Tracking phone calls as conversions

For a lot of service businesses the phone is the highest intent channel they have and the only one with no reporting. Forms get counted automatically. Ads get counted automatically. The phone rings, somebody answers or does not, and no system anywhere records what happened.

That gap distorts every budget decision downstream. A channel that produces calls looks worse than a channel that produces form fills, purely because one is visible.

What should count as a phone conversion?

A connected call, above a duration threshold, from a first time caller.

Each part of that does work. Connected removes the calls that rang out, which you should count too, but in a different column, because they are a problem rather than a result. Duration threshold removes wrong numbers, hangups, and suppliers. First time caller removes existing customers checking their order, who are valuable but are not new demand.

Pick the threshold against a real distribution rather than a guess. Maple's analysis of 1.2 million calls across more than 1,000 US restaurant and local-business locations, December 2023 to November 2025, puts the average call duration at 1 minute 45 seconds. Maple sells restaurant phone answering, so label it vendor research, but the duration is platform measured. Against a mean of 105 seconds, a threshold somewhere in the 30 to 60 second range removes the noise without cutting into real conversations. Listen to a handful of calls sitting near your chosen line before you commit to it.

Without those filters your call count will look excellent and mean nothing. A raw call volume number counts your linen supplier and a wrong number the same as a booking.

What does the filtering do to a real month?

Here is a full month with the arithmetic shown. Run it on your own call log.

Raw inbound calls logged. 214.

Remove the short ones. 63 calls came in under 30 seconds. Left with 151.

Remove the non-demand. Tagging the remainder gives 24 existing customers, 11 suppliers and 6 wrong numbers, which is 41 calls. Left with 110 qualified first-time calls.

The honest ratio. 110 of 214 is 51.4%derived. Roughly half of what a raw call counter would have reported as conversions were not new demand at all. If you have been budgeting against the raw number, you have been paying for a number that was double the truth.

Now the other column, which nobody builds. Of the 214, 29 rang out or went to voicemail. That is a 13.6%derived miss rate. Set that beside CallRail's benchmark data, drawn from analysis of 1.1 million de-identified conversations and announced in January 2025: healthcare 32%, legal 28%, home services 14%, real estate 9%. CallRail sells call tracking, so label it vendor research, and note its definition, which is an unanswered inbound call that results in a detected voicemail or is detected as abandoned. Your 13.6% is ordinary for a home services business and would be excellent for a clinic.

Then the money. $1,850 in ad spend produced 34 of the 110 qualified calls through the paid tracking number, which is $54.41 per qualified call (derived). Compare that to your own trailing three months rather than to a published average, because a published average is a national campaign-level median and knows nothing about your city or your ticket.

Why is one industry miss-rate number useless?

Because the segment decides the number, not the vertical, and the spread inside a single industry is larger than the spread between industries.

Revmo AI analysed 12,091 real restaurant call recordings and published answer rates by segment: quick service 59.9%, fast casual 75.3%, full service 91%, pizza 93.1%. That is a miss rate running from roughly 7% to roughly 40% inside one word, "restaurants." Revmo sells restaurant phone software, so label it vendor research. "Restaurants miss X percent of calls" is not a meaningful sentence.

Dental has the best-documented version of the same problem. Patient Prism scored 11,552,668 patient calls across 8,280 practice and DSO locations in calendar year 2025 and found 31 of every 100 callers hung up before reaching an agent. That is not "nobody picked up." It is hold time and routing, which is a queueing problem with a different fix. The same funnel shows where the recoverable money sits: of 100 calls, 21 book, 13 walk away without booking, 4.8 receive a callback, and 0.7 come back and book. Patient Prism sells call intelligence and the cohort is its own customers, which likely makes them more sophisticated than average, so the real market number is plausibly worse.

And this is where a claim you will hear falls apart. ServiceTitan's platform data from June 2022 puts the typical home services shop at a 42% call booking rate, rising from 24% at shops with fewer than five technicians to 59% at shops with twenty-five or more. Vendors routinely present that as evidence of a 58% miss rate. It is not. A booking rate is not an answer rate. The calls in the unbooked 58% were answered; they did not turn into jobs. CallRail, measuring the thing that actually is an answer rate, puts home services at 14%. Conflating the two produces a number four times too large, which is exactly why it circulates.

