AI Offline Conversion Tracking: Full Guide

Feed Your Ad Account the Full Funnel

Most lead-gen accounts fire a conversion on the form fill, so their ad accounts learn to buy more form fills, junk included. AI agents turn every call, text and transfer into a signal you can trust.

Last updated November 6, 202614 min readBy , CEO, Plura AI

What happens to 1,000 web leads?

200 of them have phone numbers that will never ring. Here's what your ad account learns in each case.

Real buyer Dead number counted as a conversion Dead number removed Real buyer thrown out
ANo validationPixel fires on every form fill
800 real buyers800200 junk200
200junk conversions
0buyers lost
Nonepost-call data
Your ad account learns from 200 dead numbers
BPhone-validation APIScore the number, then fire
760 real buyers760160 junk160
160junk conversions
40buyers lost
Nonepost-call data
Still learns from 160 dead numbers, and loses 40 buyers
CPlura AICall in seconds, fire on real contact
800 real buyers800200 removed200
0junk conversions
0buyers lost
Every stagepost-call data
Clean signal, plus every stage after the call

Assumes about 20% dead numbers, the average Plura sees across more than 1 million web lead phone numbers. Scenario B uses a validation API that catches about 20% of dead numbers with a 5% false-positive rate, from Plura's analysis of more than 1 million calls.

Figure 1. What happens to 1,000 web leads with no validation, a phone-validation API, and Plura AI.
On this page
  1. 01Key takeaways
  2. 02Form-fill conversions buy junk
  3. 03Phone validation APIs guess
  4. 04SIP codes expose bad numbers
  5. 05Call dispositions kill conversions
  6. 06AI funnel conversion events
  7. 07AI voice agents + SMS/RCS
  8. 08Live transfers die on hold
  9. 09Send offline conversions
  10. 10Smart Bidding needs volume
  11. 11AI lead quality signals
  12. 12Offline conversion checklist
  13. 13Free AI conversion audit
  14. 14How Plura AI automates it
  15. 15FAQ
  16. 16References

01Key takeaways

  • 1A form fill is not a conversion. About 1 in 5 web lead phone numbers never ring,[1] and ad-fraud firm Anura puts fake leads from online campaigns at 20–25%.[2] Count them as conversions and your ad account learns to buy more of them.
  • 2Pre-call phone scoring APIs catch only about 20% of bad numbers and flag about 5% of good ones as bad. The other 80% of bad numbers keep feeding bad signals to your ad accounts and your lead sources.
  • 3The first call is the real validation. The network's SIP response tells you in seconds whether a number exists.
  • 4Most marketers send nothing back after the form fill. The better ones send transfers and appointments. Even that is too little data to properly optimize an ad account, and everything in between gets lost to inconsistent human dispositions.
  • 5AI agents disposition every conversation, every time. Human dispositions are hit-or-miss, and feeding them to your ad account is as damaging as feeding it dead numbers. With AI you can send the whole funnel: contact, right party, qualified, transfer, and even what happened after the transfer.
  • 6Optimize on the middle of the funnel, not the end. Meta needs about 50 conversions per ad set a week, and 50 transfers is slow and expensive to get. Clean mid-funnel events like right-party contact and qualified lead deliver that volume, so your ad account learns faster and starts finding buyers sooner.

02Form-fill conversions train your ad account to buy junk

Your ad platform optimizes toward whatever you call a conversion. Call a form fill a conversion, and it buys more form fills, junk included.

53%of web traffic is botsThales, 2026 Bad Bot Report[3]
~20%of web lead phone numbers never ringPlura, 1M+ numbers[1]
20–25%of online campaign leads are fakeAnura[2]

The usual suspects: bots, form farms, typos, and affiliates recycling nine-month-old leads as new.

Why it compounds

Ad platforms chase the cheapest version of whatever you call a conversion, and junk is usually cheaper. Start at 10–15% bad numbers, and the algorithm serves you more every month.

From the field · Matt Beucler, Plura founder

$5.4M in Meta leads. The junk kept growing.

