Built, shipped, still running.
Every case on this page is a system that exists — in production, instrumented, measurable. And every number was read from the running system or the primary study before it was published here. No projections, no composites, no "up to." Where a number couldn't be verified, the space is empty until it can be.
By capability
- Reporting you can defend
We build reporting where each figure traces to a query or a source document we can re-read, so anyone with access can check it and get the same answer.
- Systems that read the business
We build systems that read a business's own data sources directly and act on them, with every action logged so the result can be counted.
Inventory-driven advertising · Auto dealer group
Advertising that reads the inventory
SITUATION
A multi-location dealer group whose lot changed every day — and whose advertising didn't. Content went out when someone had time, written from whatever came to mind, while the one data source that always knew what needed selling sat unread: the inventory feed itself.
WHERE THE LEVERAGE WAS
The feed already knew the thing worth saying. Nothing was reading it.
WHAT WE BUILT
A system that generates advertising and organic content directly from live inventory — what's on the lot decides what runs, automatically. The copy discipline is encoded rather than left to chance: lead with the reason a buyer cares, never the spec sheet.
WHAT IT DOES NOW
As of July 2026 it ran at thirteen dealership locations, four inventory-driven posts a day, every day, no hands. A later and larger study, on the platform's own reported numbers through 3 August 2026, found engagement below one percent for every hook type, while the boosted posts drew far more link clicks than the organic ones. We retired the earlier hook-copy figure and say so on this site. The study attributes no sales to any post.
AI content engine · Multi-location retail group
Judgment at content scale
SITUATION
Daily content across a dozen-plus locations and three platforms is a full-time job nobody was hired for. The usual choices: add headcount, lower the bar, or go dark. The operator's quality standards were real — they just couldn't be in thirteen places at once.
WHERE THE LEVERAGE WAS
The standards existed. They needed to operate without the operator.
WHAT WE BUILT
A content engine with those standards encoded — what a good post looks like, what never ships, how each location sounds. Every post is generated, judged against the standards, and scheduled automatically; every one is logged, so the system's output is auditable down to the individual post. That ledger is where these numbers come from.
WHAT IT DOES NOW
Over 1,100 posts published across 13 locations and three platforms in about fifteen weeks — roughly 400 a month, four a day per location, with the pace holding: 159 posts in the seven days to 5 July 2026.
The numbers we took back
Closed-loop paid media · a three-location golf-car dealer in a large retirement community
"Thirty-three website sales is a floor." It was not.
An early draft called 33 website-attributed sales a floor. The four matching rules give 24 to 34 on the 1,361 unit deals in the window, so a floor it was not. We retracted the sentence; the dealer's sales system governs the count.
POS intelligence · one location of a 13-location restaurant group across the Mountain West
"2.4 times a median Saturday" recomputes to 2.55
One location's Valentine's Day 2026 ran $25,858.48 net on 271 orders, a $95.42 average check. We had written "2.4 times a median Saturday". The median of the other 25 Saturdays in the 181-day window is $10,123.63, which makes the ratio 2.55. The underlying figures stand; the ratio is corrected.
Consolidated reporting · a marketing agency's data and dashboarding platform for entertainment-venue and sports-property paid media
The spend rule that zeroed multi-channel clients
A spend-calculation rule applied at the wrong data layer made some multi-channel clients' reported spend collapse to about zero. The change was rolled back for the affected clients on 1 June 2026; the rollout rule now reads: per client, with a before-and-after check.
Inventory-driven advertising · a regional auto dealer group
"Weekend Ready posts at 41% engagement", withdrawn
The April 2026 study reported a "Weekend Ready" hook at 41.1% average engagement, on six posts. The later study, dated 3 August 2026, withdrew the figure as a platform data artifact. The same August study led us to retire the hook-copy finding as well; see 'The hook-copy figure, retired'.
Inventory-driven advertising · a regional auto dealer group
The hook-copy figure, retired
The April 2026 study, 265 posts across six stores on third-party engagement data, reported hook-first copy at 4.88% against 0 to 0.2% for spec-sheet copy. The August 2026 study, on the platform's own numbers, found engagement below one percent for every hook type and called the April data sparse and unreliable. We retired the figure from this site.
Why there are no logos here
The systems are real; the names are withheld. Client data belongs to clients, and a logo wall is a cheap way to borrow credibility we'd rather earn. So the trade we make is this: anonymized cases, exact numbers, and provenance for every figure — what was measured, where, and when. If you want to see for yourself, ask in a working session and we'll pull up a live system, not a slide about one.
Find your pivot.
One working session. Bring the problem that keeps resurfacing. Leave with ground truth on where your leverage actually is — and if it helps, we'll show you systems like these running live.
