What reporting cannot tell you

Here is a pattern I have run into more than once. The details below are a composite rather than one account, but the shape of it is exact.

A dashboard reports that return on ad spend fell 18% in September. It is right. The explanation it generates is also right, as far as it goes: cost per click rose, conversion rate fell, and most of the movement sat in Shopping. Every number in that summary is accurate.

It still points at the wrong decision.

The explanation was true and useless

The obvious move from that summary is to pull back on Shopping. Costs are up, conversion is down, so reduce exposure until efficiency recovers. That is what the data supports, and it is what most people would do.

What the dashboard cannot see is that a chunk of the catalogue went out of stock mid-month. Those products stayed in the feed, kept serving, kept taking clicks, and converted at close to nothing. The blended numbers look like a performance problem because the broken part is averaged in with the working part.

Cutting budget reduces spend on the products that are still selling well, while leaving the actual problem in place. Efficiency recovers eventually, mostly because the out-of-stock items cycle out on their own, and entirely the wrong lesson gets learned.

The right move is a feed fix that takes about twenty minutes, and no budget change at all.

Why this is getting more common, not less

The tools have improved considerably. They surface anomalies quickly, write clear summaries, and are right about what the numbers say far more often than they are wrong. That is real progress and I use them daily.

But every one of them shares a limitation. They can see the account. They cannot see the business the account is attached to. They do not know which product is a loss leader, which supplier is unreliable, that the sales team stopped following up on a lead source in August, or that last year’s comparison period included a one-off promotion that will never repeat.

So the explanation arrives fluent, confident and plausible, and it is missing the one fact that changes the decision.

What experience actually contributes

Not the ability to read a chart. Anyone can read a chart, and the software reads it faster.

What experience contributes is a catalogue of how these situations have gone before. After enough accounts, an 18% drop in September stops being a number and starts being a shape you recognise. You have seen this exact pattern be a feed problem, a tracking problem, a seasonality artefact, a competitor entering the auction, and a genuine performance decline. You know which questions separate them, and roughly in what order to ask.

That is not intuition in the mystical sense. It is pattern matching against a few hundred previous cases, most of which taught something expensive.

The practical version

If you are reading automated explanations of your paid media, they are probably correct about what happened. Treat them as the input, not the answer.

Before acting on one, three questions are usually enough:

What changed outside the ad account? Stock, pricing, the site, the sales process, a competitor. Most unexplained performance movements are not advertising problems.

Is this the metric that matters, or a proxy? Return on ad spend moving is not the same as profit moving, and the gap between them is where expensive mistakes live.

What happens if I do nothing for two weeks? A surprising amount of what looks like a problem is noise, and intervening makes it worse by resetting the learning the platforms have already done.

None of those are questions a dashboard will ask you. They are the job.