10%
12%
14%
25%+
StartDead numbers counted as conversionsEnd
  • $1M+ spent on leads that could never be reached
  • Phone scoring barely dented it
  • The fix: call every lead in real time, and never fire a conversion on a number that doesn't ring
Figure 2. Dead numbers counted as conversions grew from 10% to more than 25% on $5.4M in Meta leads.

03Phone scores are guesses. A ring is proof.

Phone validation APIs guess from old data, and phone numbers change daily. Set them strict and you throw out real buyers. Set them loose and junk eats your ad spend.

Plura case study · 1M+ outbound calls
~20%of dead numbers caught by the two leading phone-validation APIs
~80%of dead numbers still counted as conversions
5–7%of real buyers flagged as bad, even by the better API
Methodology: more than 1 million outbound calls placed by Plura AI, compared with pre-call validation results from the two most widely used phone-validation APIs across multiple date ranges. "Dead" means the network returned a response showing the number had no destination.
The API is cheap. The false positives aren't.

A mortgage company spending $50,000 a month on purchase leads at about $100 each:

$50,000a month in leads
÷
$100per lead
=
500leads a month
20real buyers thrown out · $2,000
+
80dead numbers still counted · $8,000
=
$10,000a month working against you

That's $120,000 a year, a fifth of the budget. And losing even 5–10% of your real buyers is most companies' entire margin.

Assumes ~20% dead numbers, an API that catches ~20% of them, and a 5% false-positive rate on the good ones. Illustrative.

Who gets hurt

The lead buyerFights vendors over returns. The vendor already spent the ad money, so everyone loses.
The sales floorReps burn their daily leads on dead numbers, watch the rep next to them get a great one, and stop trusting the leads.
The media buyerThe client gets the junk traffic, and the agency loses the client.

04SIP codes expose bad phone numbers in seconds

Every call gets a SIP code back from the phone network. Some mean "try again." Others mean the number is dead: disconnected, mistyped or fake.

0 secLead posts
2–3 secPlura dials
Seconds laterSIP code returns
InstantlySignal goes to your ad account

You know if the number exists before the ad platform finishes counting the click.

SIP responseWhat Plura does
404 / 604: disconnected or nonexistent numberSuppressed immediately
Carrier declines, busy, temporarily unavailableRetried under code-specific rules

Not every failed call means a dead number. Plura's retry rules are tuned on more than 100 million calls, so dead numbers come out fast and recoverable buyers don't get thrown away.

The rule for 2027

Never fire a conversion off a form fill. Fire it off the first real signal: a connected call, a reply, a confirmed right party.

05Offline conversions die at the call disposition

A call disposition is the outcome an agent records after a call: not interested, transferred, wrong number.

So why don't more marketers send offline conversions? Because people record the data. Agents forget, invent new call disposition codes, or skip them. Most marketers send nothing back. The best send only transfers and appointments. An AI call center dispositions every call, the same way, every time.

Human dispositions

  • Recorded when someone remembers
  • Definitions drift between agents and shifts
  • Entered minutes or hours after the call
  • Usually nothing gets sent; at best, transfers and appointments
  • Hard to audit after the fact

AI dispositions Plura AI

  • Recorded on every conversation, every time
  • One definition per event, applied the same way
  • Fired in real time as the outcome happens
  • Every stage of the funnel can be sent
  • A recording and transcript behind every event

Transfers alone are a handful of conversions a week. Every stage is that many before lunch.

06AI funnel: an offline conversion event for every stage

AI agents work every lead across calls, texts and RCS. Each outcome fires an event the moment it happens, with the lead's click ID attached, so the ad platform can tie it to the click, ad and campaign that produced it.

