Google Ads for banks and credit unions in Singapore

Pull last month’s Search Terms report for a neighbourhood credit union in Jurong East and you will usually see the same mess: “best savings account Singapore”, “OCBC 360 vs UOB One”, “personal loan calculator”, and a handful of brand misspellings. Clicks look respectable. Funded accounts do not. The gap is not “we need more budget”. It is that Google is treating a comparison shopper, a rate tourist, and a member who is ready to open a SGD savings account as the same conversion event.

Singapore Search is expensive, MAS-aware copy is slow to approve, and smart bidding will happily optimise to form fills if that is the only signal you give it. WordStream’s 2024 benchmarks put median ROAS across industries at roughly 2x on revenue, with top-quartile accounts reaching 4x+, but that number is meaningless if your conversion is a PDF download. Singapore financial-services CPCs for product terms (personal loan, home loan) regularly exceed SGD 10-20, which makes negative hygiene worth more than most bid adjustments, and 20-30% of SEM budgets still leak on irrelevant queries, neglected negatives, or landing pages that do not match the ad. This piece is the operating version: what Singapore banks and credit unions should actually buy, how to structure the account so compliance does not freeze spend every fortnight, and how to measure funded products instead of vanity leads.

The queries that look like demand and never fund an account

Singapore searchers mix English, Singlish, and product nicknames. They type “CPF SA vs bank FD”, “HDB loan vs bank loan 2026”, “credit card 4 miles per dollar”, and “which bank for foreigner”. Those queries are not equally valuable. A credit union with a common-bond membership rule cannot profitably buy “best credit card Singapore” at the same CPA target as “join credit union Tampines”. A digital bank chasing salary crediting should not share a campaign with a mortgage specialist buying “refinance HDB loan”.

You will feel this first in CPC, not in the CRM. High-intent product terms (personal loan, home loan, high-interest savings) sit in auctions with DBS, OCBC, UOB, and the digital banks. Comparison intent inflates CPC without moving funded accounts. Automated bidding then does what it is designed to do: it finds more of whatever you labelled a conversion. If that label is “lead form”, you get more forms from people comparing 12-month FD rates they will never park with you.

Mobile is most of the click volume (63%+ of paid search clicks globally), and conversion still lags desktop by a wide margin. That lag is worse in banking because the form asks for NRIC-adjacent fields, income bands, and product eligibility. A Jurong West commuter on 4G will abandon a 14-field application that a desktop user in Raffles Place completes. If you only look at blended conversion rate, you will cut the campaign that was actually producing salary-crediting applications from office Wi-Fi.

Another leak: AI Overviews now sit on a large share of informational queries and cut CTR hard. People still research “best savings account” without clicking. You cannot out-bid an Overview on a research query. You can stop paying for it. The practical split is research versus transact. Research belongs in organic content and brand. Transact (apply, open, refinance, join) belongs in Search with tight match, dedicated landers, and negatives that strip the comparison layer.

If your team is still sending every ad to the homepage, stop. Homepage traffic is the fastest way to train bidding on the wrong people. Dedicated pages aligned to the query cluster are the minimum. If you want a second pair of eyes on structure before you raise spend, HeyLead’s Google Ads programmes are built around that split, not around “more impressions”.

How a Singapore bank Search account should actually be built

Fewer campaigns, clearer product clusters, and ruthless negatives beat a 40-campaign tree that nobody can approve. Google still needs keywords even as delivery leans on signals. Broad plus garbage landing pages will spend. Broad plus a funded-account conversion and a page that names the product, the eligibility, and the MAS-required caveats can work. Performance Max can take a share once Search is clean. It should not be the first thing you turn on while tracking is still counting PDF clicks.

Compliance is the real bottleneck in Singapore, not bid strategy. Legal will not let you claim “highest interest” without a footnote. They will not let you omit “terms apply” on a cashback card. If every RSA change waits three weeks, you cannot test. MAS Notice MAS 306 and the Guidelines on Fair Dealing require that any rate or return claim be accurate as of the date displayed and not mislead by omission - that is the legal basis for the footnote and terms-apply requirements your compliance team is enforcing. Build a pre-approved copy bank: headlines that name the product and the audience (members, salaried, foreigners with valid pass), descriptions that stay inside approved claims, and sitelinks that go to calculators, branch locator, and eligibility, not to the brand film.

Where Singapore banking ads lose the lead after the click

The auction is only half the CPA. A savings ad that promises “open in minutes” and lands on a 9MB homepage with a carousel of CSR awards will not fund accounts. Only 55.9% of origins pass all three Core Web Vitals. Banking sites are often in the failing half because of chat widgets, tag managers, and PDF rate sheets. Interaction to Next Paint on the apply button is the metric that matches the member’s patience, not a lab Lighthouse score from last year.

