Rebuilding a growth engine for a recovery-footwear brand

Nothing was wrong with the product. The brand was an early name in recovery footwear and sat in a category whose aggregate search demand grew about 22% in a year. Competitors tracked that growth. The brand diverged downward. Its new-customer base had been shrinking for more than two years.

The problem was the engine, and a six-week paid Opportunity Analysis found it in the account structure:

• Roughly 70% of Google search budget sat on brand terms, paying for conversions that were already going to happen.
• A seven-figure share of Meta budget was landing on existing customers at ad frequency 46+, the same people, over and over.
• Performance Max campaigns intended as prospecting had no audience exclusions, no audience signals, and were not using Google's net-new-customer optimization. Campaigns nominally built for growth were structurally unable to pursue it.
• Linear TV reached as much as 41% of the peak-season media mix, unmeasured, with its heaviest months landing in the year's lowest-demand windows.

Every paid channel scored 5-6 out of 10 against best practice. Web analytics, the strongest of them, scored 7.

The starting position: half the media budget was being spent talking to people who had already bought, and the brand was acquiring fewer new customers each year in a market that was expanding.

The diagnosis was that the brand was not losing to demand, it was losing execution. What we committed to was specific and time-boxed: lift channel execution into the 7-8 range within three to six months, expect a one-time 20-30% gain from cleanup alone, and hold a hard no-regression guardrail throughout.

There was no attempt to optimise the existing structure. In April it was rebuilt around a single thesis: concentrate spend, fund prospecting, prove every channel.

Meta went from more than ten active campaigns to five, heading to four, so the algorithm had enough liquidity per campaign to learn. All reservational buying was switched off in favour of auction. Existing-customer share of spend fell from over 25% to under 18% within the quarter. More than half of June's budget moved into dynamic product ads, which ran a 7% lower cost per acquisition than the account average.

Google was consolidated out of fragmented, thinly-funded campaigns into a structure organised by seasonality and funnel stage. Daily budget caps came off entirely, replaced by bid caps and ROAS targets, so spend was governed by economics rather than by arbitrary ceilings.

TikTok moved its entire budget to new customers and cut the 55+ cohort from over 30% to 21%. Programmatic was stood up beyond YouTube-only display for the first time.

In one quarter, prospecting share went from 50% to 78%, and it never went back.

The cost of that was real and it showed up immediately: cost per new customer rose 7% above the starting quarter, and new-customer volume was still down 11% year over year. That is the true shape of an acquisition-first rebuild: efficiency gets worse before it gets better, because you are buying strangers instead of harvesting your own customer list.

But the trajectory had already turned. The year-over-year new-customer gap closed from −54% to −11% in a single quarter, and by June it had crossed into positive territory. The cost was named in advance and the client funded it. That decision is what made the following quarter possible.

August is where the rebuild proved itself, and it proved itself harder than either side had planned for.

Two things landed at once: an end-of-season promotion, and a prospecting engine that was now four months into running at scale. Neither explains the result on its own. The promotion created the demand; the rebuilt engine converted it at a volume the old structure could not have reached, with almost no retargeting pool left to lean on. It is the intersection of the two that produced the quarter.

Demand ran ahead of what the operation was planned for. Paid spend was deliberately throttled and email and SMS were paused, with media and operations working in close coordination to keep fulfilment inside its capacity and protect the customer experience.

Even under that throttle, August delivered new customers up 117% year over year at a cost per new customer 27% below the prior August, the like-for-like comparison, promotional period against promotional period.

The full quarter closed with new customers up 54%, the highest quarterly acquisition volume of the engagement, with prospecting at 86% of spend and acquisition cost down 21% against the starting quarter.

The durable value was not the month itself but what it taught both teams about the ceiling. Forecasting and planning were rebuilt around the new media mix, and every promotional period that followed (Black Friday, January, May) was run with materially better predictability and no repeat of the capacity squeeze. Knowing where the limit sits is what makes it safe to push toward it.

The more consequential development that quarter was quieter. Independent geo-holdout incrementality testing went live, at a cadence the measurement partner flagged as unusually aggressive for its client base: two tests running at a time, essentially without breaks.

The first hard decision arrived almost immediately, and it was the best-performing channel in the account that had to go.

That single decision changed how budget was argued for the rest of the engagement. Platform-reported performance stopped being the deciding vote.

Q4 was the stress test: Black Friday and Cyber Monday, a compressed ramp window, heavier competitive promotional pressure, and genuine month-to-month budget volatility.

The engine held. Prospecting sat at 94% of spend, and every month delivered year-over-year new-customer growth:

• October, with media investment down about 20% year over year: +12% new customers, acquisition cost −28%.
• November: +34% new customers, revenue +6%.
• December: +25% new customers.

