Case study · Consumer membership marketplace · Google Ads, Meta and creative

Once they’d seen the ownership we took and the results we drove, handing us the creative was the obvious next step. The numbers have proven them right.

Chapter 1: Google, where we earned the trust

We started in August 2024 with search only. The audit found campaigns optimising on clicks and button presses, with no data on who actually became a member. We set up offline conversion tracking first, so Google could learn from accepted and activated members, not form fills. Then we scaled it.

In eleven months activated members from Google went from 24 a month to 81, and the share of Google registrations that became members nearly doubled, from 1.3% to 2.4%. By mid-2025 Google was producing half of all members on about a tenth of the paid budget.

Chapter 2: Meta, where the money was leaking

Meta was run by another team. In the first half of 2025 its spend more than doubled, to $1.08 million a month, and its cost per registration rose from $33 to $150. Each extra dollar was buying less.

After a year on Google, the client asked us to audit Meta, and we took it over in July 2025. We consolidated 70 campaigns into 7 so the algorithm had enough data to learn in each, then cut the spend that wasn’t turning into members.

In the quarter before we took over, the account spent $1.05 million a month and got 152 activated members at $6,890 each. In our first quarter it spent $512,000 a month and got 97 at $5,294 each. The $538,000 a month we cut had been buying members at roughly $9,700 each, nearly twice what the rest of the budget paid. From February to July 2026, cost per member averaged $5,927, still 14% below the quarter before we took over.

Chapter 3: Creative, replacing the agency

The ads still came from a dedicated creative agency. The work was well made and on brand, but it played it safe: small variations on whatever was already working, built around the same three themes. What the account needed was new angles to reach and convert new audiences, and those weren’t coming.

We didn’t wait to be asked. In September 2025 we shipped nine statics of our own, and one became the top-spending ad in the account within two weeks. In December the client told us they needed far more creative than the agency delivered. In January they ended the agency and handed creative to us.

From February to September 2026 we launched 23 new creatives a month, 20 of them new angles rather than versions of old ones. That’s 5.9 for every $100,000 of spend, against 2.4 in the agency’s final months.

We didn’t switch off the old ads. We ran ours next to them in the same campaigns and let Meta’s algorithm choose. Our creatives took 11% of spend in February and 89% by August, and brought in registrations 20% cheaper than the older ads they pushed out.

What made our creative work

Every concept started from one specific person and a hypothesis, built from the client’s best-performing search terms, community threads and competitor ads. Every batch carried safe concepts and wild ones, and a creative roadmap logged what was tested, why, and what we learned.

The breakthrough was the persona call-out. Generic ads kept reaching the same people, whatever the angle. Naming one specific kind of member in the hook reached people the account had never touched, at around half the account’s cost per registration. And because the same team ran media and creative, a test result became the next brief the same week.

The job isn’t to make each channel look good. It’s to put the next dollar wherever it buys a member.

Questions we get asked

Client name withheld. Activated members and cost per activated member come from the client’s back-end data, for members activated within 30 days of registering, across Google and Meta. Creative counts and cost per registration by creative come from every ad in the Meta account, November 2024 to September 2026, as reported by Meta; copies across ad sets, languages, formats and versions count as one creative, and a creative is new in the first month it spent $250 or more. USD.

Case study · B2B SaaS · Self-serve signup · Google Ads and Performance Max

The machines were doing a great job. The inputs underneath them didn’t match the business.

Wasn’t the model supposed to catch this?

Fair question, and the answer matters. The automation wasn’t the problem. An account this size can’t be run without it, and running paid media now without every model you can get is money left on the table.

You can see it working on the chart. Value-based bidding went live in August 2025 and cost per new customer fell from $424 to $130 by December. That was the client’s own move, before we arrived, and it was the right one.

But a model can only bid on the values it’s given. It can’t know when one of them is wrong, because the proof sits in back-end data it never sees. Someone has to bring that number to it.

The model wasn’t wrong. It was optimising accurately toward a value that didn’t match the business.

Who this is about

A B2B SaaS platform with a self-serve signup and a usage-based product. Someone signs up, connects their systems, makes a first purchase, and if the product fits they keep buying for years. That first purchase is where a signup becomes a real customer, so the account is judged on it alongside transaction volume and cohort user counts.

