CommonWealth One FCU
Google Ads Audit & Approval Sheet

Every keyword runs on broad match, without the guardrails broad match needs

All 44 active keywords ($24,199 of 90-day spend) are broad match, including the Brand campaign, and this has been the standing design since the current campaigns were created on March 31, 2026 (not a recent recovery measure; no phrase or exact keyword has spent a cent in any month). Broad match can work well in 2026, but only with two guardrails this account does not have: a strong negative-keyword net, and bidding pointed at outcomes that matter. Without them, the account buys subprime auto-lender searches, prepaid debit brands ("deep blue debit card number" cost $131 for one click), competitor banks, and "i need a loan right now" queries, and the bidding algorithm learns from every one of those clicks.

Where the wasted money actually went. We classified every search term that spent money in the last 90 days. These are the buckets, with real searches the account paid for:

None of these people can become members or qualified borrowers. Worse, some of them CALL (mistaking the ad for their own bank's service line), the call counts as a conversion, and the bidding algorithm learns to buy more searches like theirs.

Why it may feel recent: after the June pause, budgets rose and the efficient Performance Max campaigns were paused, so spend through broad search nearly doubled while the reset algorithm bought worse traffic with it. The pain got much louder in June and July; the match-type choice itself dates to March.

The negative-keyword net: a real negative list exists and has been attached since April 1, but the members that would actually block the waste above were only added July 21-22 (confirmed from both the change log and serving data). Even now the attached lists cover only 32% of the identified waste pattern. Meanwhile the best-built negative list in the account (220 competitor and rate-shopping terms) is attached only to an empty test campaign that spends nothing.

The June billing pause reset the bidding algorithm, and it is still recovering

The account served zero ads June 9 through 14 after a billing interruption, and the pause reset Smart Bidding's learned state. Because the account's conversion definitions changed several times this year, we measure the effect on a signal whose definition never changed: phone calls from ads. Cost per call had been stable around $25 for ten weeks. After the restart it tripled, and six weeks later it is still roughly 55% above baseline. Going forward this class of problem simply does not happen on our watch: billing alerts and a daily delivery check are part of our standing operating routine, so a quiet interruption cannot run for days.

One more signal worth knowing: cost per submitted loan application has deteriorated all year ($79 in April, $109 before the June pause, $355 in late July) and unlike calls it is not recovering. Part of that may be the website tracking decay in item B5 rather than real performance, which is exactly why B5 needs to run: right now nobody can say how many applications the ads actually produce.

Bidding is optimizing for phone calls, not loans

Of 1,129 conversions feeding the bidding algorithm over 90 days: 895 are phone calls, 185 are submitted applications, 25 approved, and just 24 funded loans. Worse, the per-campaign goal settings (changed July 2) mean the Personal Loan, Car, and Kasasa campaigns do not count submitted applications at all: the Personal Loan campaign recorded 39 applications in 90 days and zero of them informed its bidding. Brand is technically the only campaign optimizing toward applications, and that is not a point in Brand's favor: brand campaigns should not be on conversion bidding at all (see item A5), which makes fixing the loan campaigns' goals the more urgent. And the wasted clicks are converting: people searching another bank's customer-service line call the tracked number by mistake, that misdial counts as a conversion, and the algorithm learns to buy more of those searches. Redirecting bidding toward applications and funded loans is the single highest-leverage change in this account.

Conversion tracking is 49 actions deep and partly broken

Duplicates across three generations of tracking (Universal Analytics goals, GA4 imports, webpage tags). The "Click on Apply Now" tag fired zero times in 90 days. The submitted-application count is collapsing month over month: 127 in April, 95 in May, 51 in June, 17 so far in July, which needs a root cause. The good news: a genuine offline upload loop exists for Approved and Funded loans, so the foundation for real outcome tracking is partly built. The hardcoded analytics parameters on cofcu.org's apply links threaten the click IDs that loop depends on, and fixing them is a site-side change we will coordinate.

Budget is allocated backwards

Brand search converts at $12.08 and loses 49% of its available impressions to budget caps. Kasasa Checking search converts at $88.93 (with a $10.25 CPC) and keeps spending. Meanwhile the Performance Max campaigns, which were the most efficient tier in the account ($15 to $22 per conversion), have been paused since mid-June, so that budget moved onto more expensive broad search. Cheap demand is being capped to fund expensive junk.