A PI lawyer in Houston runs Google Ads for "car accident lawyer." Someone clicks the ad at 2:14pm, browses the site for six minutes, and calls the number on the landing page. Nineteen minutes later, the call ends with a signed retainer. That's a $400 call.
But where does the $400 actually come from? Which keyword? Which ad group? Which landing page variant?
Without attribution, the lawyer knows someone called. With attribution, she knows the call came from the keyword "houston car wreck attorney" on the mobile version of landing page B, that the caller spent 4 minutes on the FAQ before dialing, and that the intake specialist who answered was Maria, not Jake. The difference between "we got calls" and "we know exactly what's working" — that's the whole game.
That's call attribution. Not just "a call happened" but "here's the entire chain from first touch to billable event."
Why Phone Calls Are Harder to Attribute Than Web Forms
Web form attribution is easy — the submit button sends UTM parameters, gclid, session ID, and timestamp in the same HTTP request. Everything travels together.
Phone calls break this model. The visitor's browser session is one system. The phone call is another system entirely. Different device, different network, different everything. The call doesn't "know" where the visitor came from.
Annoying? Yes. But fixable.
The bridge is dynamic number insertion. DNI shows different phone numbers to different visitors based on traffic source. When a number rings, you trace it back to the pool assignment, then to the session, then to the UTM parameters. Chain complete.
But DNI is just the linking mechanism. Full attribution includes source capture, number assignment, call routing, disposition tagging, and billing qualification. Break any link and your attribution degrades.
The Attribution Chain: Six Links
Let's trace a call from ad click to invoice.
Link 1: Source Capture
For paid traffic: UTM parameters in the destination URL. utm_source=google, utm_medium=cpc, utm_campaign=pi-auto-houston. If you're running Google Ads, gclid auto-appends. For Meta, fbclid.
For organic traffic: referrer string. The browser sends the previous URL along with the request.
For direct traffic: nothing. No referrer, no UTM. Direct is the attribution graveyard.
Your tracking platform (VeloCalls, Ringba, CallRail, Invoca) reads these parameters on page load and attaches them to the visitor session. If your landing pages strip UTM parameters on redirect — which happens more than you'd think — the chain breaks here. I've debugged campaigns where 30% of calls showed as "direct" because a redirect dropped the query string. Maddening. Three hours of my life I won't get back, staring at Chrome DevTools watching query params vanish into the void.
For paid traffic cleanup, ClickzProtect catches invalid clicks before they hit your landing page.
Link 2: DNI Number Assignment
Visitor lands. The page loads with your main business number. Then your DNI JavaScript runs and swaps that number for a tracking number.
Source-based pools assign one number per traffic source. All Google Ads visitors see 555-0101. All Facebook visitors see 555-0102. Simple, cheap, tells you which channel.
Session-based pools assign a unique number per visitor session. More expensive (bigger pool), but ties the call to a specific session — including pages viewed, time on site, and exact ad click. Worth it if your calls are worth $50+. Probably overkill if you're tracking $15 leads.
The assignment gets logged: session ID + tracking number + source parameters + timestamp. If your DNI script fails to load (JS blocked, CDN issue), the visitor sees your main number and the call can't be attributed. Silent failure. No error message. Just... missing data.
For a deeper look, see our dynamic number insertion explainer.
Link 3: Call Routing
The visitor calls the tracking number. The call hits your routing layer.
In pay-per-call setups, the call enters a ring tree — buyers bid in real time, highest bidder gets the call, all in under 500 milliseconds. During routing, metadata gets enriched:
- ANI (caller ID) — who's calling
- DNIS (dialed number) — which tracking number they called
- Geographic data — area code lookup, carrier ID
- IVR path — which options the caller selected
This metadata plus the original source parameters — all of it travels with the call. Your routing platform matches the DNIS to the session record and pulls the attribution data.
VeloCalls handles this with real-time bidding, visual IVR builder, and multi-carrier support (Twilio, Telnyx, Bandwidth). Managed carriers start at 4¢/min, BYOC at 2¢/min, both stepping down as volume scales.
