigger version of it this year.

In July 2026, VideoAmp and Nielsen One Ads both withdrew from MRC accreditation, the independent audit of TV measurement (MediaPost). In India, the broadcast ministry ordered BARC to stop publishing TV ratings in July pending re-registration, so this quarter's television deals are being struck on stale currency (exchange4media, 31 August 2026). On 16 September, a large US agency announced it would pipe Roku's viewing data straight into its planning tool in near real time. Roku also sells the inventory that data will plan, target and measure (AdExchanger).

In each case the question is the same as my Tuesday morning question. If the measurement changed, which conclusions are you still allowed to draw?

The rule we built

After enough Tuesday mornings we wrote the rule down and then built it into the system. Never grade marketing on broken measurement.

In practice: every finding sits downstream of something. A cost-per-acquisition finding depends on the conversion tag. A conversion finding depends on site tracking. A return-on-ad-spend finding depends on revenue data landing on time. If an upstream signal breaks, every finding below it is withheld for that scope until the signal is back. The system does not say "CPA is up 900%". It says "conversion tracking failed on 3 of the last 7 days, so CPA verdicts are suspended for these campaigns".

Thin data gets the same treatment. Not enough of it to make a call means a watch-list entry, not a verdict. Every finding carries a confidence score out of ten, and when the evidence is thin the honest answer is "not enough data". I would rather the system said that than sounded sure and was wrong.

Three things to do this week

1. Ask for tag uptime as a line in your weekly report: hours the conversion tag fired, out of 168. Your tag manager or server logs have it.

2. Pick one metric that does not depend on your conversion tag. Clicks, paid sessions, add-to-carts from a separate source. When conversions "crash", look at that number first.

3. Write down, on one page, which conclusions are off limits when which signal breaks. Give it to whoever has the authority to pause campaigns. That page has saved more budget than any bid strategy I have ever set.

What we do about it

This rule is one of the 84 checks that run on every campaign we manage, every day. It is the reason I trust the alerts our team gets. When the system says Critical, it has already ruled out the outage.