Field notes: the week of September 30, 2026

The most useful thing we did this week was kill a report that three people were maintaining and nobody was reading. It had been on a weekly cadence for four months. When we asked which decision it had ever changed, the honest answer was none. That question, "what decision does this change," turned out to be the through-line for most of what happened this week.

The report nobody read

The report tracked twelve metrics across four channels for one client. It looked thorough. It was updated every Monday morning, which cost roughly two hours of someone's time each week plus the context-switching around it. We built it early in the engagement because the client asked for visibility, and visibility felt like the responsible thing to give.

The problem is that visibility is not the same as decision support. A dashboard that shows twelve numbers going up and down does not tell you what to do. It tells you what happened, which you can usually feel anyway. The client skimmed it, said "looks good," and moved on. Nobody ever pointed at a line and changed a budget, paused a campaign, or rewrote an offer because of it.

So we replaced it with a shorter artifact: three numbers that map to money, plus one sentence on what we are changing next week and why. The three numbers are cost per qualified lead, close rate on those leads, and revenue per closed deal. Everything else feeds into those or gets ignored.

A report that has never changed a decision is not information, it is decoration, and decoration has a maintenance cost you keep paying.

What we look for before building anything recurring

The report cleanup made us go back through our own recurring work and ask the same question of each item. Some of it survived. Some of it did not. Here is the filter we settled on for anything that runs on a schedule:

• Does it change a decision? If we cannot name the specific decision and the person who makes it, the recurring work is theater.
• Would someone notice if it stopped? We tested this by quietly delaying a few internal reports by a day. Two got chased within an hour. The rest, silence.
• Is the cadence matched to the decision? A metric you can only act on once a quarter does not need a weekly report. A metric that moves daily does not survive a monthly review.
• Can the underlying number be trusted? A precise report built on a broken attribution model is worse than no report, because it launders a guess into something that looks like a fact.

That last point cost us the most time this week, so it gets its own section.

Attribution keeps lying, and we keep believing it

We spent a chunk of the week untangling why two clients had conflicting numbers for the same channel. The ad platform claimed one figure. The analytics tool claimed another. The client's own CRM claimed a third. All three were "correct" inside their own model, and all three were being quoted in meetings as if they were the truth.

The mechanism is boring and it matters. Ad platforms count a conversion when someone who saw or clicked an ad later converts, inside their own attribution window, using their own view of the customer. Analytics tools assign credit based on the last non-direct click, or whatever model you configured and then forgot about. The CRM only knows what someone typed into a "how did you hear about us" field, which is a survey, not a measurement.

None of these is the real answer. The real answer lives in the gap between them. When we see the platform claiming far more conversions than the CRM can confirm, the platform is probably taking credit for demand it did not create. When the CRM shows deals the platform never counted, the tracking is leaking somewhere.

What we actually did about it

We picked one source of truth per client, and it is almost always the money system: the CRM or the payment processor, whatever holds actual closed revenue. Everything else becomes a directional signal, useful for spotting movement, not for settling arguments. We wrote that decision down in each client's workspace so the next person who opens a platform dashboard does not restart the debate.

This is not elegant. It means we sometimes report smaller numbers than a channel would let us claim. Growth-stage founders can handle a smaller honest number. What they cannot handle is discovering six months in that the number they raised on was a platform's self-report.

Small software changes that removed friction

On the product side, most of what we shipped this week was subtraction. The pattern held: the improvements that mattered were the ones that removed a step, not the ones that added a feature.

We changed how the project manager surfaces stalled work. Previously it showed everything, sorted by date, which meant the important stuck items sat next to routine ones. Now anything with no activity for five business days floats to the top with a plain label. No color-coded severity system, no scoring. Just "this has not moved, look at it." The first day it ran, it caught two things that had genuinely fallen through, which is the whole point.

We also cleaned up the site grader's output for one specific failure mode. When a site blocks the crawler, the old behavior was a vague error. Now it tells you exactly what it hit, a robots rule, a firewall, a login wall, so you know whether the problem is the site or the tool. Small change, but "the tool is broken" support messages dropped to near zero after it.

What this week actually taught us

The thread connecting the dead report, the attribution mess, and the software subtraction is that most teams accumulate work that once made sense and never gets re-examined. The report made sense when we did not know the client. The attribution debate made sense before anyone picked a source of truth. The feature clutter made sense when we thought more visibility was the goal.

Reviewing recurring work with the "what decision does this change" question is not a one-time cleanup. It is something we now do on a schedule, which is a little ironic given the topic. The difference is that this particular recurring task keeps earning its place, because every time we run it, something gets cut.

If you are carrying reports, meetings, or dashboards that nobody can tie to a decision, that is worth an afternoon of honesty this week. And if you want an outside read on where your marketing is actually leaking, the free grader is a fast start. If the deeper problem is that nobody owns the "what changes this week" sentence, that is the job we take on: apply for partnership and we can talk about whether it fits.