Link 4: Agent Connect
The call connects to a live agent. At agent connect, the billable timer usually starts. Your platform logs which agent answered and the connect time.
In legal intake, agents qualify: "Were you injured? When? Have you spoken to other attorneys?" In home services, they're scheduling. What happens on the call doesn't directly affect attribution — attribution is about source, not outcome. But the next link ties them together.
(Side note: I used to think agent connect was the simple part. Then I watched a call center where half the "connects" were agents picking up, saying hello, and immediately getting hung up on. Billable timer started, caller gone. Attribution was perfect. The metric was useless.)
Link 5: Disposition Tagging
Call ends. Someone (or something) tags it.
Disposition codes categorize the outcome: qualified lead, not qualified, wrong number, voicemail, appointment set, sale closed. Disposition can be manual (agent selects post-call), automated (AI tags based on transcript keywords), or hybrid.
VeloCalls includes AI conversation intelligence — transcription at 4¢/min, AI call summary at 10¢/call, sentiment analysis at 5¢/use. These feed into disposition tagging.
The disposition attaches to the call record alongside all attribution data. Now you know: this call came from [Google Ads → pi-auto-houston → car-wreck keyword], lasted 19 minutes, agent was Maria, outcome was signed retainer.
Link 6: Billing Event
For pay-per-call, one more step: does this call qualify for payment?
Billing rules evaluate:
- Duration threshold — did it exceed 90 seconds?
- IVR completion — did the caller pass required gates?
- Geographic match — was the caller in the target area?
- Duplicate check — has this ANI called within the duplicate window?
If all conditions pass, a billing event triggers. The billing record references the full attribution chain. When the invoice arrives, you can trace any line item back through every link.
For more on billing mechanics, see our pay-per-call billing guide.
Where Attribution Breaks
Perfect attribution doesn't exist. I wish I could tell you otherwise. Honestly, 85-90% attribution is good — and that number will annoy the hell out of anyone who spent six figures on their data stack expecting perfection. Know where the gaps come from so you can decide which ones are worth fixing.
Visitor calls from a different phone. They browse on mobile, call from their office landline. The session was on one device; the call came from another. No way to link them without login-based identity resolution. Accept this loss in consumer-facing verticals. In B2B (where office numbers are known), you can sometimes match on company ANI patterns.
Cached pages show stale numbers. If your CDN caches HTML aggressively and your DNI swap doesn't run fast enough, some visitors see old tracking numbers. You'll get calls that don't map to recent sessions. Check your cache headers. Set DNI assets to no-cache or short TTL. This one bit me on a client site running Cloudflare with aggressive caching — took three days to figure out why 15% of calls showed orphaned.
Session expires before call. Visitor browses at 10am, bookmarks the page, calls at 3pm. Their session-pool number may have recycled to another visitor. The call attributes to whoever has that number now — or shows as unattributed if the number's back in the idle pool. Extend session timeouts in high-consideration verticals. Legal, home improvement, insurance — 2-4 hours isn't crazy.
JS blockers kill DNI scripts. Privacy browsers and ad blockers sometimes block tracking JavaScript. The visitor sees your main number. No attribution. You can't fix this one; just measure it. Compare total calls to your main line vs. tracking numbers to estimate your blocked rate. Tools like JustBrowser help test how your pages render across different browser configurations. (And before someone emails me: yes, you could do server-side number injection. No, most teams won't. It's a pain.)
Call forwarding strips ANI. Some phone systems forward calls in ways that replace the original caller ID with the forwarding number. You lose the true ANI. Test your forwarding chain end-to-end before you trust the data.
UTM parameters dropped on redirect. Very common with URL shorteners, tracking redirects, and CMS configurations. I spent a frustrating week diagnosing a 25% "direct" rate that turned out to be a 301 redirect stripping parameters. The fix took 10 minutes — the debugging took forever. Don't assume your pages preserve query strings; verify it.
Attribution Models
Last-touch credits the final session before the call. If someone clicked a Google Ad Monday, came back organic Wednesday, and called — organic gets credit. This is what pay-per-call billing uses. Simple, defensible. Is it "fair"? Depends who you ask. The Google Ads team hates it. The SEO team loves it. Pick a side or waste time arguing.
