Field notes: the week of September 16, 2026

A client asked us this week to explain why their cost per lead went up 40% in August while their conversion rate also went up. Those two numbers moving in the same direction confuses people, so we spent an hour on the whiteboard tracing it back. The answer was boring: they had paused their branded search campaign to save budget, which had been quietly absorbing cheap, high-intent clicks and dragging the blended average down. Remove it, and the remaining traffic was more expensive per lead but also more qualified. Nothing was broken. The metric just stopped hiding what was underneath.

That kind of thing filled most of our week. Here is what we worked on and what we changed.

The branded search trap

We see this pattern often enough to name it. A team looks at their account, notices branded search converts at a very low cost, and concludes it is their best campaign. It is not a campaign in any useful sense. It is demand that already exists, captured at the last moment before it would have arrived anyway. Some of those clicks would have come through organic results for free. Pausing branded search does not kill demand, it just moves the cost around and, sometimes, hands a few conversions to a competitor bidding on your name.

The honest way to evaluate branded search is to ask what share of it is incremental. That requires a holdout test or, at minimum, a period where you turn it off and watch organic branded traffic, direct traffic, and total conversions rather than the campaign's own reported numbers. Most accounts never run that test because it feels like turning off a machine that prints money.

We ran the math for this client and kept branded on, but capped the budget so it could not eat spend that should go to prospecting. The point is not that branded search is bad. The point is that a low cost per lead inside one campaign tells you almost nothing about whether that campaign is creating value or just recording it.

What we shipped in the software

Two changes went into the portal this week, both driven by things clients kept asking us to explain manually.

• Blended vs. channel view in reporting. Every performance chart now has a toggle. Channel view shows each source on its own terms. Blended view shows the number the business actually feels: total spend over total qualified leads, regardless of attribution. Both are true. Keeping them on the same screen stops the argument we had this week from happening in the first place.
• Change annotations pulled from the project manager. When someone pauses a campaign, ships a landing page, or changes a bid strategy, that event now drops a marker onto the reporting timeline automatically. Before this, a client would see a line move and ask why, and we would go digging through logs to reconstruct what changed on that date. Now the answer sits on the chart.

Neither of these is clever. Both remove a recurring conversation that added no value. That is most of what good internal software does: it deletes the meetings you were only having to reconcile two views of the same fact.

A rescue that turned into a smaller rebuild

We took on a site this month that graded poorly on structure and speed, and the plan was a rescue: fix the technical debt, keep the content, avoid the cost and delay of a full rebuild. Halfway through, we found that the theme was running three page builders stacked on top of each other, each one loading its own copy of jQuery and its own layout engine. You could strip two of them out, but the content was authored inside all three, so removing them meant re-entering the pages by hand anyway.

At that point the rescue was more expensive than a rebuild, so we changed course. This is worth saying plainly because it cuts against our own incentive to look decisive:

A rescue that costs more than a rebuild is not a rescue, it is a sunk cost you are defending because you already started paying it.

We caught it early because we scoped the technical audit before committing to a path, not after. The lesson we keep relearning is that the choice between fixing and rebuilding is not a philosophy, it is a number, and the number changes once you open the hood. We were wrong about the starting estimate. Being willing to say so within the first week is cheaper than being right about it in the third.

Skeptical note: the AI content flood

A prospect showed us their content plan this week, which was 120 articles for the quarter, all generated and lightly edited. The theory was that volume would win because search rewards freshness and coverage. We told them we did not think it would work, and here is the mechanism, not just the opinion.

When everyone can produce 120 articles a quarter, the marginal article is worth close to nothing. The scarce thing is not text, it is a reason for someone to link to you, quote you, or remember you. Generated coverage of common questions competes against every other site generating coverage of the same questions, plus the answer engines that now summarize all of it before the user ever clicks. You do not stand out by matching the flood. You stand out by publishing the thing the flood cannot produce: a specific number from your own data, a named trade-off you actually made, a process you can describe because you ran it.

We suggested they cut the plan to 20 articles, each anchored to something only they know. That is a harder brief and a slower one. It is also the only version that survives contact with an answer engine.

Small things worth noting

A few items that do not need a section:

• The site grader now flags when a page has a canonical tag pointing at a URL that returns a redirect. Small bug, common on migrated sites, quietly wastes crawl budget.
• We moved our internal standup notes into the same workspace clients use, so nothing about how we work is hidden behind a wall the client cannot see.
• Two campaigns hit their compensation trigger this month, which means we got paid because the client did. That structure keeps us honest about which of the above actually mattered.

If your reporting shows numbers you cannot explain, or you are staring at a rescue that keeps growing, that is the kind of problem we like. You can apply for partnership, or start with the free grader and see what your own site is hiding first.