Top of funnel

Is this a real person?
number_ringscontact_connectedright_party_contact
Fired byOutbound callInbound callCallbackSMS replyRCS tap

Mid funnel

Is this a real buyer?
intent_confirmedhigh_intent_buyerqualified_leadrcs_engaged
Fired byVoiceSMSRCS

Bottom of funnel

Did it move toward a sale?
live_transfertransfer_goal_metappointment_set
Fired byVoiceCallbackRCS booking

The top of the funnel gives you the most volume, and the bottom gives you the strongest signal. A lead that reaches a live transfer should already have sent number rings, contact connected and every stage in between. Meta's CRM integration asks for every stage as it happens, including the first one, and it gives your campaigns a full funnel to learn from.[4]

07AI voice agents + SMS/RCS: message first, then call

The real power is in the handoff: each channel does the part it's best at, and every step fires its own event.

Get consent before the first call. The FCC treats AI voices as artificial voice under the TCPA, so marketing calls need the lead's prior express written consent. Capture it on the form, store the proof with the lead, and honor every opt-out. Our TCPA compliance guide for AI calling covers the rules, and TCPA opt-out rules for AI agents covers how to honor a stop.

11:18
Northline Law
Text Message · RCS
Today 11:18 AM
Hi Sarah, it's Northline Law. We got your accident claim. Two quick questions?
Sure
contact_connected
Were you hurt, and was the other driver at fault?
Yes and yes. Rear-ended last Tuesday.
qualified_lead
You may have a strong case. Free for a quick call with an attorney?
Yes, call me now
high_intent_buyer
Northline Law calling…Branded caller ID · seconds after her yes
number_rings
Connected to attorney
live_transfer
+Text Message · RCS
Personal injury (auto accident)
70%answer rate on the AI's follow-up call
95%conversion rate on those transferred calls

The text did the qualifying. By the time the AI called, she'd already said yes. It became the client's highest-converting call type.

Figure 3. An RCS conversation where each step fires an offline conversion event, ending in a live transfer.

Missed call, callback later

Day 0AI calls, leaves a voicemail
Day 0Text with one-tap callback
Day 14Lead calls back
Day 14AI qualifies and transfers

Late callbacks still tie back to the original click, within each platform's import window. Put your AI agent's number anywhere: your website, emails, mailers. Known leads match their original click by phone number; new callers from offline campaigns need their own tracking number.

08Live transfers die on hold. Call timers never notice.

35–40%of transfers never reach a human. They hang up on hold.

Call timers were built for billing leads, not optimizing ad accounts. Counting every transfer is no better. Both wreck your attribution. Plura listens to the transfer instead, so it knows what actually happened.

Call AThe buyer you lose
AI qualifies · 2:10 Hold · 0:45
Timer45 seconds on the transfer. No conversion.
PluraQualified, then abandoned in queue. Conversion fired.
Call BThe junk you pay for
Auto attendantIVR Human · billing question · 2:50
Timer3:20 on the transfer. Conversion fired.
PluraCustomer-service call, not a sale. No conversion.
Figure 4. Call timer vs. Plura AI on two live transfers.

Red marks the wrong call, blue the right one. In Plura's data, duration timers misjudge about 1 in 3 transfers, in both directions.

What Plura hears on every transfer

Hold musicAuto attendantHuman picks upRequalified, then sent to a closerStraight to a closerCustomer service handled

You set the rule

Fire the transfer conversion when QualifiedAND( Human connectedORAbandoned in queue)

A good transfer, transfer_goal_met, goes one step further: a human connected, the buyer engaged, and they agreed to a next step.

Bill on duration, optimize on qualification

Keep billable duration (90 seconds, 2 or 3 minutes) for paying partners. Send qualification to your ad account.

09How to send offline conversions to Google, Meta, TikTok and more

Hooking Plura up takes one field. Pass the ID with the lead, and every outcome flows back to where you need it.

