Conversion tracking for paid ads: Google, Meta, Bing, and GA

When a Tampines aircon firm spends SGD 18,000 a month on Search and Meta, the marketing lead is not hunting for a definition of a pixel. They need one answer: which click became a WhatsApp enquiry, a PayNow deposit, or a booked servicing slot in an HDB or condo, and which click just burned budget. That is the job of paid ads conversion tracking in Singapore. The four systems you open after lunch (Google Ads, Meta Ads, Microsoft Advertising, GA4) are not scorekeepers for the same event. They are bidding engines and a website analytics product, each with its own clock, consent rules, and idea of a “conversion.”

Get the signal wrong and Smart Bidding and Advantage+ keep buying the cheap form fill. Get it right and you can still live with imperfect totals, because the thing you optimise toward is the same quality bar your sales team uses. This guide walks through how that chain actually works, what a healthy Singapore setup looks like, what you can ignore when the numbers refuse to match, and a compact audit you can run without rebuilding the account.

How a Singapore click becomes four different conversion counts

Someone in Punggol taps a Performance Max ad on mobile. Google records the click, stamps a GCLID, and waits for a conversion tag, Enhanced Conversions payload, or offline import. If your thank-you page fires generate_lead in GA4 and you import that as a primary conversion, Google will bid on it. If the same person later messages on WhatsApp and never hits the thank-you page, Google never sees the sale unless you send that event back. Meta never saw the GCLID at all. It saw a click ID of its own, a browser pixel if consent allowed it, and whatever you posted through Conversions API. Bing is sitting on UET plus, if you set it up, offline goals. GA4 is counting sessions and events on your domain, not ad platform credit.

Three clocks run at once. Platform conversion windows (often 7 days click / 1 day view on Meta, 30 or 90 days on Google depending on the action) are not GA4’s session or key event windows. Last-click in GA4 will give Search the win after a Meta view-through that Meta still claims. Modelled conversions fill gaps when iOS and ad blockers drop 20% or more of pixel fires, which is now normal, not an incident. You did not “lose the data.” The browser never handed it over. Server-side events and first-party hashes (email, phone) are how you put some of it back without pretending cookies still rule the auction.

Consent and PDPA sit in the middle of every Singapore stack. If GTM fires ads tags before the visitor accepts, you get inflated early numbers and a compliance problem. If tags never fire after accept, bidding starves. The practical pattern is: Consent Mode (or equivalent) gates Google tags, Meta’s pixel and CAPI respect the same banner, Bing UET follows, and GA4 still receives a cookieless ping so modelled key events are not empty. WhatsApp clicks need their own event. A tel: tap and a form submit are not the same quality. If both are marked primary in Google Ads, tCPA will hunt the cheaper tap.

CRM close is the last hop most teams skip. A HubSpot or Salesforce stage of “qualified” or “won” is not a browser event. You export hashed leads with GCLIDs and timestamps, or you stream them. Until that import exists, Google is still bidding on “someone typed in a form.” For Meta, Event Match Quality is a diagnostic for match keys, not proof the lead was profitable. For Bing, UET goals without offline import will always undercount phone-heavy SMEs in the East and North-East who close on WhatsApp after the session dies.

Landing pages change the signal before any platform counts it. A slow LCP on a mobile form (only around 55-60% of origins globally pass all three Core Web Vitals on mobile (Google CrUX report, Q1 2025)) means the click never becomes an event. If the ad promised same-day aircon servicing and the page is a homepage with three CTAs, you will still get “conversions,” just not the ones operations can fulfil. HeyLead Insights is useful here because session recordings and heatmaps show whether people rage-click the form, bounce at GST-inclusive pricing, or never see the proof block. That is post-click evidence, not another dashboard total. If the plumbing from ad to CRM is what you want owned rather than patched in slides, HeyLead’s analytics and CRM integration work in Singapore is built around that handoff.