The gradient in the ServiceTitan numbers is the more useful finding anyway. A shop with fewer than five technicians books 24% and a shop with twenty-five or more books 59%. Missed calls at the small end are a capacity problem, not a discipline problem. The owner is on a roof.

How does call tracking actually work?

Two mechanisms, and you probably want both.

Static numbers assign one tracking number per channel. One on your website, one on printed material, one in your ads. Calls route through to your real line, and the system logs which number was dialled. Cheap, simple, and it survives someone writing the number down.

Dynamic number insertion swaps the number displayed on your website based on how the visitor arrived. Same visitor, different number, so the call ties back to a specific source and campaign. This is what lets you say a call came from paid search rather than from the website generally.

Dynamic insertion has a real cost worth naming. It changes a number on your page after load, which can conflict with structured data and with local listing consistency if implemented carelessly. The rule is that your Google Business Profile and your citations keep the real business number, always. Tracking numbers go on the website and in campaigns, never in your listings. That rule is not optional for a local business: Patient Prism's attribution data puts 90% of attributable patient calls in the Google ecosystem, with the Business Profile alone at 54%, and a profile with an inconsistent number is a profile you have damaged for a reporting convenience.

For dynamic insertion to say anything useful, your campaign tagging has to be right first. Consistent UTM tags are what the call tracking system reads to attribute the call. Sloppy tagging produces confidently wrong call attribution, which is worse than none.

What do I do with the recordings and the transcripts?

Listen to them, but with a purpose and a limit.

Twenty calls, once a quarter, is enough to learn more about your business than a year of dashboards. You will hear the question nobody on your website answers. You will hear the price objection your team handles differently every time. You will hear how long the phone rings before someone picks up.

Tag each call with an outcome. Booked, quoted, information only, wrong number, existing customer. That tagging turns your call log into a conversion funnel, and it is the only way to get from call volume to call quality.

Then look at when the calls arrive, because the timing usually explains the miss. Maple's data puts 68% of restaurant calls inside the lunch and dinner rushes, 28% at lunch and 40% at dinner. The phone peaks exactly when the floor is least able to answer it, which is a staffing finding rather than a marketing one.

One legal note, and it is not optional. Recording law varies by state, and California requires all-party consent under the Invasion of Privacy Act, so an announcement at the start of the call is required if you record anywhere in the Bay Area. Check the rule for every state you take calls from, not only your own.

Then resist the urge to draw conclusions from a small pile of calls. Twenty calls is a qualitative sample, useful for finding problems and useless for measuring rates. Small samples mislead in a specific and predictable way, and call data is where owners get fooled most often, because each call feels significant.

Why does the call number rarely match the ad platform's?

Because they are counting different things over different windows.

Ad platforms count calls initiated from an ad, inside their attribution window, using their model. Your call tracking system counts calls that arrived, with duration, from a number assigned to that source. Someone who sees an ad on Tuesday and calls from your website on Friday appears in one system and not the other.

Neither is lying. They answer different questions. Pick the one you will plan from, which should be your own call log, and treat the platform's number as directional.

The same logic applies to the form side of your site. Call tracking often reveals that people are phoning because a form failed them, which is a finding you can act on. Form abandonment analysis without expensive tools is the cheapest way to check that theory.

What to do this week

Count first. Before you buy anything, get your phone system or carrier to give you a report of inbound calls with answered status for the last thirty days. Most systems have it. That is your missed call baseline and it costs nothing.

Then add tracking numbers by channel. Start with three: website, one for paid, and one for print. Your Google Business Profile keeps the real number.

Set a duration threshold and write it down beside the number so nobody changes it quietly later.

Add answered and missed as two separate tiles on whatever you review weekly, and run the filtering arithmetic above once so you know what share of your raw count is real.

Decide what a call is worth to you, roughly, using the same discipline you would bring to reading a P&L line by line. A call value that is a guess is fine, as long as everyone knows it is a guess.

If this is part of a new engagement with an agency, get call tracking live in week one. It belongs in the short list of things to measure in the first ninety days, because starting it in month four means you have no before.

Questions about your calls?

Email me at eric@seod.com with your industry and how many people can pick up the phone during business hours. I will tell you the cheapest way to get your actual missed call rate this month without buying software, and what number would be worth worrying about in your category.

I answer these myself. It usually takes one short email and there is nothing attached to it.

More on measurement sits in the analytics library.