The conversion column in a Search account can look healthy at 11am on a weekday in Raffles Place. CPL in SGD is down. Lead forms are stacking. Ops still has gaps on the calendar, and the WhatsApp thread is full of people asking for a price they will never pay, a service you do not offer, or a job in Johor that was never going to book. That is not a “volume problem”. It is a quality problem wearing a green conversion badge.
Singapore Search is expensive enough that junk traffic hurts twice. You pay the click. You also pay the person who has to reply, qualify, and then write off the enquiry. Industry ROAS benchmarks typically cited in SEM audits range from 3x to 6x depending on sector - which sounds healthy until you strip out the junk traffic share. In accounts we’ve audited, 20-30% of spend routinely leaks to irrelevant queries before negatives are tightened, neglected negatives, or a landing page that does not match the ad. Mobile takes most of the clicks. Conversion on mobile often lags desktop. If your “lead” is a half-filled form on a slow page, smart bidding learns the wrong lesson and spends more of it.
This piece is about the specific Google Ads mistakes that fill forms but not jobs for Singapore companies. Not a rant about “strategy”. The actual settings, conversion actions, match types, and post-click gaps that pad the dashboard.
Account settings that convert in Ads and stall in the ops queue
You do not need seven new campaigns. You need to stop teaching Google that a tyre-kicker in a condo lobby is the same as a booked job. Each mistake below is a real configuration or skip. The bite is always the same: cheap CPL, noisy WhatsApp, empty pipeline.
Common mistakes
Action checklist
- Primary conversion is the form submit, not a booked job You fire a conversion on thank-you page load, or on every WhatsApp click, then hand Maximize Conversions a diet of junk. Smart bidding chases that event. In Singapore that often means after-hours enquiries, students, and “how much for aircon servicing” shoppers who never intended to book. Fix: keep the form event as a secondary observation. Make the primary action something closer to money: qualified lead from the CRM, PayNow deposit, confirmed appointment, or a call over a minimum duration. If you cannot get CRM feedback yet, at least exclude micro-events from bidding. Enhanced Conversions help when 20%+ of conversion data is already leaking to ad-blockers and privacy settings. Clean signal beats more signal.
- Broad and phrase match with a thin negative list Search terms reports in SG accounts still fill with jobs, salary, course, DIY, free, cheap, and competitor brand variants you never meant to buy. Broad match is not the villain by itself. Broad match with no weekly negatives and a homepage as the landing URL is. You will see queries from Central that look commercial and still dump you into “near me” intent you cannot serve, or HDB vs landed language you never wrote for. Fix: mine search terms twice a week until the list is boring. Add shared negative lists for careers, training, wholesale, and out-of-country. Keep phrase and exact on high-intent clusters (book, quote, emergency, same-day) and let broad run only after the conversion action is qualified. “How do you handle negative keywords?” is the question serious buyers ask for a reason.
- “Performance Max drives the bulk now” is true in a lot of SG accounts. It is also how Display and YouTube inventory quietly eat budget that used to sit on high-intent Search. If you did not exclude brand, did not feed customer lists, and did not set Search as the priority for query capture, PMax will harvest your brand terms and report them as “new” conversions. Fix: brand Search stays in its own campaign. Feed first-party lists. Use account-level URL exclusions for careers pages and blog posts that cannot convert. Read channel reporting, not just the campaign total. If PMax ROAS looks 4x-8x while ops says the jobs are all existing customers, you are counting remarketing as acquisition.
- Ad copy promises a quote, the page asks for a 12-field form The RSA says “WhatsApp us for a same-day quote” and the landing page wants company UEN, GST status, three budget fields, and a 9% GST lecture before anyone can send. Mobile is most of paid search clicks. A heavy form on a slow origin is how you convert browsers and lose buyers. Only 55.9% of origins pass all three Core Web Vitals in May 2026 CrUX data. If INP is ugly, people tap WhatsApp in the header and never complete the tracked form, so bidding never sees the real closer. Fix: match the offer in the ad. Quote request, not a novel. Put proof above the fold (reviews, licences, response time in hours, not “we care”). Use HeyLead Insights to see where scroll dies and which fields get abandoned, then cut those fields. Dedicated landing pages aligned to the ad, never the homepage. If someone says they just send traffic to the homepage, that is your cue to stop the meeting.
