What useful GA4 looks like for AU lead-gen sites

Finance in a Sydney or Brisbane B2B shop does not celebrate a 22% lift in engaged sessions. They ask whether last week’s Google Ads spend, billed in AUD and carrying GST on the agency line, produced enquiries a salesperson would actually ring. That is the job GA4 has on a lead-gen site here: turn a form, a click-to-call, or a booked demo into a first-party signal you can trust when Performance Max is already doing the bulk of delivery and CPCs keep climbing.

Most Australian properties already have a GA4 container. Few of them measure the same thing sales counts as a lead. Pageviews still sit in the default reports. Scroll depth is marked as a key event. A thank-you URL fires twice because the tag and the native form listener both fire. Meanwhile Smart Bidding is chewing on that noise, and the in-platform CPA diverges from what landed in the CRM. Useful GA4 is not a prettier dashboard. It is a small set of events that match real enquiries, named the same way in GTM, GA4, and Google Ads, with enough user-provided data that Enhanced Conversions can survive ad-blockers and privacy loss that routinely wipe 20%+ of conversion hits.

This guide is for the person who owns the marketing number: how the measurement actually works, what a healthy Australian lead-gen property looks like, what you can ignore, and a compact check you can run without waiting for a full rebuild.

How an Australian enquiry actually becomes a bid signal

Start with the object, not the report. Someone on a Melbourne landing page submits a form, taps Call on mobile, or books a 15-minute slot. That action has to become a GA4 event with a stable name, a handful of parameters, and a conversion flag that Google Ads can import. GTM is usually the switchboard: a trigger (form submit, thank-you page, tel: click, Calendly postMessage), a GA4 Event tag, and a separate Google Ads conversion tag or the GA4 import. If those three disagree on the name or fire on different conditions, you will spend the next quarter arguing about CPL instead of improving it.

GA4 records hits as events. A page_view is an event. generate_lead is an event. The difference that matters is whether you marked the second one as a key event and whether you sent it to Ads as a primary conversion. Key events in GA4 are not automatically Ads conversions. You still map them in the Google Ads interface, pick a count setting (usually one per session for lead gen), and wait for enough volume before bidding algorithms treat the series as reliable. For Australian accounts, that mapping also has to survive Consent Mode v2: if ads_storage is denied, you still want a modelled conversion path rather than a silent zero, which means the tag must load in a consented or cookieless-pings configuration, not sit behind a hard block that never fires.

Parameters do the unglamorous work. lead_type (demo, quote, callback), form_id, page_location, and value (even a placeholder like 1 for a raw enquiry) let you split “contact us” spam from high-intent demo requests without inventing a second property. User-provided data (hashed email, phone) is what Enhanced Conversions uses. Skip it and you donate more of that 20%+ loss to Smart Bidding’s guesswork. Mobile is not a footnote: Google’s global benchmark puts mobile at roughly 60% of paid search clicks - Australian B2B verticals tend to run lower, but click-to-call gaps on iOS are still the most common silent leak we find in AU lead-gen audits, and conversion rates still lag desktop by a wide margin, so a tel: click that never fires on iOS Safari is a real leak, not a reporting quirk.

Attribution inside GA4 will not match last-click in Google Ads, and it will not match Meta. That is expected. Last-click still double-counts the same enquiry across channels if you mark the same form as a conversion in three places. Useful setup treats GA4 as the first-party ledger of what happened on-site, Ads as the auction optimiser, and the CRM as the quality filter. You reconcile weekly on unique enquiry IDs or hashed emails, not by forcing every tool to print the same number. When in-platform CPA and CRM-qualified CPL diverge, you debug the event, not the bid strategy first.

Landing pages sit in the same loop. If the form is below a slow hero and INP is poor, you will under-count because people bounce before the listener attaches. Only 55.9% of CrUX origins pass all three Core Web Vitals; lead-gen pages that fail INP on mid-range Androids in Perth or Adelaide quietly starve the event stream. Session recordings help you see the abandoned field, not just the missing hit. That is where HeyLead Insights earns its keep: heatmaps and recordings show whether the generate_lead gap is a tag or a form people never finish.

If the wiring already feels heavier than a part-time marketer should own, a focused GA4 configuration pass (events, key events, Ads import) is usually cheaper than another quarter of bidding on thank-you duplicates.

What a healthy lead-gen GA4 property looks like when CPL is the scoreboard

A useful Australian property is boring on purpose. One GA4 property per brand (or per clearly separate P&L), not a property per campaign. Data streams split web from app if you have both; most lead-gen sites only need web. Internal traffic from the Sydney office and the Melbourne warehouse is filtered. DebugView is used during deploys, then switched off so testers do not inflate generate_lead. Time zone is Australia/Sydney or the market that owns the P&L, not the default US setting that makes “today” lie to you every morning.

