Google Ads review · published August 21, 2026 · charts and headline figures updated September 18, data through September 17
Cost per completed app
$226 → $127
-44% July vs August, by application date
Brand cost per click
$0.72 → $0.14
-81% after the Jul 27 cap
July junk search spend
47%
of non-brand $ went to searches now blocked
Negative list
969
terms, ten rounds since Jul 27, reviewed daily
The one thing that unlocks everything else
The landing pages need to be fixed.
The search traffic is now clean; the page experience is the constraint. All 20 of August's tracked completed applications came from brand clicks; non-brand's 716 clicks produced zero, consistent with cold shoppers meeting a 285-link page with no rates and leaving. The rebuilt pages, files, and setup guide have been ready since July 28 and can be implemented as-is. Two requests: an implementation date from the web team, and viewer access to the GA4 property so engagement on these pages can be measured directly.
Daily spend: deliberate descent, a stable cap, then the scale-up
Total account spend per day, July 1 to September 17. Hover for values. July ran $12.2K; August closed at $8.5K by design; September is pacing about $12.3K with the growth budgets in place (Personal Loans $4K, Auto $6K, Credit Cards $1.5K).
Brand: half the story is price, the other half is coverage
Weekly windows through September 17, two panels, one metric each. The Jul 27 manual cap ended automated bidding's $0.72 average clicks; coverage roughly doubled while cost collapsed to ~$20/day, and both have held through September. Verified this morning: 99.8% of August brand-campaign clicks were true brand searches (commonwealth / cofcu variants; the remainder is misspellings of the name).
Brand campaign
What the money bought: July's worst offenders are now impossible
Real July spend on searches that the list now blocks (969 terms as of September 11). Hover any bar. Outcomes follow searches on a lag; this is the leading indicator of the applicant quality the CRM team wants to see.
July search
Spend
Category
i need a loan right now
$129
desperate intent
credit acceptance
$65
subprime auto lender
onity mortgage
$60
wrong institution
americredit
$46
subprime auto lender
used cars with zero down payment
$44
desperate intent
driveway finance
$43
wrong institution
credit one bank español
$42
wrong institution
fidelity bank customer service
$40
wrong institution
westlake financial customer service
$37
subprime auto lender
title loans near me
$31
title lending
suntrust
$30
wrong institution
Non-brand efficiency by month
Cost per tracked conversion by campaign. What "conversion" means here: for the non-brand campaigns this is entirely the Calls from ads action (connected calls), because the application/approved/funded imports have credited almost exclusively to Brand. This is the proxy metric the account came with, useful for direction, not the destination: the plan is to move optimization and reporting to product-split outcomes (deposits vs loans, application and funded-based) as that tracking lands. September note: Kasasa Checking was rebuilt to phrase match on September 11 and its target CPA was removed on September 18 to let it relearn, so its September figure is a small-sample relearning number, not a trend.
JulyAugustSep 1-17
Where funded loans have actually come from (trailing 12 months, as of August 21)
Every funded and approved outcome imported into Google Ads, by campaign. Brand is two-thirds of proven production at the campaign level, and the true gap is wider: at the search-term level, branded searches also produced funded loans inside the non-brand campaigns ("commonwealth one federal credit union" converted in Personal Loan, "www cofcu org" in Car). Counting by what people actually searched, roughly 84% of visible funded outcomes trace to branded searches. Brand now costs ~$400/month.
BrandNon-brand
Reading the outcomes, and our recommendations
Status as of September 18
The recommendations below are as written on August 21. Since then: the three budget increases are live (Personal Loans Aug 26; Auto Sep 4, then to $6,000 on Sep 16 at the lending team's request; Credit Cards Sep 18). The deposits-versus-loans conversion split is still open with Alpharank. The rebuilt landing pages (three loan pages plus the new Visa Signature Rewards page) are with the web team. The brand incrementality test has not been scheduled.
Reading August approved / funded numbers
July imports finished at 17 approved / 16 funded. August reads 5 / 4 so far because outcomes lag applications by weeks and imports arrive on a rolling basis: funded moved from 2 to 4 just between the Aug 19 and Aug 20 data pulls. August is a month still being written, not one that ended at zero.
One thing to confirm together: the deposits question
July's 16 "funded" included roughly 10 deposit accounts; loans-only, July was about 4. We flagged early on that deposits and loan applications were mixing in the conversion feed (loans are worth thousands, deposits hundreds). Before comparing July to August, we should confirm together whether the import definition changed recently so deposits no longer pass back as funded. If it did, August's 4 funded is loans-only and roughly level with July's true loan count. Either way, the fix we recommend is the same: separate conversion actions for deposits vs loans, so each campaign optimizes toward its own outcome and every report reads clean.
The requested budget increases: our recommendation
Personal Loans to $4,000: yes, starting week of 8/24. It is the strongest non-brand campaign and has improved through the cleanup ($32 to $27 per tracked conversion).