- Location targeting set to “presence or interest” You think you are buying Singapore. Google is also showing ads to people interested in Singapore from elsewhere, plus travellers planning a trip, plus people whose VPN last touched Changi. Search terms look English. The phone country code does not. Ops wastes mornings on enquiries that cannot PayNow and cannot show up in Tampines, Punggol, or Woodlands. Fix: presence only. Exclude locations you cannot serve. If you cover East and North-East but not the West, say so in the ad and on the page. Add a location question early on the form so sales does not discover it on the fifth WhatsApp message.
- Last-click thank-you pages counted in Ads, GA4, and the CRM as three wins CM360-style last-click counting has been getting noisier, and last-click still double-counts the same form across channels. Marketing celebrates 40 leads. Sales says 11 were real, 4 booked, 2 paid. Automated bidding then inflates spend because it thinks quality improved. Fix: one source of truth for “qualified”. Import offline conversions or at least a weekly hashed dump of booked jobs. Do not optimise to the Ads column alone. Attribution will not be the CEO answer, but it still has to be clean enough that tCPA is not chasing ghosts. Unified Enhanced Conversions plus a CRM status beat another dashboard tile.
DIY tools that catch these mistakes
Run these on the live landing URLs and the UTMs you actually use in Search
If the account is already on Maximize Conversions and the landing page is the homepage, pause the spend increase until those two are fixed. Adding budget on a dirty conversion action is how CPL looks cheaper while cost per booked job climbs. A tight SEM / Google Ads structure with fewer campaigns and cleaner signals usually beats another layer of ad groups.
What to repair first when CPL in SGD looks cheap
Do not rebuild the account on a Friday because a founder screenshotted 80 form fills. Sequence the work so bidding does not relearn garbage every time you touch a campaign.
This month, conversion hygiene comes first. If the primary action is a form or a WhatsApp tap, every other optimisation is theatre. Map the event in GTM, confirm it fires once, import a qualified status if you have it, and switch bidding only after two weeks of cleaner data. Parallel to that, lock location to presence and dump the search terms into a shared negative list. Those two changes stop the worst leak without resetting learning on every RSA.
Next, match type and landing match. You can live with broad on a qualified conversion. You cannot live with broad plus a 12-field form that talks about GST 9% before it talks about the job. Align headline to H1, cut fields, and check Core Web Vitals on mobile from a real SG connection, not an office Wi-Fi story. Speed before new creative. Match types before net-new campaigns. Tracking before scale.
What can wait: a full Performance Max rebuild, a Display experiment, a new brand video, another tool in the stack. Those are tempting because they look like “activity”. They do not fix the mechanism that filled the dashboard with people who will never PayNow. Realistic Search quality work in a live account is a 3-6 month arc, not a guaranteed CPL from a pitch deck. No one can promise a specific ROAS before they have seen your conversion actions, landing pages, and unit economics.
If you already know the landing handoff is the leak, a short look at on-page behaviour is enough to brief the next test. You do not need a redesign to learn that the CTA sits below a 2,000-word block nobody on MRT data will scroll. When you want that channel owned rather than poked at, Google Ads management in Singapore is the workstream that should sit with the same people who own negatives, RSAs, and the conversion import, not a separate “media plan” that never touches the form.
Get a free marketing audit - we review your search, ads, and landing pages and send back what to fix first.
Get a free audit
The growth lead in Tanjong Pagar who stopped treating every form as a job
A Head of Growth at a B2B services firm near Tanjong Pagar had a Search account that looked like a win on paper. CPL had dropped. Forms were up. Sales was quietly routing most of the WhatsApp enquiries to a junior who spent the afternoon asking “are you in Singapore?” because location was set to presence or interest and the primary conversion was every form submit, including the ones with a Gmail address and a one-word message.
The one change that flipped the week was not a new campaign. They demoted the form to observation, imported “qualified - meeting booked” from the CRM as the primary action, and added a presence-only geo plus a negative list for jobs, intern, and course. Bidding got worse for 11 days. Then cost per booked meeting fell even though form volume looked “worse” in Ads. Ops stopped celebrating the thank-you page. The calendar started filling with people who could actually show up.
That is the pattern. The dashboard lied because you asked it to. Once the conversion matched the job, Google had something useful to buy. Creative and landing copy still needed work, but they were no longer fighting a bidding model trained on tyre-kickers.
