Chapter I — III
A senior-authored reading on signal loss, in-house MMM, and weekly recalibration.
Three short chapters. Each one closes a loop the standard GA4-and-Meta-pixels stack leaves open. No vendor pitch, no replatforming fear — just the version of the buying decision we run internally, now written down.
CHAPTER 01
What "signal loss" actually means after iOS 14+
The cleanest way to describe what happened after App Tracking Transparency is that the marketing org didn't lose data — it lost resolution. Aggregate numbers still arrive; the chain of evidence that connects a click to a checkout broke across roughly 41% of post-click touchpoints on a standard GA4 install. That gap is large enough to invert the conclusion of a weekly buying meeting. A channel that looks like a cost center on Monday becomes a profit center once the unobserved conversion path is reconstructed, and vice versa. Most teams either shrug through this or throw more pixels at it. We chose the third option: rebuild the tracking stack against a fixed reference set, recover what we can server-side, and stop pretending the rest is gone. That habit is where the Signal Recovery Framework begins.
"Standard GA4 setups lose roughly 41% of post-click signal against our audited reference set — the gap that flips a channel from cost center to profit center in a single Monday meeting."
CHAPTER 02
Why in-house MMM beats black-box attribution
A media mix model is not a dashboard. It is a hypothesis about how your channels compound, expressed as coefficients you can interrogate, stress-test, and rerun when the assumption breaks. That property — interrogability — is the entire sale. When a vendor sells you an attribution product, you are buying a closed world whose internal weights you may not audit. When you run MMM in-house, the weights are a spreadsheet you can hand to a finance counterpart and argue with in plain English. We have operated our own MMM since 2020, have weathered three platform-side measurement deprecations without rebuilding the model from scratch, and now maintain a standing 41.6-month rolling training window that resets weekly. It is not exotic work. It is unglamorous, repeatable, and durable — which is exactly why senior teams defend it.
"A model you can interrogate on a Tuesday is the only model a CFO will underwrite on a Wednesday."
CHAPTER 03
Weekly recalibration as a buying discipline
A model that recalibrates once a quarter tells you what worked ninety days ago. Marketing moves at a weekly cadence, and so do we. Every Friday, the analyst on each engagement reruns the MMM against the current week's incrementality tests, updates the channel coefficients, and writes a one-page memo to the media buyer specifying the marginal dollar allocation for the next seven days. That memo is the buying decision. Not a platform-native forecast, not a CMO-instinct spreadsheet — an explicit written argument that lands in Slack each Friday at 4 p.m. local. After three years of running this cadence across 180+ active engagements, the average ROAS improvement inside the first 90 days clocks 3.4x against the prior baseline — not because the model is clever, but because a weekly rhythm of recalibration is faster than a monthly rhythm of regret.
"A weekly rhythm of recalibration is faster than a monthly rhythm of regret — and shows up as 3.4x ROAS inside the first 90 days, audited across 180+ engagements."