We took the account over in April 2026. ROAS on this page means first-month cohort revenue divided by acquisition spend, which is a leading indicator in a business where value accrues over years, not a payback figure.

What made the difference

We audited what the model had been told, not how it was tuned

The instinct with an underperforming automated account is to tune it. Adjust the targets, add another tool, restructure the campaigns. We went the other way and looked at the inputs.

Put the platform’s valuation next to the business’s own outcomes, segment by segment, and look for where the two disagree. That’s the method. We ran it across first purchases, transactions and cohort users, broken out every way the account could be split. In one dimension the two had been disagreeing for months. Here it was device. In another account it could be geography, landing page or time of day.

The fix was a value adjustment, not another tool

Mobile conversions were being priced as though they were worth what desktop ones were. The back-end numbers said otherwise, so we adjusted the value assigned to mobile and let the model reallocate on its own. No restructure, no new platform, no extra layer. One correction to a price the system had been carrying since before we arrived.

The model optimises. Someone still has to decide what it’s optimising toward.

Did that cost us volume?

No. July 2026 brought in roughly as many new customers as March 2026 did, on around 40% of the spend.

The June and July figures aren’t only better than April either. They’re the best in the whole nineteen-month series. Combined cost per new customer had never gone below $130 before, and that came in December and January, the peak of the season. June and July beat it in the quiet months.

Questions we get asked

Client name withheld at the client’s request. All figures are actual reported results for the periods shown. ROAS refers to first-month cohort revenue divided by acquisition spend and is used as a leading indicator, not a measure of payback.

Case study · EdTech · Meta · Performance creative

Nobody can pick the winning ad in advance, so we stopped trying to.

Isn’t 704 ad tests just spray and pray?

It’s the fair objection, and volume on its own would have produced nothing.

Most accounts fail in one of two directions. The first is over-investing in a handful of ads. Weeks of concepting, rounds of internal review, four beautiful launches. It wastes the opportunity, because nobody can identify the winning concept in advance. Not the strategist, not the designer, not the client. Limiting yourself to a few attempts is an expensive way to be confidently wrong.

The second is shipping everything to see what sticks. That produces the occasional winner and no knowledge, because without a stated hypothesis you can’t tell whether an ad won on the angle, the format, the offer or the timing. So you can’t do it again on purpose.

The split matters here. Of the 704 tests, 461 were new concepts or angles and 243 were variants of something already working. Two thirds of the output was a genuinely different idea rather than a resize or a colour change.

Nobody knows in advance what will win. That’s the argument for volume, and it’s also why every test needs a reason attached.

Who this is about

An EdTech business where the conversion that matters is an enrollment. Paid social hadn’t been used at any real scale before, so there was no library of proven ads, no history of which messaging worked, and nothing to build on.

What made this work

The design team sat under the performance team

One of the key factors behind volume, data driven hypotheses and fast testing is having the design under the performance team. Not a brand team operating on completely different KPIs. Not a creative agency that cares more about how the work looks than how it performs.

You need them working desk by desk, one Slack message apart. From the moment a data insight lands to the brief being written should take hours, not days or weeks.

At 27 ads a week, a week between reading the data and acting on it costs you a full round of tests.

Two senior people held the account every week, no rotation and no juniors. A performance lead who reads the whole funnel from CTR down to CAC and knows how the algorithm spends. A designer who turns the idea into something that does its job and builds statics and motion at test speed. Both answering to the same numbers, which in a lot of accounts is the whole difference.

Every hypothesis started from data

No spaghetti at the wall. Every concept started from something observed. Reddit threads and other external discussion where people talk about the decision in their own words. The client’s customer survey. The questionnaire applicants fill in when they apply, which is a large pile of first party language about what people want and what is stopping them.

Each of those sources produced several hypotheses rather than one, and we tested most of them. That is where a lot of the volume comes from. It is not 704 random ideas, it is a much smaller set of real signals worked through properly.

Every test got a postmortem

Results were read back against the hypothesis, not just against CPA. That’s the step that turns 704 launches into an accumulating body of knowledge instead of 704 separate events, and it’s the one almost everyone skips.

Questions we get asked

Client name withheld at the client’s request. All figures are actual reported results for the periods shown.