What a trustworthy stack looks like when you open all four tabs

A healthy Singapore setup does not make Google, Meta, Bing, and GA4 print the same integer. It makes them bid on one business definition. Pick a primary action per funnel: for lead gen, that is usually a completed enquiry with a valid SG mobile number, not a newsletter tick. Mark only that action as primary in Google Ads. Secondary actions (scroll, button, WhatsApp click) can be observed, not bid on, until you prove they predict booked jobs.

Google side: gtag or GTM, Enhanced Conversions on with hashed first-party fields, Consent Mode v2, and offline conversion import for CRM stages that actually matter. Performance Max will eat whatever you call a conversion. If that is a 3-second landing, PMax will look like a 4x ROAS hero while sales complains. Google Search typically returns higher ROAS than Meta broad in lead-gen verticals - but only when the conversion event fed to Smart Bidding is close to revenue, not form volume. They are not a target you promise a board before you have seen the account.

Meta side: pixel plus Conversions API for the same event name, value in SGD when you have it, and a clean event for “Lead” that matches the form, not PageView. Deduplicate with event_id. Meta recommends one optimisation event per ad set at the pixel level - running multiple custom conversions against the same objective splits the learning budget and lets Advantage+ default to whichever event is easiest to record, not most valuable. Meta’s Advantage+ delivery consolidates targeting signals so the creative itself is the primary differentiator - tracking cannot rescue an ad the algorithm has already deprioritised. Tracking cannot save an ad that dies in five days, but dirty events will make the next creative look worse than it is.

Bing: UET on every paid landing page, goals aligned to the same thank-you or CRM import, and UTM discipline so GA4 can still see bing / cpc. Pair the UET tag with Microsoft Clarity so session recordings sit next to the same click IDs you bid on. Microsoft auto-tagging appends msclkid; that parameter must survive every redirect or offline import and GA4 reconciliation will break. LinkedIn Profile targeting inside Microsoft Advertising is a differentiator worth the smaller volume for B2B and higher-ticket home services in Singapore. Ignore Bing and you still leak branded Search in Singapore that Google does not fully cover.

GA4: one key event that matches the primary ads conversion, plus a separate event for WhatsApp and phone if those are real paths. Do not mark 15 events as key events. Use UTMs consistently (source, medium, campaign, content) so the ads click and the session can be reconciled even when the click ID drops. Attribution in GA4 is for journey diagnosis, not for paying the media invoice. When last-click and data-driven disagree, believe the CRM for “did we get the job,” and believe the ads platform for “what did we bid on.”

You will still see gaps. Modelled conversions, view-through, different windows, GST-inclusive vs exclusive revenue, and people who enquire twice. A good week is not four matching CSVs. A good week is Smart Bidding receiving enough qualified events that CPA on sales-accepted leads stays in a band you can live with, while GA4 explains the path. If form friction is eating mobile (conversion rates often lag desktop by a wide margin even when mobile is most of the clicks), fix the page before you add another pixel.

What to ignore when the dashboards refuse to match

Do not run a weekly ritual of forcing GA4 last-click to equal Google Ads conversions. They will not. View-through on Meta is not fake, but it is also not a booking. Treat it as a directional assist, not a line you add into ROAS for finance. Event Match Quality going from 6 to 8 does not mean the leads got better. It means more hashed emails matched. Useful. Not a KPI for the CMO.

Drop cargo-cult settings: firing ten conversion tags “for data,” counting button clicks as primary, importing every GA4 event into Google Ads, and screenshotting in-platform ROAS for the board pack while CRM says half the leads were students or job seekers. Automated bidding will spend into garbage if you feed it garbage. That objection is fair. The fix is the conversion definition, not a return to pure manual CPC as a religion. Manual bidding on a new account can be a training-wheels phase. It is not a measurement strategy.

Ignore vanity session counts and “all conversions” columns that mix add-to-cart with newsletter with lead. Ignore the story that attribution is dead so tracking does not matter. Last-click as a CEO answer is weak. Signal quality for bidding is still the difference between wasting 20-30% of SEM on junk queries and landing pages, and giving PMax something true to learn from. Also ignore guaranteed CPL before anyone has seen your offer, your GST pricing on the page, and your speed to WhatsApp reply. No serious partner should sell that.