Event hygiene is the tell. You will see page_view, session_start, and user_engagement because GA4 creates them. You will also see a short list you defined: generate_lead (or a tighter name like request_quote), click_to_call, and maybe book_demo. Those three, if they exist, are key events. Scroll, file_download, video_start, and outbound clicks stay as events, not conversions. Count is Once per session for leads unless you truly sell multiple distinct products in one visit. Value is either blank or a conservative AUD estimate you can defend to finance, never a fantasy ROAS input.

Naming matches across tools. If GTM sends generate_lead, GA4’s key event is generate_lead, and the Ads conversion is also generate_lead (or a clearly mapped import). Parameters are documented in a one-page sheet: who owns the tag, what the trigger is, whether Enhanced Conversions is on (unified toggle), and whether the CRM writes a qualified_lead event back via Measurement Protocol or a later offline import. Offline import is what separates “form filled” from “sales accepted” for longer B2B cycles. Without it, bidding optimises for tyre-kickers on the Gold Coast who wanted a PDF, not the demo that closed.

Reporting is narrow. The acquisition report is sliced by session source/medium and campaign with UTMs that actually parse (utm_source=google, utm_medium=cpc, utm_campaign=brand-au-search, not “Google Ads / CPC” pasted from a spreadsheet). You keep an exploration that answers three questions: how many unique generate_lead events this week, from which landing pages, at what estimated cost when you join spend from Ads. You do not build a 14-card Looker Studio that restates Sessions. Cadence is weekly enough to catch a broken thank-you tag before the learning period eats a five-figure AUD budget, not a monthly PDF that buries the miss.

Consent and modelling are visible, not hidden. You can say what share of conversions is observed versus modelled. You know Enhanced Conversions is collecting email on the form. You have tested a denied-consent path so the site still sends cookieless pings. You have not “solved attribution.” You have a ledger that is good enough for Smart Bidding and honest enough that a founder in Canberra can compare channel contribution without pretending last-click is the CEO answer.

Two patterns show up when this is healthy. First, Ads primary conversions drop in volume and rise in quality after you kill scroll-as-conversion, and CPA looks worse for a fortnight before qualified pipeline per dollar improves. Second, mobile click_to_call finally appears, and you discover a chunk of “non-converting” Search was people who rang from the SERP-style call button and never loaded the thank-you page. Neither pattern needs a new tool. It needs the event list to match how Australians actually enquire.

Vanity hits, cargo-cult Admin, and the reports that never book work

Engaged sessions, average engagement time, and “users who viewed 3 pages” do not pay GST on a media invoice. They are fine as diagnostics when a landing page is clearly broken. They are not conversions. Marking them as key events teaches Performance Max to hunt for scrollers. You will get cheaper “conversions” and a CRM that is quiet.

Admin theatre is the other trap. Creating five custom channels, twenty audiences you never use in Ads, and a full e-commerce schema on a site that sells consults looks sophisticated in a screenshot. It does not change booked work. Neither does copying a US measurement ID “best practice” list that fires begin_checkout on a quote form. Cross-domain measurement matters if you send people from the .com.au site to a Calendly subdomain; if you do not, leave it. Duplicate properties “for testing” that still receive production tags will split your conversion history and reset learning when someone points Ads at the wrong one.

Last-click scorekeeping across GA4, Meta, and Ads will never reconcile to the dollar. Teams that force that reconciliation waste weeks and still cannot explain incrementality. Use GA4 to see on-site behaviour and unique events. Use Ads to see auction CPA. Use the CRM to see accepted leads and revenue. Accept that modelled conversions exist because 20%+ of hits never arrive. Do not turn modelled conversions off to “keep the data pure.” You will starve bidding and then blame automation for inflating spend without improving quality, which is the objection finance already has.

Ignore guaranteed CPL promises that assume the current event stream is true. If generate_lead includes newsletter signups, the number is fiction. Ignore daily structural tweaks in GA4 “to help the algorithm.” Constant changes reset the same way campaign restructures do. Ignore exporting 400 generate_lead rows and calling it a forecast. Deduplicate on user_pseudo_id plus timestamp plus form_id, then match to CRM, then talk about cost per qualified lead.

Also ignore the homepage as the default landing for paid. If the team says they just send traffic to the homepage, the measurement problem is downstream of a positioning problem. Dedicated pages with one form, one offer, and UTMs that survive the click are what make GA4 reports readable. When those pages stall, look at form friction before you add another event. Recordings beat another custom dimension.

A short mid-article check: if your key events list is longer than your sales-accepted lead types, cut it. If you cannot name the GTM trigger for click_to_call on iOS, fix that before you ask for more budget. Clean conversion setup is still the phrase buyers use for a reason.

A compact check you can run on the live property without a rebuild

You do not need a six-week measurement programme to see whether the current setup is useful. Run this on the production site, in an incognito window and on a real phone, with DebugView open and Google Ads conversion diagnostics in another tab. Stop when you can answer “does this event equal an enquiry sales would ring?” for every key event.

DIY playbook

1. List every GA4 key event. Delete or unmark anything that is not generate_lead, click_to_call, book_demo, or a CRM-qualified import. Keep scroll and file_download as events only.

