Can you trust your ad numbers? A 15-minute tracking check for Inner West businesses

Why a dashboard can say great while the phone stays quiet

Google Ads and Meta run automated bidding that learns from the actions you report, so a wrong signal produces wrong learning at machine speed. A form tag that fires on every refresh teaches the platform to find people who refresh pages.

If one enquiry is counted three times, every optimisation after it is built on sand.

Nothing looks broken: cost per conversion falls and a summary says the campaign is performing. Only the owner can compare that story with the shop, reception or the job book. So this guide has two parts: a 15-minute test of whether the tags work, and a slower three-source reconcile that tests whether the numbers mean anything. Once the numbers are trustworthy, why a good ROAS can still mean bad growth covers what they do and do not prove.

The 15-minute test enquiry

You need viewer access to Google Ads, GA4 and Meta (or a screen share), plus a phone and a desktop, because some tracking breaks only on mobile.

Set up and trigger (5 minutes)

1. List every action that counts as an enquiry: phone call, form, booking confirmation, quote request, email tap, direction request. Each needs its own tracked event.
2. Open a test log (time, device, action, result), and use a labelled name such as "TEST Do Not Call".
3. Do not click your own live ads, which costs a click and adds noise. Visit the landing page directly with a test tag on the end, for example ?utm_source=test&utm_medium=test&utm_campaign=tracking-check.
4. On a phone, in a private window, submit the form, tap the number and walk a booking through to confirmation. Repeat the form on a desktop.

Check GA4 (3 minutes)

Open Reports, then Realtime. Google describes it as monitoring activity as it happens, covering the last 5 and 30 minutes, so your visit and event names (such as generate_lead or form_submit) should appear within a minute or two. Then open Admin, Events and confirm the event is marked as a key event. Google defines a key event as one that measures an action particularly important to the success of your business, and an event that is not marked is not counted as a conversion.

Check Google Ads (4 minutes)

Open Goals, Conversions, Summary. Each action carries a status label. Labels include Recording conversions, No recent conversions and Inactive. Google says No recent conversions might mean no paid ad clicks led to conversions in the last 7 to 14 days, and Inactive often points to a misconfigured or missing tag. Google has changed its label set over time, so read whichever your account shows. A direct visit with no ad click may not appear as an ad conversion: this step proves the tag is healthy, and the reconcile proves the counts.

Check Meta (3 minutes)

In Events Manager, open your pixel, go to Test Events, enter your site address and repeat the enquiry. Each action should arrive once, including if you also send events through the Conversions API (see below).

Conversion setup: what is actually being counted

Primary versus secondary actions

Primary actions are reported in the Conversions column and used for bidding. Secondary actions are for observation only, appear in All conversions and are not used for bidding. If a tap on the phone number is primary, the system bids to maximise taps. If the goal is booked appointments, make the booking confirmation primary and the taps secondary.

One count or every count

Each action has a counting option. Google describes "every" as counting every conversion after an ad interaction and recommends it for sales, and "one" as counting only one conversion per ad click and recommends it for leads. A lead form set to count every conversion will report a person who submits three times as three enquiries.

Double counting

The most common cause of inflated numbers is two routes reporting one action. A Google Ads tag on the thank-you page and an imported GA4 key event for the same submission, both primary, count one enquiry twice. Add a Meta Pixel event and a Conversions API event and one submission can reach three or four counts across platforms, each plausible on its own. Choose one source per platform for each action and set the rest to secondary or remove them.

Phone calls, which are the real conversion for many businesses

Three different things get called a phone conversion, and they are not equal.

• A tap on the number. A click-to-call event records that someone pressed a button, not that a call connected or came from a new customer. Google Business Profile's Calls metric is likewise the number of times a customer clicked the call button, and Google discontinued Business Profile call history on 31 July 2024, so the profile cannot give you a call log.
• A call from an ad. Google Ads says you set a minimum call length and every call lasting at least that long is counted. Set it to separate a real enquiry from "are you open today", which means listening to your own calls first.
• A call recorded by your own phone system. This is the ledger you control: time, duration and number, plus a note from a person on whether it was a new enquiry.

For a dentist or plumber, the second and third matter far more than the first. A site with a polished form and no call tracking can report ten conversions while the phone handles forty.

Meta Pixel and Conversions API: deduplication

The Meta Pixel sends events from the visitor's browser and the Conversions API sends them from your server or a connected system. Meta's guidance recommends sending events from your own systems as well as through the pixel, but sending both without matching identifiers means Meta receives the same enquiry twice.

Meta's developer documentation sets the rules. The browser eventID must match the server event_id, and the event names must match. Events are only deduplicated if received within 48 hours of the first event with that identifier, and the system generally keeps whichever arrived first. A mismatch such as Lead from the pixel and lead from the server breaks it quietly.

To check, run your test enquiry with Test Events open and confirm the browser and server events are treated as one lead. If a plugin offers a Conversions API option, ask its support how it sets the event identifier. A vague answer means the server events are unproven.

Consent banners and a privacy caution

Two opposite faults occur. A banner that blocks every tag until someone accepts can record almost nothing. Or tracking runs with no thought for the personal information it sends to advertising platforms. This is general information, not legal advice. Check with the Office of the Australian Information Commissioner (OAIC) or a privacy professional about your own situation.