If the argument in the room is only “Meta says 40, Google says 22,” you are having the wrong meeting. Ask which 22 would sales take. Then ask whether those 22 are in the conversion action the algorithm can see. When that gap is the story, a short look at paid ads conversion tracking for Singapore teams is usually more useful than another Looker Studio skin.

What marketing leaders are seeing

“We had generate_lead as primary and WhatsApp as primary at the same time. tCPA fell, booked slots did not. The cheap event was the tel tap from people who never sent a unit number.” - Head of Growth, home services, Singapore

“CAPI stopped the overnight cliffs after iOS, but we still over-trusted Meta’s 7-day click total in the QBR. Finance only calmed down when we imported ‘survey booked’ from the CRM and bid on that.” - CMO, education group, Singapore

FAQs

Should Google Ads, Meta, Bing, and GA4 ever show the same conversion total?

No. Different windows, modelled conversions, view-through, and consent mean the totals will diverge. Align on one lead definition and one primary action. Use CRM stages to judge quality. Use in-platform numbers to judge what you asked the algorithm to buy.

Do I still need the Meta pixel if I have Conversions API set up?

Keep both and deduplicate. The pixel catches browser-side behaviour CAPI cannot see. CAPI recovers events the pixel loses to blockers and ITP. Pixel-only is how you quietly donate 20%+ of signal.

How should WhatsApp enquiries be tracked in Google Ads Singapore?

Track them. Do not make them the only primary action unless operations truly closes there and you can qualify the thread. Pair the click event with a later CRM or “job booked” import so bidding is not addicted to chat starts.

Is Enhanced Conversions enough without offline import?

It helps match the people who convert on-site. It does not tell Google that a lead became a paying customer two weeks later. Longer cycles in B2B and higher-ticket home services still need the import.

What about GST in conversion values?

Pick one rule and document it. If revenue in the CRM is GST-inclusive at 9%, do not send GST-exclusive value into Google Ads purchase events. Inconsistent value trains value-based bidding to chase the wrong dollar.

Putting it to work

You do not need a full rebuild to see whether bidding is learning the wrong thing. Run a tight audit, then a short sprint.

Compact tracking audit

1. Write one sentence: “A conversion is a [form / WhatsApp / PayNow] that sales would accept.” Anything else is secondary.

2. In Google Ads, list primary vs secondary actions. Demote button clicks and page views. Keep one primary per campaign objective.

3. Confirm Enhanced Conversions and Consent Mode are actually firing after accept, not in preview only.

4. Match Meta pixel and CAPI event names and event_id. Kill duplicate Lead events.

5. Check Bing UET on the same landers you pay for on Google. Orphan pages are common after a web refresh.

6. In GA4, keep one key event aligned to that primary. Build UTMs with a shared naming sheet.

7. Sample 20 CRM leads from last month. How many have GCLID or UTM? If fewer than half, import and form hidden fields are the gap, not “attribution is dead.”

8. Watch five mobile sessions on the paid lander. If the form is below a slow hero or GST pricing shocks people, fix that before you add spend.

This week

• Export the last 30 days of Google Ads primary conversions and match 25 of them to CRM records by time and phone.

• Turn off primary status on any action that is not that sales-accepted event.

• Standardise UTMs on every live Google, Meta, and Bing campaign with the UTM builder.

• Verify CAPI and pixel share event names on one live Lead event.

• Run the Core Web Vitals checker on your top three paid landing pages.

Next 30 days

• Stand up offline conversion import for “qualified” or “won,” including GST-consistent values where you have them.

• Separate WhatsApp and form events in GA4 and in ads, then bid only on the one operations trusts.

• Rebuild any homepage destinations into message-matched landers.

• Reconcile Bing UET goals with the same definition, even if spend is smaller.

Pull the last 60 days of paid leads, keep only the ones sales would take, and check whether that exact event is still marked primary in Google and Meta. If it is not, the algorithm has been shopping for a cheaper proxy. HeyLead can take that conversion-definition and CRM-import loop off your plate so bidding follows booked work rather than form volume. Chat with us on WhatsApp