Wiring a bank's first program before claiming results

A community bank started working with us on its brand strategy. It had never run digital advertising. Once the strategy was approved, we proposed a single campaign to test the market: one month, up to three static ads, on Meta with an optional search layer. The bank approved it within two days.

The harder constraint was what the bank could see. Accounts open on a platform hosted by its core banking provider, not on its website, so nothing placed on the site could see an account being opened. Analytics had been pulled from the site earlier in the year and were restored days before the campaign started. Both sides went in knowing the first campaign would report what the ad platforms could see, and not much more.

So we separated what could be claimed from what would have to wait. The first campaign would report platform metrics only. The larger pilot that followed was set up awareness first, with its early numbers treated as a baseline and conversion as something to optimize toward once measurement was standing. Chasing sign-ups sooner would have meant optimizing against incomplete data.

We also defined what would count as a conversion before any existed: intent we could actually observe, like an application started, a call tapped or a lead form reached, rather than accounts opened, which no tag could see. Anything not yet measured would be reported as missing, not as zero.

Everything ran inside the rules banks advertise under. Meta places financial services in a special ad category that removes ZIP-code and narrow-radius targeting and lookalike audiences, so every market was drawn as a wide radius. The FDIC mark went into the frame of every ad at every size – the bank asked for it before its compliance officer confirmed it – and fee messaging carried the disclaimers compliance required. The pilot ran in English and Spanish.

The tracking had to be built without editing the website ourselves. The bank's site sits with its core banking provider, so we built the server-side tracking and consent setup, the bank's IT team placed it through the provider, and we published once they confirmed. The bank's chief information officer set the terms before any of it started, and it went live partway through the pilot.

The first campaign ran about two months, extended twice by the bank. Its cost per landing-page visit fell from the first half of the flight to the second. By the numbers the platforms could see, it was working.

What it couldn't show is whether any of it opened an account. Roughly nine-tenths of its spend ran before the server-side tracking went live, and account opening happens on a platform no tag can see. The platforms' conversion counts couldn't observe an account being opened, so their silence proves nothing either way. This page doesn't claim deposits, accounts or cost per customer, because the program can't yet stand behind them.

When the flight was due to end, we recommended keeping it running for another month, the first full month of post-click data. The bank agreed, to get meaningful data from it.

We'd keep defining a conversion before one exists, and reporting what's missing as missing. And we'd keep letting the client set a report's altitude. Asked for something to share with its board, the bank wanted a bird's-eye view of what had been done and what the goals were, which was the right call for data that short.

We'd watch rules that travel past their scope – a landing-page rule meant for one kind of ad ended up sending business-facing ads to personal-account pages – and campaigns tied to an occasion, which need an end date. One event-driven push kept spending after the event.