• On 24 June 2026 the OAIC found that Medmate and Monash IVF breached privacy law. Its finding was that using tracking pixels on health websites and then targeting visitors with social media ads amounted to a collection of sensitive information for which the website provider must obtain users' consent.
• The OAIC's pixel guidance, published 4 November 2024, says organisations cannot take a passive "set and forget" approach: they must understand how the product works, identify the risks and put mitigations in place.
• The Privacy Act's small business exemption covers annual turnover of $3 million or less, but health service providers are covered regardless of turnover. Dentists were not named in the summary we read, so do not assume either way. Check the OAIC's definition.
• A statutory tort for serious invasions of privacy commenced on 10 June 2025 and, the OAIC says, can reach entities outside the Privacy Act.

Google's consent mode documentation and user consent policy are framed around the European Economic Area, the UK and Switzerland, and Australia is not named in the policy's scope. That does not settle what Australian law requires of you, so a banner copied from a European template is not a substitute for advice. For health, dental and wellness sites, do not send details that reveal a condition or treatment to advertising platforms without advice. Optimise on the fact that an enquiry happened, not what it was about.

UTM naming: keep the sources readable

Google Ads auto-tagging adds a Google Click Identifier (GCLID) to the address people click through. Google says it is needed to track website conversions on all browsers and to import key event data from GA4, and that in GA4 auto-tagging is prioritised over manual tagging. Leave it on, and use manual UTM tags for everything else: Meta, email, QR codes, partner sites, magazine listings.

UTMs fail through inconsistency: Facebook, facebook and fb become three sources. Write down a convention: lowercase with hyphens; utm_source for the platform or publisher (meta, newsletter, qr-window); utm_medium for the type (paid-social, email, print); utm_campaign as year, month and theme (2026-10-spring-clean-newtown). A tagged QR code on a window or menu shows whether offline material sent anyone online.

The three-ledger reconcile

The 15-minute test shows whether tracking runs. The reconcile shows whether the numbers are honest. The method is to put three records side by side for the same four weeks and look at the gaps.

1. The platform ledger. What Google Ads and Meta say they delivered. This is the most generous record, because each platform has an incentive and a method to claim credit.
2. The site ledger. What GA4 recorded as key events. It is less generous because it uses its own attribution and cannot see phone calls, walk-ins or anything off the website unless you set those up.
3. The business ledger. The phone log, the inbox, the booking system and the till. It is the only one that records what happened. It is also the only one that cannot see where a customer came from unless someone asks.

Treat the business ledger as the anchor and the other two as claims to test. Copy this table into a spreadsheet and fill it in weekly.

After four weeks, divide spend by each column. Cost per platform conversion is what the dashboard shows. Cost per real enquiry and per booking are the numbers that matter, and the distance between them is the size of the problem.

Worked example (illustrative)

An illustrative Marrickville Road dental practice spends $2,400 on Google Ads over four weeks. The figures are invented to show the sums.

The dashboard said $48 a conversion. The business said $104 an enquiry and $300 a booking. Nobody lied: one form was counted twice, taps were counted as calls, and no tracked action captured the outcome. Fixes, in order: set the imported GA4 event to secondary, replace tap counting with a call-length threshold, and make the booking confirmation primary if the booking system can send it.

That last fix is harder when the booking widget sits on another domain. Google's cross-domain measurement attributes activity to a single user as they cross domains, and requires the domains to share the same web data stream tag ID. GA4 can also list unwanted referrals, so people returning from a booking or payment provider are not counted as a new source. Ask your provider what it supports.

How to read the gaps

• Platform higher than the business ledger. Expect this: double counting, taps counted as calls, view-through credit and existing customers who clicked. Our working rule is to investigate when the platform claims more than about a third above the business ledger. That is a rule of thumb, not an industry standard.
• Platform lower. Something is missing: calls that never reach the site, a form on another page, or a consent banner blocking tags.
• GA4 and the platform disagree wildly. Two tags or two counting methods are probably measuring one action. Check counting options and primary status first.
• Enquiries match but bookings do not. Tracking may be fine. Look at response time and the offer.

AI-assisted checking helps with the tedious part. An agent can read a full export of conversion actions, flag duplicates and match phone-log timestamps to site events in minutes. A person decides which calls were real enquiries and which gaps are worth fixing.

What we would do first, and what to do on a small budget

You do not need a paid call tracking service to start. This week:

• Run the 15-minute test on a phone and a desktop, and keep the log.
• List every Google Ads conversion action, mark primary or secondary, and demote any pair measuring the same enquiry, and check each counting option (one for leads, every for sales).
• Confirm nothing still depends on Universal Analytics, which stopped processing hits on 1 July 2023.
• Start the business ledger today: a shared sheet with one line per new enquiry (date, time, source, language, outcome).
• Add "How did you hear about us?" to the booking script and the till.
• Agree a UTM convention, tag every non-Google link, and run the Test Events check if you use the Conversions API.
• Book 30 minutes in four weeks to finish the reconcile table.

If you run Meta and Google together, do not add their reports together. Each uses its own windows and method and will claim some of the same enquiries.

Limits of this advice

Some enquiries cannot be traced. A person sees your ad on a bus down King Street and phones two weeks later. No setup solves that, and a vendor who claims otherwise is guessing. The "how did you hear" question is imperfect too: people forget, or name the brand and not the channel. Read it in patterns, not single entries.

We have not seen your accounts, so this guide describes common faults and how to test for them, not a diagnosis. Interfaces change: Google has relabelled its conversion statuses before, and per its help page Enhanced conversions for leads moved its upload route to the Data Manager API from 15 June 2026. If your screen differs, trust the screen. The privacy points are general information, not legal advice, so get proper advice before running pixels or retargeting in health, wellness or any sector handling sensitive information.

Our sample audit (a fictional report with invented figures) shows how tracking findings are written, and the Inner West research sets out our method. If tracking is fine and leads still do not convert, read why Meta ads get clicks but no customers.