InLead + one IDClick ID, attribution ID or affiliate ID
Plura AIWorks the lead
CallsSMSRCS
BestAttribution platformFirst-party tracking feeds every ad platform
GoodAd platforms directPixel or Conversions API
AlsoAffiliatesPostback by affiliate ID and sub ID
Pass us one ID. That's the integration.
gclidfbclidttclidmsclkidli_fat_idtwclidAttribution IDAffiliate / sub ID

Pixels and direct uploads Good

  • Works, and Plura supports it
  • Browser privacy, ad blockers and iPhone protections drop data
  • Click IDs on your landing page can get stripped or lost
  • One setup per ad platform

First-party attribution Recommended

  • Tracking runs on your own domain, so far less gets blocked
  • One ID covers every ad platform
  • Shows the whole journey, not just the last click
  • Works with Cometly, Hyros, Triple Whale, Northbeam and similar tools

How we run it at Plura. We pass the Cometly ID into Plura with each lead. Plura sends every call, text and RCS outcome to Cometly, and Cometly feeds the ad platforms.

Native lead forms (Meta Instant Forms, TikTok and Google lead forms) carry the platform's own lead ID, so matching is near-certain. Website forms depend on the click ID surviving the trip, which is the strongest case for first-party tracking.

Platform-by-platform details

Google Ads

gclid · gbraid · wbraid · hashed email/phone

Use enhanced conversions for leads, Google's upgraded version of offline conversion import. Google recommends lead-gen advertisers use the Qualified lead or Converted lead goal. Since June 15, 2026, these uploads run through the Data Manager API.[5][6]

Meta

fbclid · _fbc · _fbp · lead ID · hashed email/phone

The Conversions API handles offline events. For Instant Form leads, the Conversion Leads integration takes your stage names and optimizes on lead quality. Meta reports about 15% lower cost per quality lead for campaigns set up this way.[4][7]

TikTok

ttclid · TikTok lead ID · hashed email/phone

The Events API takes web and offline events server-side. For TikTok lead forms, CRM events tie each stage back to the original lead ID.[8]

Microsoft Advertising

msclkid

Create an offline conversion goal, then upload against the Microsoft Click ID by file or API. Wait two hours after creating a goal before sending conversions to it.[9]

LinkedIn

li_fat_id · hashed email · CRM ID

The Conversions API takes offline outcomes, such as a call taken or a meeting booked, against a conversion rule you set up in Campaign Manager.[10]

X

twclid · hashed email/phone

The Conversion API sends events server-side, matched on the X Click ID or hashed identifiers.[11]

Suggested event mapping

EventGoogle AdsMetaTikTok
contact_connectedCustom conversion actionContact / lead stageContact
right_party_contactCustom conversion actionLead stageCustom event
intent_confirmedCustom conversion actionLead stageCustom event
high_intent_buyerCustom conversion actionLead stageCustom event
qualified_leadQualified lead goalLead stageCustom event
rcs_engagedCustom conversion actionCustom eventCustom event
live_transferCustom conversion actionLead stageCustom event
transfer_goal_metConverted lead goalLead stageCustom event
appointment_setCustom conversion actionScheduleSchedule

Suggested mappings only. Your media buyers make the final call for each account.

10Smart Bidding needs volume: the cost of learning

Google's Smart Bidding and Meta's AI-powered bidding both need volume to learn. A campaign that only reports transfers often never sends enough events to get out of the learning phase. The full funnel fixes that: there's an event with enough volume at every stage of a campaign's life.

What 50 conversions costs you, home-services example
Optimize on bookings$17,50050 × $350 per booking. Weeks of spend on weak or no signal before it learns.
Optimize on form fills$87550 × ~$17.50 (1 in 20 books). Cheap, but it trains on junk.
Optimize on qualified leads$3,50050 × ~$70. Real buyers, at a volume you can afford.

Illustrative. Assumes about 1 in 4 form fills qualify. Most companies can't spend $17,500 a week waiting for an ad set to learn.

Figure 5. What it costs to reach 50 conversions, by optimization event.
~50
optimization events per ad set per week is what Meta looks for to exit the learning phase.[12] Below the red line, the ad set stays "Learning limited."
contact_connected
110
right_party_contact
95
qualified_lead
50
live_transfer
25
transfer_goal_met
12
Illustrative weekly events for one ad set receiving 250 leads a week. Your rates will differ; the shape won't.
Figure 6. Weekly events per ad set by funnel stage vs. Meta's ~50-event learning threshold.
Launch

contact_connected or right_party_contact

Highest volume. Cuts bad-number traffic first, and gives a new campaign enough events to learn.