From the account
“We had Maximize Conversions eating every WhatsApp tap. CPL in SGD looked brilliant until sales showed us 9 of 12 ‘leads’ were asking if we hire. We switched the primary action to booked consult and the cheap CPL vanished, which was the point.” - Head of Growth, B2B services, Singapore
“PMax was reporting 4x and half of it was our own brand plus people already on the customer list. We split brand Search out and the ‘new pipeline’ number got honest overnight.” - CMO, professional services firm
FAQs
Should we pause Performance Max if form quality is poor?
Not as the first move. Fix the conversion action and brand cannibalisation first. If PMax is still flooding Display with cheap, unqualifiable traffic after that, cap it or exclude the junk URLs. Killing PMax while Search is still bidding on form spam just moves the same problem. For Singapore lead quality, separate brand into its own Search campaign so PMax cannot harvest branded queries and report them as new pipeline. Upload a customer match list so existing buyers are not counted as acquisition. Add account-level URL exclusions for blog and careers pages that cannot convert a booked job. Those three signals - brand campaign separation, customer match, and URL exclusions - are the practical PMax brand cannibalisation fix before you touch the campaign on/off switch.
Is broad match always wrong for Singapore Search?
No. Broad match with a dirty conversion and no negatives is wrong. Broad match on a qualified offline conversion, with weekly search term hygiene, is how a lot of accounts keep CPCs from running away while still capturing the way people actually type (HDB, condo, landed, aircon servicing, same-day).
Why does mobile traffic convert worse even when CPC looks similar?
Most paid clicks are mobile. Forms, GST copy walls, and slow INP punish thumbs. Desktop conversion can look 30-40% stronger in mixed reports and hide the real landing problem. Check mobile CWV and form completion separately before you blame the keyword.
Can we keep using WhatsApp click as a conversion?
As an observation event, yes. As the thing tCPA or Maximize Conversions optimises to, only if you have already proven those taps become booked work. In Singapore, WhatsApp is how buyers talk. It is also how tyre-kickers talk. Do not let bidding treat them as equal.
How long until quality work shows in cost per booked job?
Expect a messy 10-14 days after you change the primary conversion, then a clearer 4-8 weeks if negatives, geo, and landing match are in place. Promising a number before the account has run on clean signals is how retainers lose trust.
Putting it to work
Execution sprint
This week
- Pull 30-90 days of performance for google ads mistakes that fill forms but not jobs (Search Console, ads, CRM, or call logs - whatever you have).
- Flag the top leak: wrong intent, weak page, slow response, or dirty conversion tracking.
- Ship one fix on the highest-traffic money path (page, campaign split, or response rule).
- Run the free tools below on that same URL or account and log the findings.
Free tools for this sprint
Next 30 days
- Expand the fix to the next one or two money paths only after the first one shows cleaner bookings.
- Align creative, keywords, or content with the same offer the page now states.
- Review booked outcomes weekly; cut anything that still only produces unqualified volume.
Pull the last 60 days of Google Ads conversions next to booked jobs in the CRM, not next to form volume. If the two lists barely overlap, you already know the primary action is lying.
This week
-
Export search terms and add negatives for jobs, salary, course, free, DIY, and any country you cannot serve.
-
Switch location to presence; exclude what you do not cover in East, West, North, or North-East.
-
Demote form and WhatsApp clicks to observation if they are currently primary.
-
Check landing URLs with the Core Web Vitals checker and cut one form field that sales never uses.
-
Tag the next RSA traffic with the UTM link builder so GA4 and Ads stop arguing about source.
Next 30 days
-
Import qualified or booked-job conversions and give bidding two weeks on the new signal.
-
Split brand Search out of PMax and read channel reporting, not the campaign total.
-
Rewrite one landing page so the H1 matches the RSA promise (quote, WhatsApp, same-day) instead of a generic homepage.
-
Agree with sales on a weekly “junk vs booked” count in SGD so marketing stops celebrating the wrong column.
If the gap you cannot staff is the loop from Search query to booked job, the conversion import, the negatives, and the landing match that currently teach Google the wrong buyer, HeyLead can run that Google Ads quality work as an ongoing programme so you are not re-teaching bidding every quarter. Chat with us on WhatsApp
Free marketing audit, or reach Martin directly:
Get a free audit Connect on WhatsApp · martin@heylead.com