An agency lead joined a call a few weeks ago and skipped the greeting. "He cut it," she said. I asked who, and what. The founder of one of her clients had cut the programmatic budget by nearly all of it. He did it over a weekend. Her team found out when a revised insertion order landed in her inbox.

I asked how it happened. He had sat down on a Sunday with a coffee and opened Google Analytics. He is a founder who looks at the numbers himself, and she said that like it was a credential. He filtered the report to the programmatic vendor and looked at the conversions assigned to it. Almost nothing. A thin line along the bottom of the chart. He did what any owner does with a line item that appears to produce nothing. He cut it.

Nobody in that story was careless. He read the report he had access to and drew the conclusion the report supported. That is what makes it worth telling.

Then an independent read of the same period came back. It showed a meaningful lift in sales while the programmatic ads were running. The sales were real. They were landing in a large online marketplace. People saw the ads, went to the marketplace, and bought there. Google Analytics never saw those purchases because they never touched the company's website. By construction, it can't.

The tool answered a smaller question

The founder asked whether the channel worked. The tool answered a smaller question: how many visitors arrived on the company's own site and completed a purchase there. For a brand whose customers finish the sale on a marketplace, that number can sit near zero while the channel does real work.

The tool had no bug. It counted what crossed its own threshold, and it counted accurately. A security camera pointed at the front door records nobody breaking in through the back. Nobody calls the camera broken. They call the footage incomplete, and they stop treating it as proof of a quiet night.

Absence of evidence in a system that cannot see something is not evidence of absence. A quiet dashboard tells you the dashboard saw nothing. It says nothing yet about whether anything happened.

Every agency has this executive

I have yet to meet an agency owner who doesn't have a version of this founder. He opens the dashboard himself. He trusts the number because he can see it, and he trusts it more than a model he can't. The chart is clean and it is his, and that makes it persuasive.

Cutting also feels like discipline. Nobody gets questioned for trimming a line that reported nothing. The cost arrives later and arrives quietly. When the ads stop, the sales they were feeding thin out in the same marketplace the report couldn't see. The damage never shows up in the dashboard that justified the cut. He will never get the screen that says he made a mistake.

The more confident the person, the more expensive the error. A hesitant executive asks someone first. A confident one with a filtered report in front of him acts on Sunday.

Ask what the tool can see

Before you cut a channel on analytics data, establish what fraction of that channel's likely conversions the tool can even observe.

It takes an afternoon. Write down every place a customer can finish a purchase. Your website. A phone call or an appointment. A physical store or a dealer. A marketplace. Next to each one, mark whether your analytics package can see it. Be strict about that column. If a customer can complete the sale without ever loading a page you instrument, the tool is blind to that path.

Now look at the channel you were about to cut. If most of its likely conversions happen in the unmarked places, a low number in the report means very little. Get an independent read before you touch the budget. A lift study or a modeled estimate will do. Anything that measures the business rather than the click.

Put one question to every report that is about to justify a cut: what would this tool have missed even if the channel worked perfectly? If the answer is a lot, you have not looked at the channel yet. You have looked at the tool.