2. Submit the real form on the highest-spend AU landing page. Confirm one hit in DebugView, one in the realtime report, and one in Google Ads (or the GA4 import) with the same name. If you see two, find the duplicate listener before you touch bids.

3. Tap the click-to-call button on iOS Safari with consent accepted and denied. If denied consent zeros the event entirely, fix Consent Mode so pings still flow; if accepted never fires, the trigger is wrong.

4. Check Enhanced Conversions is on and the form actually collects email or phone that can be hashed. Empty user-provided data is why modelled conversions look heroic and CRM looks empty.

5. Pull 14 days of generate_lead, dedupe, and spot-check 20 rows against the CRM. Note the miss rate. That miss rate is your real tracking tax, not the Ads CPA card.

6. Inspect UTMs on paid landing URLs with a builder so source, medium, and campaign survive redirects. Broken UTMs make GA4 blame “direct” for Search spend.

7. Run Core Web Vitals on those same URLs. If INP is poor on mobile, fix the page before you declare the event “low volume.” Tags cannot fire on a thread that never becomes interactive.

8. Document the three remaining events on one page: trigger, parameters, Ads mapping, owner. If nobody owns it, it will drift the next time someone installs a chat widget.

That list is enough to brief an analyst or to decide the property needs a rebuild. Do not add audiences, custom channels, or a second property while those eight items are red. Signal quality beats another dashboard. If paid social is in the mix, keep the same generate_lead name there too so you are not teaching two auctions two different definitions of a lead.

What changes after the cleanup

Two patterns we see after a GA4 event audit:

“We had generate_lead firing on the newsletter footer as well as the demo form, so Smart Bidding looked cheap until sales said the calendar was empty. Cutting it to the demo thank-you made CPA jump 40% in nine days, then qualified pipeline per dollar actually moved.” - Head of Growth, B2B SaaS

“iOS click-to-call was never in GTM. Search looked like it died on mobile in Brisbane. It hadn’t. People were ringing and we were optimising to desktop forms only.” - Founder, Australian services brand

FAQs

Should generate_lead be a Google Ads primary conversion or only a GA4 key event?

If you bid for enquiries, import it (or fire a matching Ads tag) as a primary conversion with once-per-session counting. A key event that never reaches Ads is a report, not a bid signal. Keep secondary conversions for assist diagnostics, not for the algorithm.

How do we handle modelled conversions when finance wants “real” numbers?

Show both: observed events matched to CRM, and modelled fill for denied consent. Turning modelling off does not make the data more true. It hides the 20%+ you already lose to blockers and privacy. Use CRM-accepted leads as the quality scoreboard and GA4/Ads as the volume and auction feed.

Do we need a separate GA4 property for Google Ads versus organic?

No. One property, one event dictionary, UTMs that distinguish paid from organic. Splitting properties splits conversion history and makes Enhanced Conversions weaker. Filter internal traffic instead of cloning the measurement ID.

When is Measurement Protocol or offline conversion import worth the work?

When sales-accepted leads lag the form by days or weeks, which is normal in Australian B2B. Upload qualified or closed-won against the original GCLID or hashed email so bidding can prefer the demo that closed, not the PDF download that looked like a conversion at the time.

What if Meta’s CPA and GA4’s generate_lead never match?

They will not. Different windows, different modelled paths, different click IDs. Align the event definition and the landing page, then compare direction week to week. A simple weekly check: if both channels show generate_lead moving in the same direction (up or down) week-over-week, the signal is real; if they diverge sharply, audit the landing page UTMs and form trigger first before changing bids. Forcing a single number is how teams stall while CPCs keep rising.

Putting it to work

This week

• Export key events and mark only enquiry-type events as conversions in both GA4 and Google Ads.

• Submit the live form and the click-to-call path once on desktop and once on iOS; screenshot DebugView.

• Dedupe 14 days of generate_lead against the CRM and write down the miss rate in one sentence for finance.

• Fix UTMs on the top three paid URLs with the UTM builder so GA4 stops dumping spend into Direct.

• Note INP on those URLs; if the form is slow to become interactive, put that in the same brief as the tag work.

Next 30 days

• Turn on Enhanced Conversions with real email/phone fields and verify Event-level user-provided data is populating.

• Add a single offline import or Measurement Protocol hit for sales-accepted leads if the cycle is longer than a few days.

• Kill duplicate listeners (native form + GTM + chat widget) until each enquiry is one hit.

• Rebuild the weekly view around unique enquiries, landing page, and cost, not engaged sessions.

Pull the last 14 days of key events, keep only the ones a salesperson would ring, and make Ads bid on that list before you change another campaign. If the mapping between GTM, GA4 key events, and Google Ads conversions is the thing that keeps slipping while you are trying to run the rest of growth, HeyLead can own that enquiry-signal wiring so you are not debugging thank-you duplicates every time a form plugin updates. Chat with us on WhatsApp