Case studiesGrowth & Measurement

2.8x growth in 10 months, on measurement the whole company trusted.

Home & furnishings DTC brandClient unnamed10 monthsBuilt in their stack
The situation

Every channel slide was green. The business was flat.

Weekly spend was assembled semi-manually with hand-typed overrides, one analyst finishing the report at 11pm on Mondays. The Monday marketing meeting was a row of positive channel slides built on each platform's own metrics. An MMM vendor updated twice a year, reconciled to nothing. A multi-touch attribution tool had just been bought, but it was one more tab to check, not something a decision ran through. GA4 sessions mostly went unused. Nothing connected to the topline, which was not moving.

The Monday meeting, reconstructedWhere it stood
What the meeting heard: each channel, on its own platform's numbers
Meta
Platform ROAS
4.1
"Strong week"
TikTok
Platform ROAS
3.6
"Scaling well"
Pinterest
Platform ROAS
3.2
"Efficient"
Reddit
Platform CPA
-12%
"Improving"
Snap
Platform ROAS
2.9
"Testing up"
What the business saw, the same weeks: total revenue
Every channel up and to the right. The sum of them, flat.
Illustrative reconstruction of the weekly readout. Every channel positive by its own scoreboard, while revenue stayed flat. Both were technically true.
Each channel was winning by the platform's numbers. Nobody was measuring the game.
The foundation

First, a CAC everyone could plan on. Then a model that earned the budget meeting.

Governed spend came first: automated dbt-modeled digital spend combined with a simple, clear process for the manual channels (influencers, linear TV, direct mail), automated further over time. The weekly number was done by noon Monday instead of 11pm, and finance and marketing stopped debating whose CAC was right. Then an MMM embedded in the business rather than delivered to it: reconciled every 2 months against business trends and inputs, predicting weekly revenue and CAC within ~3%. That gave the monthly finance and marketing meeting a baseline everyone accepted, seasonality included, so it became a budget decision meeting instead of a numbers debate.

Governed spend & CAC
Automated digital spend in dbt, clear manual process for TV, direct mail, influencers. Done by Monday noon, not 11pm
An MMM that reconciles
Reviewed every 2 months against what the business actually saw. Weekly revenue and CAC predicted within ~3%
First-party tracking
Own web pixel: ~20% more sessions than GA4, and one user stitched across visits instead of fragments
One naming convention
Ad naming rebuilt to be MECE, so creative questions like "is UGC actually working" have answers
MMM predicted vs actual, weekly revenueWhy it was trusted
1
2
Actual weekly revenue
Model prediction
~3% average error, seasonality included
1Actual pulls away from the model, giving an early confident read that something has improved in the business.
2Recalibrated to the new baseline, 1-3 times a year, and it accurately forecasts performance again for budget planning.
The practical payoff: a bad week stops triggering a war room. It is either inside normal error against a baseline everyone already accepts, or it is a real change with a date on it.
Weekly predictions against actuals over the period, drawn illustratively. Built on the budget the twice-a-year vendor and the unused tools were already costing. A model that tracks the business week by week is a model whose budget guidance gets used.
The lens

One yardstick for every digital channel: who actually brings new, qualified visitors.

The first-party pixel changed what a journey looked like: the average measurable customer journey went from 1 day to 10, with some past 30. That made TV's long consideration cycle visible, and it gave every channel one leading indicator measured on our own site: cost per first-time quality visitor, a brand-new visitor who views 2+ pages including a product page. The platforms said Reddit, Pinterest, Snap, and TikTok were doing great. The yardstick said they were barely bringing new qualified visitors, and they had quietly drifted to ~17% of spend.

The same channels, on the platform's numbers vs oursIndexed, illustrative
ChannelThe platform reportedOur yardstick: cost per first-time quality visitorThe call
Meta~67% of its first-time visits engage
ROAS 4.1
~$4.20
Scaled
Google non-brand~60% engage
ROAS 1.5 last-click
~$5.80
Scaled
Pinterest~50% engage
ROAS 3.2
~$15
Cut
Snap~45% engage
ROAS 2.9
~$17
Cut
TikTok~45% engage
ROAS 3.6
~$19
Cut
Reddit~40% engage
CPA down 12%
~$24
Cut
The four channels cut had quietly drifted to ~17% of spend. Post-purchase surveys agreed with the yardstick: even customers whose tracked journeys touched them rarely said that is where they first heard of the brand. Two independent reads, same verdict. The pixel and the yardstick are also the piece that stands alone: they paid for themselves before the MMM conversation even started.
Dollar values and engage rates are illustrative; the ranking, the gap, and the ~17% spend share are from the real analysis. Nobody distrusted these channels. Spend had just drifted to them slowly, the way it does.
What it changed

Where the money moved, and why it moved with confidence.

Each call below was made because independent reads kept agreeing: first-party quality, the attribution model (no longer a spare tab once its journey credit could be explained: who initiated, who held, who closed), the MMM, and post-purchase surveys. Where the reads agreed, budget moved. Where they disagreed, that was the finding to run down, not the number to argue with.