Scale

qualified_lead

Move down once you're clearing the volume threshold. Now you're buying real buyers, not just real people.

Optimize

transfer_goal_met

The strongest signal before a sale. The one to scale once volume allows.

Keep sending every stage the whole time, even after you switch the optimization event. The upper stages still help the platform model your funnel, and they're your early warning when a source starts to slip.

11AI lead quality signals: credit bad leads, not abandonment

Not every outcome belongs in your ad account. These signals go to your own reporting and your lead partners, so you can clean up sources, sub IDs and placements. Don't send them to ad platforms as conversions. For how this plays out by industry, see our lead intelligence guides for mortgage, solar and legal.

SignalMeaningWhy it helpsCredit the lead?
bad_numberDisconnected, invalid or not in serviceThe fastest way to find junk traffic by sourceCredit
wrong_personThe person who answered isn't the leadSpots fake or mistyped submissionsCredit
duplicate_leadYou already have this personShows overlap and resold leadsCredit
not_interestedRight person, not buyingShows weak-intent ads and pagesNo
opt_outAsked not to be contacted againFlags low-intent sourcesNo
out_of_areaOutside where you sell or serveTightens geo targetingPer your terms
no_contactNo live contact after the full sequenceSeparates unreachable from bad numbersPer your terms
transfer_not_metTransfer graded not good, with the reasonShows where leads fall apart after transferPer your terms
abandoned_in_queueQualified, then hung up on holdYour queue lost a good lead. Can still fire your transfer conversion (section 8)No credit

Credit bad leads, not abandonment

When a qualified lead hangs up in your queue, the recording proves the lead was good. Credit dead numbers and wrong people; don't credit your own hold times. Typical terms shown; set yours with each partner.

12AI offline conversion tracking checklist

13Free AI prompt: audit your Google Ads and Meta conversions

Paste this into ChatGPT, Claude or Gemini. It interviews you about your current setup and tells you where your conversion data is costing you money.

offline-conversion-audit.txt
You are a paid media and attribution auditor. I run lead generation campaigns, and I want to know whether my conversion tracking is training my ad accounts correctly.

Ask me one question at a time about:
1. Which event I currently fire as a conversion on each platform I buy on (Google Ads, Meta, TikTok, Microsoft, LinkedIn, X).
2. Whether I capture click IDs (gclid, gbraid, wbraid, fbclid, _fbc, _fbp, ttclid, msclkid, twclid, li_fat_id) and hashed email and phone with every lead.
3. What share of my leads are bad phone numbers, and how and how fast I find out.
4. Whether I use pre-call phone scoring, and whether I've measured how many good numbers it wrongly flags.
5. Which post-lead outcomes I send back (contact, right party, qualified, transfer, appointment, sale), how fast, and who records them.
6. How consistent my dispositions are across agents and shifts.

Then give me:
- A score out of 10 for my conversion setup.
- The three biggest gaps, ranked by likely wasted ad spend.
- A stage-by-stage event plan, including which event each campaign should optimize toward today.

Source: Plura AI, "AI Offline Conversion Tracking: Full Guide," https://www.plura.ai/guides/ai-offline-conversion-tracking

14How Plura AI automates offline conversion tracking

Plura AI works every lead with omnichannel AI agents across voice, SMS and RCS, and turns every outcome into a conversion event you can count on: auditable, repeatable, and fired every single time. See how Plura AI voice agents work, or see pricing.