The Black Friday play was deliberately counter-intuitive. Rather than concentrating budget on promotional days, spend was ramped into the build-up, accepting a temporary efficiency hit to build funnel and awareness ahead of the offer. Acquisition cost climbed to 38% above the starting quarter, the highest point of the engagement, before the promotion launched. On launch it collapsed to nearly half that, and the month closed comfortably below the starting-quarter baseline.

By quarter end, nine geo-holdout tests were complete, enough to compare channels against each other and across high- and low-demand periods. Budget moved on those reads within weeks rather than at the next planning cycle (section 6 sets out what each test changed). Performance Max was also rebuilt in the quarter, from product-type asset groups into persona asset groups: nurses and healthcare workers, athletes and active recovery, runners, all-day comfort, travel.

Q4 also surfaced the next constraint. Two Black Week ads ran on the same platform, in the same window, at the same budget, and returned materially different cost per acquisition, click-through and return. With media, targeting and timing held constant, creative was the only variable left. That forced creative work to stop hunting for individual winning ads and become a system: Audience × Angle × Product: who it is for, what problem it speaks to, which product solves it.

The quarter closed with new customers up 25%, with acquisition cost 31% below the starting quarter.

Q1 is seasonally the brand's lowest-volume quarter. That is precisely what makes it the strongest proof point: efficiency in a thin quarter is a property of the engine, not of the season.

• New customers up 78% year over year, nearly double the volume of the same quarter a year earlier, the engagement's trough.
• Cost per new customer 33% below the starting quarter, the lowest of the engagement.
• January alone: new customers up 120%, at the cheapest acquisition cost of any month in the engagement, 41% below the starting-quarter baseline.
• Media investment up 18% year over year. At the prior year's cost structure, the same volume of customers would have cost materially more.
• Prospecting 95%.

Meta and Google ran opposite plays in the same quarter and both worked. Meta scaled +59% as the primary acquisition engine. Google budget was cut 25% and still grew revenue, with platform-reported return reaching its highest level of the engagement. Bing was taken to zero on its incrementality read, and reopened later when demand shifted. Concentration, not addition.

The constraint Q4 exposed also got its answer this quarter. If creative volume was the ceiling on how far the engine could scale, the fix had to be a supply line rather than a better brief. A creator programme was built to feed it: sourcing shifted from a single external vendor to a directly-managed roster, deliberately weighted toward a large number of small creators rather than a handful of large ones, because the algorithm rewards genuinely distinct executions over variations on a theme. The binding constraints turned out to be contracting and payment onboarding rather than creative quality, so those were the processes that got rebuilt.

By this point, fourteen completed geo-holdout tests sat behind these decisions. Every channel in the mix, and every channel out of it, had evidence attached.

Platform-reported performance tells you what a channel claims. It does not tell you what would have happened without it. From the third quarter onward, every meaningful channel in the mix was put through independent geo-holdout testing, at a cadence the measurement partner flagged as unusually aggressive for its client base.

The point was never the tests themselves. It was that each read had a decision attached to it, taken within weeks, and that a poor read cost a channel its budget regardless of how good its own reporting looked.

That last row matters as much as the first. A channel failing a test was not a verdict on the channel, it was a verdict on how it was being run. Several were rebuilt and put back through testing rather than written off, and the ones that improved earned their budget back on evidence.

Because the brand sells through Amazon and wholesale as well as direct, reads were assessed beyond the direct-to-consumer line wherever the data allowed it. The clearest case was linear TV. The concern about cutting it was that retail footfall depended on the awareness it bought, and that pulling it would show up in stores. It did not. That said, cross-channel effects are the hardest thing in this discipline to prove cleanly, and the programme treats them as directional rather than settled.

What changed between the quarter before the engagement and the quarter that closed it.

Total investment grew. Its composition changed entirely, from broad-reach and retargeting toward concentrated, measured prospecting.

Two markets, not one

The same engine ran the Canadian business alongside the US one, on its own targets and its own efficiency ceiling rather than as an afterthought bolted onto US campaigns. Canadian revenue grew 79% year over year over the same period, comfortably outpacing the growth rate of the much larger US business.

Three measures, all indexed to the starting quarter at 100, so the shape of the trade is visible in one view.

All three measures indexed, with year-over-year change alongside:

Full year over the prior year:

Name the cost before you spend it. The rebuild-quarter efficiency dip was forecast, funded, and explained in advance, not rationalised afterwards. That is the only reason the client stayed in the trade long enough for the third quarter to arrive.

Let measurement outrank the platform. A 6.0× platform-reported return against a 0.23× measured incremental read was the test case. Pausing the best-looking channel in the account, on evidence, is what made every subsequent budget argument credible.

Concentrate rather than spread. Ten-plus Meta campaigns to four. Budget caps replaced by return targets. Bing to zero. Google cut a quarter while growing revenue. Nearly every gain came from removing something.

Efficiency and scale are sequenced, not opposed. The final quarter invested 18% more than the prior year and acquired 78% more customers. That gap is the whole return on the rebuild.