Field notes: the week of September 2, 2026

A client asked us this week why their paid search cost per lead had climbed 40 percent since June while the volume stayed flat. The account had not changed. The bids had not changed. What changed was that three competitors entered the same auction with looser budgets and worse tracking, and they were willing to pay more for clicks that would never convert for them. We were bidding against people who did not know they were losing money. That is a hard problem, because you cannot out-wait an opponent who has no scoreboard.

The auction problem nobody wants to name

Most advice about rising ad costs assumes the fix is on your side of the account: tighter match types, better negative keywords, a smarter bidding strategy. Those help at the margin. But the base rate of an auction is set by everyone in it, and when a competitor with venture money or a bad agency decides to spend, your cost per click goes up regardless of how clean your account is.

Here is what we did instead of fighting the auction head on:

• Shifted budget toward branded and high-intent long-tail terms where the competitor set was thinner and the conversion rate was three to four times the generic head terms.
• Moved a portion of spend into a channel the competitors were ignoring entirely (a local service directory with intent that mirrored search but a fraction of the CPC).
• Rebuilt the landing page for the two campaigns that still had to compete on price, so the conversion rate improvement offset the higher click cost.
• Set a hard ceiling on the generic campaigns and let them go dark during the hours when the loose-budget competitors were most active, which we found by pulling the auction insights report by hour.

None of that is clever. It is just refusing to treat the auction as the only place the fight happens.

You cannot win a bidding war against someone who does not know they are losing, so stop bidding where they are and go find the demand they cannot see.

A rescue that turned into a rebuild

We wrote recently about how to decide between fixing a site and starting over. This week a decision we thought was a rescue quietly became a rebuild, and it is worth showing the moment it flipped.

The site was a mid-size services business on a page builder plugin that had accumulated eleven years of edits. The plan was to fix the three templates that mattered, leave the rest alone, and ship in two weeks. Then we tried to change the header navigation and discovered it was hard-coded in four separate places, each slightly different, because past contractors had duplicated it rather than edit the shared component.

The test we use is simple: if a change you need to make in one place has to be made in five, the site has no single source of truth, and every future edit will cost five times what it should. That site failed the test. We stopped, priced the rebuild against six months of expected edit overhead, and the rebuild won on cost alone before we even counted the performance gains.

The lesson is not "rebuild more." It is that you cannot judge rescue versus rebuild from the outside. You have to make one real change and see what it costs.

What the grader keeps catching

We run a lot of sites through our own site grader, both prospects and clients, and a pattern held again this week. The most common serious problem is not slow load time or missing meta tags. It is pages that rank and get traffic but have no conversion path on them at all.

A blog post pulls in 800 visits a month, answers the question it was written to answer, and then ends. No offer, no next step, no reason to give an email address, nothing but a footer. The traffic arrives, reads, and leaves. From an analytics view it looks like a success because the pageviews are real. From a revenue view it is a leak.

Why this happens

Content and conversion are usually owned by different people, or different agencies, or the same person at different times who was thinking about different goals. The writer optimized for the ranking. Nobody optimized for what happens after the reader is satisfied. The fix is not a popup. It is a relevant next step that matches where the reader is: a related service page, a specific tool, a checklist that requires an email, a calculator that does part of the work for them.

We added conversion paths to the top eight pages of a client's blog this week. We will not have clean numbers for a month, but the mechanism is not in doubt. Traffic that already exists and already trusts you is the cheapest conversion you will ever get.

A note on our own tooling

We shipped a change to the project manager this week that we resisted for a long time. Every task now requires an expected outcome before it can be marked active, not just a description of the work. "Rewrite the services page" is not enough. "Rewrite the services page to move the quote-request rate from its current level" is.

This slows down task creation, and people complained about it, which is fair. But the reason for it is that we are paid on outcomes, and a task with no stated outcome is a task nobody can tell you was worth doing. It forces the argument to happen when it is cheap, at the start, instead of at the end when the invoice is due.

We also connected the site grader results into the workspace through one of our MCP servers, so a partner can ask Claude to pull a client's current grader score and the three lowest-scoring items without leaving chat. Small thing. It removes a five-minute context switch that used to happen ten times a day.

What we are watching next week

Three things are open going into next week:

• Whether the hourly budget ceiling on those generic campaigns holds the cost per lead down without cutting total volume too far. If volume drops more than 15 percent we will reverse it.
• The first conversion data from the blog pages we fixed, to confirm the leak was where we think it was.
• A pricing conversation with a growth-stage client who wants to move from a mixed retainer to a fully performance-based structure. We want that, but only if the attribution is clean enough to be fair to both sides.

If you are carrying a site that fails the single-source-of-truth test, or you are watching your cost per lead climb for reasons that have nothing to do with your account, run it through the free grader first and see what it surfaces. If the problems are structural and you would rather have a partner paid on the outcome than the hours, apply for partnership and we will tell you honestly whether it is a rescue or a rebuild.