Auto to $5,000: yes. The waste that made Car the laggard is exactly what this week's negative keyword batches removed, so the new budget lands on a cleaner match-mix than the trailing number reflects. We will route the increase through tightly themed rate-focused ad groups and review cost per conversion weekly.
Credit Cards at $1,500: yes, as a new build. We confirmed there has never been a card campaign in this account, so this is net-new. We will build it this week and launch it with its own conversion tracking, so card applications are measured as card applications rather than blending into the existing mixed conversion setup.
On brand, and the existing-member question
Existing members funding loans is the model working. A credit union grows by serving its members across products; an existing checking member taking a vehicle loan is precisely the outcome Consumer Lending is tasked with. In every lead-generation account we operate, a meaningful share of paid conversions are existing customers of another product.
The member tag says who they are, not why they came. What the data shows: they searched, clicked, and applied. Members comparison-shop; the account's own search terms are full of other lenders and rival credit unions. The brand ad defends the loan at the rate-shopping moment.
The auction does not pause when we do. If brand ads stop, the credit union will often not appear first for searches of its own name; competitors and aggregators bid on institution names, not every member is loyal, and the rate will not always be the best. A member in the shopping phase sees a rival first. How many applications that would cost is genuinely hard to estimate, which is why we would rather measure it than debate it.
The scale of the decision: across the trailing 12 months, roughly 84% of the funded outcomes visible at the search-term level trace to branded searches, and brand now costs about $400 a month at $0.14 clicks.
Our proposal: a matched-zip brand incrementality test, designed this week: pause brand in a matched set of zips for a few weeks and compare total applications. That answers the "would they have come anyway" question with data instead of assumptions. Alongside it, we fully support the month-end framework from the CRM side: cost per approved and funded loan, brand vs non-brand, deposits excluded from loan KPIs, and adding the GCLID to the weekly export so applicant quality ties back to exact keywords.
Landing pages: the biggest lever still on the table
Status check (live, this morning): the loan pages still run the full site experience: 285 links and the complete mega-menu on the vehicle-loans page, no rates shown anywhere on the loan path, and vehicle-type cards that navigate to separate pages. The rebuilt page mock-ups, files, and setup guide delivered July 28 are ready to implement as-is.
What four months of this account's own data shows: non-brand completed applications ran 24 in May, 12 in June, then zero in July and zero in August (brand: 72, 39, 30, 20). The May-June non-brand applications trace, term by term, to searches like "fha loan with 500 credit score," "guaranteed car finance no credit check no deposit," and other-lender names, exactly the applicant quality the CRM data flagged, and exactly what the negative keyword program has since blocked, deliberately. The clean traffic arriving now meets a busy corporate page with no rates and completes nothing; its only tracked outcome is phone calls. It looks like the current pages work for people who already know the credit union and lose cold shoppers, which makes the pages the single biggest constraint on non-brand performance.
The good searches are already arriving, and they aren't applying. August's clean non-brand spend (~$974, 205 clicks) includes exactly the demand everyone wants: "personal loan," "best online personal loans," "looking for a personal loan," "federal credit union near me," "credit union springfield va," "i want a credit card." The outcome so far: phone calls, zero completed applications.
The trailing-12-month check we ran on funded loans makes the same point: every non-brand funded outcome on record traces to a wrong-lender or instant-approval subprime search, the kind of traffic that has since been deliberately blocked. A clean, high-intent non-brand search has not yet produced a funded loan in this account. We read that as a page problem, not a demand problem.
One measurement gap to close together, the phone path: the calls those ads generate carry no click ID, so if a caller completes an application over the phone, it enters the CRM under whatever source intake assigns, not Google Ads. That means our "zero applications" figure counts online applications and may understate what callers complete by phone, and it also means paid search is structurally under-credited on exactly the path the current pages push people toward. Worth confirming: how phone applications are recorded and sourced in MeridianLink/Alpharank, and whether trackable numbers can be placed for ad traffic so call-driven applications get attributed.
Our honest assessment: even as the account seasons further, we do not believe non-brand reaches an efficient cost per funded loan on the current pages. The requested budget increases scale exactly this cold traffic. Negative keywords fix who arrives; only the landing pages fix what happens after they arrive. We'd like to leave the call with an implementation date from the web team.
Search-quality program and next steps
Negative keyword list built from zero on July 27 to 693 terms, three rounds this week alone. 47% of July's non-brand spend went to searches that are now blocked; outcomes follow searches on a lag.
Deposits-vs-loans conversion split: schedule it now; it also gates the credit card launch.
GCLID column added to the weekly CRM export (column spec available today).
Standing items: Alpharank/AdSync scope, Meridian Link page access, Apply-link site edit.
Prepared August 21, 2026 · sources: Google Ads (live), imported CRM outcomes