Multi-touch distributes credit across touchpoints. Useful for internal optimization, but keep it separate from invoicing. The moment you try to bill on multi-touch, every buyer will dispute the math.
JustAnalytics handles web session tracking for multi-touch analysis without GDPR consent banners.
Common Mistakes
Trusting platform data without spot-checking. Pull recordings. Listen. Verify the disposition matches what you heard. Automation is good; blind faith in automation is expensive. I make it a habit to audit 10 random calls per week. The patterns you catch are worth the time. For managing client payments and payouts across campaigns, VeloCards handles virtual card issuing.
Ignoring "direct" calls. If 20% of your calls show as direct/unattributed, that's 20% of your data missing. That's not a rounding error — that's a broken funnel. Investigate. Is it mobile crossover? Cached pages? Stripped UTMs? Each cause has a different fix.
Using source-based DNI when you need session-based. If every call is worth $50+, you want session-level attribution. The extra number cost — maybe $50-100/month for a reasonable pool — pays for itself in optimization data. You can't bid up a high-converting keyword if you don't know it's converting.
Not tagging dispositions consistently. "Qualified" means different things to different agents. One person's "warm lead" is another person's "callback." Define your codes precisely. Run calibration sessions monthly. Inconsistent tagging = garbage analysis. I've seen entire quarterly reports invalidated because two agents had different ideas about what "qualified" meant.
Skipping IVR for attribution's sake. IVRs add friction — some callers drop off. But IVR completion data ("caller pressed 1 for roofing, 2 for HVAC") enriches attribution. It's a tradeoff, not a no-brainer. Test IVR vs. direct connect and measure qualified call rate, not just raw volume. My take? If your calls are worth $100+, the friction cost is worth the data. Under $50? Maybe skip it.
Frequently Asked Questions
What data does call attribution actually capture?
Full attribution captures: traffic source (from UTM parameters or referrer), the specific ad or keyword (utm_campaign, utm_content, gclid), the tracking number that was displayed, caller ANI, call timestamp, IVR path, agent who answered, call duration, disposition code, and recording URL. Good platforms tie all of this to a single session ID so you can see the visitor's entire journey — pages viewed, time on site, call outcome.
How accurate is call attribution with multiple touchpoints?
Attribution accuracy depends on your tracking setup. Last-touch (the session that triggered the call) is near 100% accurate with proper DNI. Multi-touch is messier — you need cross-session identity resolution, which means cookies or login-based tracking. Most pay-per-call operators use last-touch for billing and multi-touch for internal analysis. Perfect multi-touch attribution doesn't exist; pick a model and be consistent.
What breaks call attribution?
Common breaks: visitors calling from a different phone than they browsed on (number mismatch), cached pages showing stale tracking numbers (CDN caching issues), callers who write down the number and call later (session expired), JS blockers preventing DNI script execution, and forwarded calls that strip ANI metadata. You can't fix all of these — 5-10% attribution loss is normal even in good setups.
How does attribution differ between inbound and outbound calls?
Inbound attribution tracks the source that drove the visitor to call you — DNI handles this. Outbound attribution tracks which campaign list a call came from and ties it to disposition outcomes. Outbound is simpler (you know the source because you loaded the list) but requires consent verification and TCPA compliance. Inbound is harder to attribute but cleaner legally.
Resources
- Dynamic Number Insertion Explained — DNI setup, pool types, JavaScript implementation
- How Pay-Per-Call Billing Works — billable duration, duplicate windows, disputes
- ClickzProtect — click fraud detection for paid traffic
- JustAnalytics — session-level web analytics
Try VeloCalls for Your Vertical
Pay-per-call platform built for HVAC, plumbing, roofing, PI lawyers, Medicare brokers, and insurance. Smart routing, real-time bidding, visual IVR builder, AI conversation intelligence (transcription 4¢/min, AI summary 10¢/call). Per-minute pricing — Managed starts at 4¢/min, BYOC at 2¢/min, both drop as you scale.