2–3 second speed to leadThe AI dials as soon as a consented lead posts during calling hours, so bad numbers surface in seconds.
SIP-level number validationWe read the network response on every call and suppress numbers that have no destination.
Every stage, in real timeEach funnel event posts back the moment it happens, with your click IDs attached.
AI-graded transfersEvery transfer is recorded, transcribed and graded, so only good transfers count as wins.
True omnichannelCall, text, voicemail, inbound and 24/7 RCS with calendar booking, in one platform.
Our own carrierPlura Connect is our FCC-licensed carrier, so we control routing, numbers and call data end to end.
Talk to a Plura specialist

15Frequently asked questions

What is offline conversion tracking?

Offline conversion tracking sends outcomes that happen after a lead form, such as a connected call, a qualified lead, a live transfer or a booked appointment, back to your ad platform. The platform matches each outcome to the click that produced it and optimizes toward people who reach that outcome.

Why shouldn't I fire a conversion on a form fill?

A form fill counts bots, bad numbers, typos and recycled leads the same as real buyers. In Plura's analysis of more than 1 million web lead phone numbers, about 20% never ring. Reporting those as conversions trains your ad account to find more of them.

Do phone validation APIs catch bad numbers?

Some. In a Plura analysis of more than 1 million outbound calls, the two leading phone-validation APIs caught roughly 20% of the numbers that never connected, and both flagged some numbers that did connect. Calling the number and reading the network response is the definitive test.

What is a SIP code?

A SIP code is the response the phone network returns to every call attempt. Codes such as 404 Not Found and 604 Does Not Exist Anywhere mean the number has no destination. Others, like a busy signal, mean try again later.

Which offline conversion events should I send to my ad accounts?

Send every funnel stage in order: number rings, contact connected, right-party contact, intent confirmed, high-intent buyer, qualified lead, RCS engaged, live transfer, transfer goal met and appointment set. Keep quality signals such as bad number and opt-out in your own reporting, not your ad accounts.

Which event should my campaigns optimize toward?

Start with a clean mid-funnel event such as right-party contact or qualified lead. Meta looks for about 50 optimization events per ad set per week, and bookings or transfers are usually too slow and expensive to reach that. Move down to a graded good transfer as volume allows.

Can an AI call center send offline conversions to Google and Meta?

Yes. An AI call center dispositions every call, text and transfer in real time, so each outcome can be sent back as an offline conversion. Pass the lead's click ID or attribution ID with the lead, and every stage, from the first ring to a graded transfer, flows to Google, Meta, TikTok or your attribution platform.

Why are AI agents better than human agents for offline conversion data?

Human agents disposition inconsistently: they forget, invent new dispositions or skip them. AI agents disposition every conversation the same way, in real time, with a transcript behind every event, so the data is complete, repeatable and auditable.

16References

  1. Plura AI internal analysis of more than 1 million web lead phone numbers. Rates vary by lead source; higher-quality sources run lower.
  2. Anura, "How Many Fake Leads Come From Form Bot Fraud?" (fake lead rates in online ad campaigns typically 20–25%), and Anura's Global Ad Fraud Report, May 2025. anura.io · report announcement
  3. Thales, 2026 Bad Bot Report (bots were 53% of web traffic in 2025), as reported by IT Brief UK, April 30, 2026. itbrief.co.uk · thalesgroup.com
  4. Meta for Developers, Conversion Leads integration and payload specification. developers.facebook.com
  5. Google Ads Help, "About qualified leads and converted leads" and "Offline conversion imports FAQs." support.google.com
  6. Google Ads Help, "Configure the Google tag for enhanced conversions for leads." support.google.com
  7. Click Here Digital, "Meta's Shift to Lead Optimization," citing Meta's reported results for CRM-connected Instant Form campaigns. clickheredigital.com
  8. Datahash documentation, TikTok CRM Events API and Events API. datahash.com
  9. Microsoft Learn, "ApplyOfflineConversions Service Operation." learn.microsoft.com
  10. Datahash documentation, LinkedIn Offline Conversions API. datahash.com
  11. X Developer Platform, Conversion API for Web. docs.x.com
  12. Code3, "Meta learning phase," on Meta's roughly 50 optimization events per ad set per week. code3.com