Match the page to the query. “Open SGD savings account” should hit a page that states eligibility (Singaporean, PR, valid pass), the interest structure without illegal superlatives, documents required, and a short form or deep link into the secure apply flow. “HDB refinance” should hit rates, lock-in, and a callback for a housing specialist, not the same savings form. Offer mismatch is how you burn 20-30% of spend after you already paid the CPC.

Form design is a marketing problem, not only a digital-banking problem. Ask for product, approximate amount, and contact first. Push NRIC and income into the authenticated flow. If you must collect more, show progress. Call tracking on branch-intent campaigns (credit unions still live on appointments) should fire a separate conversion so Search does not treat a 40-second “what are your hours” call as a funded member.

Session recordings and heatmaps tell you whether people stall on the disclaimer, the CAPTCHA, or the “upload payslip” step. HeyLead Insights is useful here because you can see the exact scroll depth where the MAS footnote sits on top of the CTA on mobile. Fix that overlap before you argue with bidding. Last-click will still double-count if the same person hits brand Search after a generic loan click. Import the funded event with a unique application ID and accept that channel scorekeeping will not match the CFO’s view of incrementality.

A realistic pattern: a regional credit union in the east ran one campaign for “join” and “personal loan”. CPA looked fine until operations showed that 7 of 10 forms were rate shoppers who failed the common-bond check. Splitting membership Search from lending, adding “not a member” as a negative on the loan campaign, and sending join traffic to a bond-explainer page cut junk applications without cutting spend. One mechanism. That is the work.

What marketing leaders are seeing

A pattern we see repeatedly: teams optimise to completed applications until operations show half of them failed eligibility on the common bond. CPA looks cheap. Funded members do not move until “issued membership” is the only primary conversion. (Composite example based on client patterns.)

A pattern we see repeatedly: legal holds RSA tests for weeks, so the account only gets enough conversions for tCPA after shipping from a pre-cleared headline bank - without rewriting claims every week. (Illustrative scenario.)

FAQs

Should Singapore banks use Performance Max before branded Search is clean?

No. PMax will harvest brand and comparison queries and report a blended ROAS that hides waste. Clean brand and transact Search first, then add PMax with brand exclusions and product-specific assets. Google reports case-study ROAS in the 4x-8x range for retail and travel; banking funded-account ROAS will depend heavily on product margin and attribution window, and only with conversions finance trusts.

Is automated bidding safe when CPCs keep rising?

It is safe when the primary conversion is a funded product and you have volume. It is not safe on form fills. If quality drops after you switch to tCPA, inspect the conversion action and the landing page before you go back to manual forever. Constant structure changes reset learning and make the test unreadable.

How do we handle MAS and advertising approval without freezing the account?

Pre-approve a copy bank per product. Pin only required legal lines. Use sitelinks for calculators and eligibility rather than stuffing disclaimers into every headline. Marketing still owns the query map and the lander; legal owns the claim list. That split keeps tests moving.

What should we do about AI Overviews killing clicks on “best savings” queries?

Stop buying those queries in Search. Put the comparison content in organic and brand. Pay for apply / open / refinance intent. Visibility without clicks is not a CPC problem you can bid through.

Can we promise a CPA before the account has run?

No one can guarantee a specific ROAS or CPL before seeing your account, landers, and unit economics. Three to six months of clean tracking and query control is a realistic window. Anything faster is usually a form-fill story.

Putting it to work

Start by exporting 90 days of Search Terms and tagging every query as transact, compare, research, or junk. That single file tells you whether Google Ads for banks and credit unions in Singapore is buying funded accounts or rate tourism.

This week

• Pause junk queries and add them as shared negatives.

• Confirm the primary conversion is funded or issued, not “submit”.

• Check that each transact ad group hits a matching lander, not the homepage.

• Run the Core Web Vitals check on apply URLs, especially mobile INP on the first CTA.

• Build UTM conventions so GA4 and the ads account describe the same product cluster.

Next 30 days

• Split membership, savings, lending, and cards into separate campaigns with their own CPA targets in SGD.

• Import delayed funded conversions and switch bidding only after 30+ of those events.

• Stand up a legal-approved RSA bank so tests do not wait on email chains.

• Review Search Terms every week for 30 days, then decide if PMax deserves budget.

If you want the query map, funded-event tracking, and MAS-aware RSA testing taken off your plate, HeyLead can run the Singapore Search programme around that exact leak between comparison clicks and accounts that actually fund. Chat with us on WhatsApp