Has this worked before?
Twice, at very different sizes: inside a $50MM consumer brand over five months, and in a Brooklyn retailer built from no ecommerce at all.
A $50MM consumer brand: five months, one person
The brand was running on Excel exports and platform dashboards, triangulating between what Meta, Google Ads and Shopify each reported. Its analytics tools cost more than $10,000 a month before anyone wrote a query, and nothing checked what the ad platforms were claiming.
of planned CTV budget reallocated, after the first $300k showed no incremental revenue
inflation in reported Google Ads conversions, found and corrected
of paid sessions turned out to be bots
less subscription churn from failed payments
What it found
CTV wasn't working. Incrementality testing and the media-mix model showed the first $300k of a roughly $1MM CTV plan returning almost nothing incremental. The remaining budget was reallocated.
Reported conversions were inflated 3-4x. A Google Ads tracking error was multiplying purchases, so bids had been rising against sales that never happened. Tracking was fixed and the history rebaselined.
More than half of paid sessions were bots. That distorted every channel comparison built on session data until paid was measured on real visitors.
The entry offer was buying worse customers. A cheaper first-order offer had improved conversion and CAC while lowering what those customers were worth over time. The offer went back up, and LTV:CAC improved.
Branded paid search was paying for free clicks. The brand was buying traffic it would have gotten from organic search, and the budget moved accordingly.
Retention
Failed payments were a marketing problem nobody was treating as one. Reworking the recovery sequence, including retry timing, the message at each step and pacing text messages against email, cut that churn by nearly half. Winback conversion improved 20%.
What got built
A warehouse unifying every core source, a Bayesian media-mix model, geo-holdout incrementality tests, first-party attribution and behavioral customer segmentation, plus a daily health scorecard in Slack and automated data-quality checks. One person built it in five months, and it cost about $4.50 a month to run.
A Brooklyn retailer: from no ecommerce to 20%+ of revenue
This one answers a different question: whether the approach works when there's no analytics function to improve, only a business. I co-founded a Brooklyn wine retailer that had no ecommerce and no marketing function. From 2023 to early 2026 I built both, hands-on.
year-over-year growth, two years running
of total revenue from ecommerce, starting from zero
return on ad spend on delivery marketplaces
I ran paid media on Meta and Google myself until both were contribution-positive, built the catalog syncing to UberEats, DoorDash, Grubhub and Instacart, and launched a loyalty program connecting the shop and the site.
Why these are described but not named
The findings belong to the businesses that paid for them. The numbers are real and the work is mine; the brand details stay out of public view.