If you run paid search for a dental group, a law firm, or a multi-location service business, you’ve probably had this conversation with yourself in the last six months: Why are my Performance Max campaigns underperforming when my conversion volume looks fine?
The answer is usually hiding in your phone system.
In May 2026, with Google having now fully retired First Click, Linear, Time Decay, and Position-Based attribution models, every Smart Bidding strategy on your account — Target CPA, Target ROAS, Maximize Conversions — is running on Data-Driven Attribution. That model is, as Google puts it, only as accurate as the conversion signals you give it. And for service businesses, the conversion that actually matters most is the one most marketers still aren’t tracking properly: the phone call.
This isn’t a small gap. For a typical dental practice running local search ads, somewhere between 35% and 60% of inbound leads come in by phone. For personal injury law firms, it’s higher. For real estate brokers, higher still. And yet most marketing teams are still feeding their ad platforms a steady diet of form-fill conversions while pretending the calls aren’t happening.
The platforms aren’t pretending. They’re optimizing for what they can see. And what they can’t see, they can’t reward.
The “lagging metric” trap
There’s an interesting shift happening in how the smartest revenue teams talk about call data. For most of the last decade, call tracking was a reporting function — you’d pull a monthly report, see that 47% of your conversions came from calls, nod, and go back to optimizing form fills because that was what the dashboard surfaced.
Industry analysts at companies like Marchex have started calling this the difference between lagging and leading indicators. A call volume report tells you what already happened. A real-time, AI-graded call signal piped back to your ad platforms tells the algorithm what should happen next.
Here’s the practical version of that distinction. Imagine two paid search managers working on the same dental account this week:
Manager A has Smart Bidding set to Maximize Conversions. The conversion event is “form submitted on /contact.” Every form-fill counts as one conversion, regardless of whether the person was a real prospect, a wrong-number tire-kicker, or a competitor’s intern. Smart Bidding learns: “Send more people to the contact form.” It does. CPL stays flat. Booked-appointment count stays flat. Revenue stays flat. Manager A starts blaming “the algorithm.”
Manager B has the same campaign, but every inbound call from a TrackNotion-assigned tracking number flows through automatic AI transcription, gets a quality score, gets tagged “qualified” or “not qualified,” and the qualified ones get pushed back into Google Ads as enhanced offline conversions with the full revenue value attached. Smart Bidding learns: “Bidders who clicked this ad called and booked a $4,200 implant consult.” It bids accordingly. CPL goes up. CPA on a qualified booked appointment goes down by 30%.
The campaigns are identical. The signal isn’t.
Why “more conversions” has become the wrong goal
Google’s own guidance has been increasingly direct on this point. Their advice for Performance Max and Smart Bidding now explicitly recommends differentiating the value of conversions — a phone call worth $50 should not be treated the same as a phone call worth $500, and an in-store visit worth $75 should not be lumped in with a low-intent newsletter signup.
The problem is that for the vast majority of agencies and operators, “differentiating call value” still means manual tagging in a spreadsheet at the end of the month. By that point, Smart Bidding has already spent two weeks bidding on the wrong audience.
This is where the architecture of modern call tracking has had to evolve — and where the gap between traditional call tracking and AI-powered attribution starts to widen sharply.
What “AI-powered” actually has to mean in 2026
A lot of platforms slap “AI” on their landing pages. Most of them mean “we transcribe the audio.” Transcription is table stakes. The actual work AI has to do, end-to-end, looks more like this:
Identify the source. Dynamic Number Insertion swaps in a unique tracking number for every visitor based on their channel, campaign, keyword, and even ad creative. The Google Ads click that generated the call is bound to the call from the moment it rings.
Transcribe accurately. Modern speech recognition is now reliable enough to handle accents, hold music interruptions, multiple speakers, and the chaotic acoustics of a dental front desk on a Monday morning.
Grade the conversation. This is the part most legacy systems still don’t do well. AI sentiment analysis isn’t just “happy or angry.” It’s parsing whether the caller asked about pricing, whether the rep handled the objection, whether an appointment was booked, and whether the caller’s tone shifted during the call. Each of these signals correlates differently with revenue.
Score the lead. A call gets a quality score not based on duration (a 12-minute angry call is not better than a 3-minute booked appointment) but on outcome plus conversation quality.
Push the signal back. This is the part nobody talks about enough. The qualified-call event has to flow back into Google Ads, Meta, TikTok, GA4, and the CRM with revenue values attached, ideally within hours. Otherwise Smart Bidding is still flying blind.
The fifth step is what closes the loop. It’s also the step where most call tracking implementations quietly fall apart, either because the integration was never built, or because nobody on the marketing team knew it existed and turned it on.
What this looks like for the practices and firms we built TrackNotion for
The pattern is remarkably consistent across the verticals we serve.
A dental practice running a $6,000/month Google Ads budget typically discovers — once dynamic number insertion is fully deployed — that something like 40% of their booked new patients came in via phone, not through the form. That’s a year of campaign optimization built on half the data.
A personal injury attorney we onboarded last quarter was paying close to $180 per form-fill lead. Once call tracking with AI grading was in place, the team realized that 65% of their actual signed cases originated as phone calls, that two specific intersection-billboard campaigns were generating most of those calls, and that one of their highest-spend keyword groups was generating volume but no qualified conversations at all. Reallocating the budget didn’t increase ad spend; it nearly doubled signed cases per month.
A regional marketing agency managing 14 dental offices found that the calls they thought were being missed at the front desk weren’t being missed at all — they were being answered, mishandled, and never followed up on. AI call grading and automated SMS alerts to the practice manager surfaced this in week one. Not a marketing problem. An operations problem with a marketing fingerprint.
The privacy angle nobody’s talking about
There’s one more thread worth pulling on, because it’s going to matter more in late 2026 and 2027 as state-level privacy laws keep tightening. The traditional response to “we need better attribution” has been “let’s collect more demographic and identity data on every visitor.” That approach is on a collision course with how privacy legislation is moving.
Call tracking through dynamic number insertion is genuinely different. It doesn’t require harvesting personal data from anonymous browsers. The attribution signal is bound to the channel, not the individual — which campaign drove the call, what was said on the call, what the outcome was. No tracking pixel resurrection. No cross-device fingerprinting. The conversation itself is the data.
This was deliberate when we built TrackNotion. The platform’s mission, from day one, has been to give marketing teams attribution depth without forcing them into the data-collection arms race that most of the adtech industry is still running.
Where to go from here
The blunt question to ask yourself this week: if you pulled your last 30 days of paid search reporting and stripped out every conversion that wasn’t a phone call, would your campaigns still look like they were working?
For most service businesses, the honest answer is no. The campaigns are working — the calls are coming in. The signal is just not making it back to the platforms doing the bidding.
That’s the gap call tracking with AI conversation intelligence is built to close. Smart Bidding doesn’t need more conversions. It needs better ones. And right now, the best ones are ringing your front desk while your reporting dashboard quietly ignores them.
If you’re running ads for a dental group, a multi-location practice, a law firm, or a marketing agency that handles any of the above, the next 14 days are a reasonable window to find out what your actual qualified-call number looks like. We’re still running complimentary setup for the first 100 trial users with a Fahrenheit Marketing partnership. The free trial runs 14 days. The math usually settles itself in the first 30.
Stop letting Smart Bidding starve.