Call tracking: the missing half of local conversion data

Why does this matter so much for local?

Because local intent resolves fast and physically. In widely cited local-search data, 76% of people who run a "near me" search visit a business within 24 hours, and 88% of mobile local searchers visit or call within a week. Much of that action lands on the phone: directions, opening hours, and above all bookings. The gap in lead quality is just as stark as the volume story. A caller converts at 37% during the call, 46% in home services; a form fill at 1.7% often gets a reply tomorrow, by which time the searcher has rung a competitor who answered.

For a multi-location brand the distortion is uneven, which makes it dangerous. Branches with older demographics, urgent services, or trade customers skew heavily to phone; branches with younger, form-comfortable audiences look artificially better in form-only data. Budget follows the reported numbers, so the phone-heavy branches get starved by an accounting artifact. That is precisely what the gym chain in this diary was about to do to three of its best sites.

What are the three layers?

1. Google forwarding numbers on call assets and call ads: free, native, and they attribute calls from the ad itself to campaign and branch.
2. Dynamic number insertion on landing pages: each session sees a trackable number, so calls made after the click, which is most of them, still attribute to their source.
3. Offline conversion imports from your booking system or CRM: the upgrade from "call happened" to "appointment booked, worth €X," which is what bidding should optimize toward.

Layer three is where most setups stop short, and it is the layer that changes bidding behavior. Smart Bidding fed only "calls" optimizes for ring volume, which any wrong-number afternoon can satisfy. Fed booked outcomes with values, it prices a 40-minute consultation differently from a cancelled trial, and the difference shows up within two bid cycles. Where a booking system exists, the import is a weekly CSV at worst and an API sync at best; either way it is hours of setup, never a project.

Which thresholds keep the data honest?

Duration first: 30 to 60 seconds depending on how long a genuine enquiry takes, calibrated against a week of real call logs. Local Services Ads use their own 30-second bar for valid leads, a useful reference point. Then dedup rules, so the same customer calling twice does not count as two leads, and business-hours flags, since a missed after-hours call is real demand with a different fix, usually a rota rather than a bid.

Review the thresholds quarterly. One franchise client discovered its 30-second bar was counting voicemail drops at two rural branches whose greeting message ran long; the fix was a 45-second bar plus a shorter greeting, and the branches' reported CPL rose to its honest level overnight. Thresholds are boring precisely until they decide a budget.

What changes once calls count?

Budgets move toward the truth. The gym chain had marked three sites for budget cuts on form-only data; corrected counting showed their real cost per trial near €24 against a claimed €55-plus, and those sites finished in the chain's top ten. In benchmark terms, phone-heavy verticals live well inside their bands once counted: health and fitness at a $67.36 median CPL, dentists at $72.97. The per-branch ledger from the structure guide only tells the truth when this layer exists underneath it.

The second-order effect is operational visibility. Answer rates per branch become a report column, and they vary more than anyone expects; a branch that answers 60% of its ad-driven calls is burning close to half its call budget on the phone system rather than the auction. We surface it because the fix belongs to operations, and because it also drives Local Services Ads rankings, where responsiveness outweighs bids.

Where does consent fit?

The Consent Mode v2 regime has governed EEA and UK advertisers since March 2024, and non-compliant accounts lost conversion modeling for declined users in March 2025, per Google's consent documentation. Number swapping and call counting sit inside that same consent framework. Our default is counting without recording: number, duration, branch, campaign, and nothing else, which keeps the compliance surface small while giving bidding everything it needs. Where a client wants recordings for training, that runs on their own phone system under their own lawful basis, outside the measurement stack. The full install runs through the call and lead tracking service, including the release-day regression checklist that keeps later site updates from silently breaking the wiring.