Field notes: the week of September 23, 2026

A client asked us this week why their cost per lead had gone up 30% while their conversion rate on the site had barely moved. The honest answer was uncomfortable: the leads were fine. The problem was that their sales team had stopped calling back within an hour, and the leads had gone cold. We had been staring at the ad account when the leak was three steps downstream. That is the theme of the week: the metric you watch is rarely the metric that pays.

The lag between a good lead and a dead one

We pulled timestamps for one client's inbound form fills against the first outbound call from their team. The median gap was just under four hours. For a home services business where a homeowner is comparing three companies at once, four hours is a lifetime. By the time the call goes out, two competitors have already booked the estimate.

The fix was not a marketing fix. It was a routing fix. We wired the form submission straight into their chat workspace so a human sees it in under a minute, and we set a rule: if nobody claims the lead in five minutes, it escalates. The ad spend did not change. The close rate did.

This is the kind of thing that makes performance-based compensation clarifying. When you get paid on outcomes, you cannot pretend the handoff is somebody else's job. You either find the leak or you do not get paid.

What we changed in the software

Two things shipped this week, both small and both born from the lead-lag problem above.

• Lead timestamps in the portal. Every inbound now carries a first-touch and first-response timestamp, visible to the client. When a client asks why leads feel slow, we no longer argue from memory. We point at the number.
• Grader copy pass. We rewrote the recommendations the site grader hands back so they read like instructions instead of scores. "Your hero section has no phone number above the fold" is more useful than "Conversion: 62/100." A score tells you that you are sick. An instruction tells you what to take.

We also spent an afternoon on something less glamorous: cleaning up how the MCP servers hand context to Claude when we draft a competitor teardown. The model was pulling stale page snapshots on a couple of sites that had relaunched. We added a freshness check so it refuses to analyze a cache older than seven days and re-fetches instead. Small, but it kept us from writing a teardown of a site that no longer exists.

The rebuild-versus-rescue question came back

We had two intake calls this week that both looked like rebuild candidates on paper: old sites, slow load times, dated design. In both cases the right call was a rescue, not a rebuild, and the tell was the same in each.

When we ran their analytics, the pages that carried the traffic were converting fine. The ugly pages were not the pages people landed on. A rebuild would have spent three months and a large budget to replace a working engine because the paint was chipped. The rescue was two weeks of fixing the specific pages that got traffic and left money on the table.

A slow, ugly page that converts is worth more than a beautiful page that nobody has tested. Fix what earns before you replace what offends you.

The trap is that rebuilds feel like progress. There is a launch, a reveal, a before-and-after. Rescues feel like maintenance, so they get deprioritized even when the math favors them. We push clients toward the boring answer because the boring answer is usually the profitable one.

How we tell them apart quickly

The heuristic we use now, before anyone gets attached to a redesign:

• Is the traffic already there, or does the site need traffic it does not have? Traffic present favors rescue.
• Are the top five landing pages converting below their category norm, or is the whole site structurally broken? A few weak pages favor rescue. Broken information architecture favors rebuild.
• Can we ship a meaningful fix in under two weeks? If yes, rescue and measure before committing to more.

If two of those three point to rescue, we do not open a rebuild conversation until the rescue has run for a month.

A note on attribution honesty

A prospective partner asked us this week to guarantee a specific number of leads per month. We said no, and it is worth explaining why, because it is the same reason our compensation works the way it does.

We can control inputs: the offer, the targeting, the page, the response time, the follow-up. We cannot control the market's demand this quarter, the competitor who just doubled their budget, or whether the client's phone gets answered. A lead guarantee pretends those variables do not exist. It sets up a relationship where we quietly optimize for cheap leads that look good in a report and convert badly, because the contract rewards the count and not the cash.

Tying our pay to outcomes is the opposite bet. It means we would rather deliver fewer, better leads that close than a bigger number that impresses nobody's bank account. It also means we say no to work where the outcome depends on something we cannot touch, like a sales team we have no visibility into. That is not caution. It is refusing to get paid for luck.

What we are watching next week

Two open questions we are carrying into the next few days:

• Whether moving the lead-response escalation from five minutes to two is worth the alert fatigue it creates for small teams. Faster is better in theory, but a rule people ignore is worse than a slower rule they follow.
• Whether the grader's new instruction-style copy actually changes what clients fix. We can see who reads the report. Next we want to see who acts on it, and which instructions get ignored so we can rewrite them.

The pattern across all of it is the same one we started with. The scoreboard is not the game. The number on the dashboard is a proxy, and proxies drift from reality the moment you stop checking the thing underneath them.

If you have a site that converts on the pages that matter but leaks somewhere you have not found, run it through the free grader and read the instructions, not just the score. And if you want a partner whose pay depends on finding that leak, apply for partnership.