The long-tail cut
Reddit, Pinterest, Snap, and TikTok were cut hard from ~17% of spend, redeployed mostly to Meta. Site conversion and the quality-session rate improved almost immediately, and CAC followed. Their share of post-purchase survey responses did not move after the cut.
TV & direct mail flighting
With 10-30 day journeys now visible, the adstock read held up, so long-cycle channels moved 4+ weeks ahead of Memorial Day, Labor Day, and Black Friday, with Meta and Google taking the peak weeks instead.
Google non-brand
First-party data showed it driving top-of-funnel quality while already sitting at ~$1.5 last-click ROAS. Top-of-funnel that pays for itself on the worst-case measure got scaled immediately, the easiest confident win of the year.
Direct mail
The MMM consistently credited it and the reconciliation meetings kept agreeing, so it scaled under the same early-flighting rules.
Landing journeys
Closing the click-to-session gap (from ~20% of paid clicks lost to ~3%) showed which landing pages leaked. Personalized journeys with the ecom team grew Dressers, a smaller category that turned out to acquire the highest-LTV customers.
Before and afterDraft figures
MeasureBeforeAfter
Weekly spend & CAC reportMondays 11pm, manual overridesMondays noon, governed
MMMVendor update 2x a yearEmbedded, ~3% weekly error
Measurable customer journey1 day10 days avg, some 30+
Spend on channels failing the yardstick~17% and drifting upCut, redeployed to Meta & Google
Measurement budgetVendor + tools, detachedSame spend, running the decisions
Draft figures to confirm before publish. These are the operating changes. What they added up to is below.
What we deliberately skipped

Why geo tests didn't make the measurement stack.

We evaluated incrementality testing seriously; in principle it is the gold standard, and the hype is understandable. In practice, at the sample sizes most DTC brands have outside their 1-2 biggest channels, the results were too wide to change a decision. That is not the vendor's fault, it is the math. The decision input is not "did sales lift", it is incremental return against what the spend cost, and that answer came back as a range.

One geo test, read two waysIllustrative
"+5% sales lift, +$350k revenue"the headline version, and it looks like a winner. The budget question is dollars back per dollar spent, and on that question the answer below is a range
The same result in the platform's confidence view: incremental return per $1 spent
The test designer's own power estimates
Holdout2 weeks3 weeks4 weeks
10%54%54%54%
30%57%57%57%
50%59%59%59%
Power = the chance the test detects a real effect at all. Getting it near ~68% meant simulating roughly double the channel's spend for a month.
The quoted number is just the curve's peak. This one looks like a clear winner, and the same test still leaves roughly 1 in 3 odds that the spend lost money. The test cannot tell you which happened.
At ~59% power, a real effect goes undetected roughly 2 times in 5, and the test costs the same either way. On mid-size channels the design was underpowered before it started.
The triangulation above answers the same question weekly, from reads that already exist. The test budget went there instead.
We put the money into the triangulation instead: first-party quality, MMM, and surveys agreeing. A test that cannot change the decision is not worth its cost.
The outcome

Revenue climbed 19%. The price of growth barely moved.

Growth is easy to buy at worsening prices; that is what the flat years before had been avoiding. This year added ~$31M of revenue, up 19%, while blended CAC moved ~4%. That pairing is the report card: spend kept moving to where it worked, so scale did not cost efficiency.

Revenue vs the price of growth, through the yearIndexed, draft figures
Revenue, indexed
Blended CAC, indexed
The gap between the lines is the point.
Indexed and drawn illustratively; the +19%, ~$31M, and ~4% are the real draft figures to confirm. The growth had many parents across the company. The claim measurement makes is the gap between these two lines.
Held

The operating rhythm became the system: every 2 weeks, first-party quality, first-party MTA, MMM, and post-purchase survey ROAS reviewed side by side. Where they disagree is treated as the finding, not the annoyance.

Where it went next

A similar model now predicts quality organic traffic as TV's leading indicator: Meta and TV each drive ~25% of it, and that read guides when TV spends up and down between the twice-yearly MMM refreshes it used to wait for.

[Client unnamed: a home & furnishings DTC brand. Sam to confirm: how identifiable the description can be, how the +19% / ~$31M / +4% CAC year relates to the 2.8x-in-10-months window (same period or sequential), the illustrative yardstick values, and the "same measurement budget" claim vs what the old vendor and attribution tool cost. Owner: Sam]

Get in touch.

Tell us what you are trying to figure out. If we can help, we will say how. If we are not the right fit, we will say that too, and point you somewhere better.

01
Free work first
We start by doing a real piece of work on your data, free, so the first thing you judge is output, not a pitch.
02
First build, fixed scope
A trusted core metric layer in weeks, not quarters, at a fixed price. Small enough to be safe, real enough to matter.
03
Scale when it earns it
Month to month from there, from a few thousand a month. You own every model, dashboard and definition.