A multi-location garage services company, storage and organization systems, across four major U.S. metropolitan markets, engaged January 2026. Roughly $250K in H1 media managed across Google and Meta.
An established multi-location advertiser spending six figures a quarter with call tracking already in place. The trouble sat one layer up.
They could not tell what their previous marketing companies had actually been doing, and the numbers those companies reported did not survive contact. Prior campaigns were built around keywords with no connection to buying intent, and the conversion goals counted duplicates, so the reported cost per lead looked far lower than the true one. The business was steering off instruments that read wrong.
Once we cleaned the lead flow, deduplicated conversion counting, and rebuilt tracking so every source reported honestly, the marketing measurement problem was solved. What that exposed was the layer behind it: a sales operation and company structure that never picked the leads up. 2,739 leads reached the client's own CallRail in six months; 5 were marked qualified, and three systems (the ad platform, the CRM, and call tracking) each held a different lead count with no reconciliation between them. The constraint was never lead volume. It was everything after the lead.
The company's CRM had built-in call tracking. It had never been used. The prior marketing operation, run under the same fractional-CMO layer, marketed itself as specialist in that exact platform, and the platform's most basic measurement feature sat switched off for the entire relationship. When we cleaned the phone tracking and the attribution, years of broken automations surfaced at once, and the fix was blamed for the mess it revealed. That is the pattern: when measurement finally works, it exposes everything built on top of the old numbers, and the messenger takes the hit.
Nothing on the accounts was touched for the first two weeks. Then five moves ran in sequence so every later result could be attributed to a specific decision, not to the sum.
Merged five market-siloed websites into one multi-location site, pooling five domains' worth of fragmented organic authority into a single domain that could actually rank.
Same logic on Google Ads: unified the market-siloed account structure so the platform sees the whole business's conversion data, not five starved slices of it.
As the accounts stabilised, we built exact-match conversion actions across Google, Meta, and Google Business Profile, one definition of a lead, counted once, everywhere. This is what made every number on this page comparable.
With clean measurement in place, we ran a phased Meta experiment: 12 days on lookalike audiences alone as the control, then first-party customer data layered in, so any change in cost per lead could be attributed to the data, not to luck.
Did not raise total budget to chase volume. Did not adopt the weekly evaluation cadence or the 2021 baseline. Did not take over the client's internal lead qualification, which sat outside the ad accounts and turned out to be the binding constraint.
The account the way an auditor reads it: every stage from search to conversion, last 90 days against the prior 90, all four markets, enabled and paused campaigns included so nothing is cherry-picked.
Reading the chain: more people searched, we captured a slightly larger share, bought clicks 7% cheaper at a higher click-through rate, and converted them 26% more efficiently into 67% more leads. Cost rose 23% by design, to fund the added volume. Every controllable metric improved. The one zero on the board is the number only the client's own systems could supply, and it never arrived.
Leads up 35% to 123% in every market. Cost per conversion fell in three of four. In Metropolitan 4 spend nearly doubled to capture a 135% surge in search demand, and cost per conversion held within 5% while leads rose 89%. Buying 89% more leads at roughly flat unit cost during a demand spike is a win, and it is disclosed as the trade it was.
Preview outline below; each ships in the same rhythm as the sections above, using the numbers already verified on Dimi's mockup.
This case study is one account. Yours will look different. Same method: read the raw account before touching a lever, run the diagnostic chain end-to-end, structure a controlled test, ship it, and hold the reporting layer honest to what the platforms actually reported.
45 minutes. No deck. No pitch. If we cannot find the money in your account, we say so on the call.
