Skip to content
Guides · Attribution

Attribution models for lead generation: how to choose the right one

First click, last click, linear, time decay and more compared with a worked example, plus which model and window fit the question you need answered.

Daniel Kalcher13 min read

This guide is for anyone running ads to generate leads who wants to know which channel is worth the money: performance marketers, agencies, founders managing their own ad accounts. You'll see what the common attribution models do with one and the same lead, why the platforms' numbers never add up, and how to choose a model and window for the question you're trying to answer.

Each model has its own glossary page with the full definition. This guide is about the comparison: what shifts when you switch, and what decision you make afterwards.

What a model actually decides

Before someone fills in your form, they've often had several touchpoints with you: an ad, a search, a newsletter, a direct visit. An attribution model is the rule that splits one conversion across those touchpoints. It settles three things:

  • Which touchpoints count: only clicks and visits, or also ads that were merely seen.
  • How far back: the attribution window, say 30 or 90 days before the conversion.
  • How credit is split: all to one touchpoint, or shares across several.

The shares of a conversion always add up to exactly one lead. A model doesn't create leads, it moves them between channels. No model can prove which channel "really" brought the lead. It can only answer a clearly framed question consistently.

The seven models at a glance

ModelRuleAnswers the questionWeakness
First click100% to the first touchpointWho brings in new prospects?Ignores everything after
Last click100% to the last touchpointWhat happened right before the enquiry?Direct visits and brand searches take it all
Last non-direct click100% to the last touchpoint that isn't a direct visitWhich channel last brought them back?First contacts get nothing
Last paid click100% to the last ad clickWhich ad was involved last?Hides organic channels
LinearEqual shares for every touchpointWho was involved at all?Every touchpoint looks equally important
Position-based40% first, 40% last, 20% for the middleWho opens, who closes?Weights are a convention, not measured
Time decayThe closer to the conversion, the more creditWhat worked just before the decision?With long cycles, almost the same as last click

Last paid click has no glossary entry of its own: it looks for the last click on a paid ad among the touchpoints and gives it the whole conversion. It fits when you only want to compare paid channels with each other and deliberately leave organic visits out.

Then there's Google's data-driven attribution, which splits the conversion using a model Google learns from your past data. According to Google Ads Help it's the default for most conversion actions, and first click, linear, time decay and position-based are no longer supported in Google Ads. Google Analytics 4 removed the same four models in November 2023. Keep in mind that data-driven attribution only knows the touchpoints Google can see, and you can't recalculate its weights yourself. How GA4 and TSO differ beyond that is covered in TSO vs. Google Analytics 4.

One lead, seven answers

Say the operations manager of a regional logistics company is looking for fleet management software. Her path to a demo request:

  1. Day −21: She clicks a Meta ad and reads your home page.
  2. Day −14: She searches Google for "fleet software comparison" and clicks your search ad.
  3. Day −5: She clicks an article in your newsletter.
  4. Day −1: She sees one of your Meta ads in her feed but doesn't click.
  5. Day 0: She types in your address and books a demo.

Only clicks and visits count here, so the view on day −1 doesn't. With a 30-day window, all four clicks are inside. Here's how each model splits that one lead:

ModelMetaGoogle AdsNewsletterDirect
First click100%0%0%0%
Last click0%0%0%100%
Last non-direct click0%0%100%0%
Last paid click0%100%0%0%
Linear25%25%25%25%
Position-based40%10%10%40%
Time decay (7-day half-life)6.3%12.6%30.7%50.4%

Four models, four different winners. That's not a bug, it's the point:

  • First click tells you Meta found this prospect. Judge Meta by last click and you cut exactly the campaign that started the journey.
  • Last click credits the direct visit. That's useless for budget decisions, because you can't spend money on "direct".
  • Last non-direct click puts the newsletter first. Useful if you want to know what brings people back.
  • Last paid click, and Google's own reporting, would give the lead to Google Ads.
  • Time decay weights each click by 0.5 to the power of (days ÷ 7): the click 21 days back counts for an eighth of the direct visit. Meta gets very little, even though it made the first contact.

Scale this up to a month of leads and a campaign with "expensive" leads can turn into your best one, depending on the model. That's why choosing a model is a budget decision.

The attribution window

The window sets how long a click before the conversion still counts. It often changes the numbers more than the model does.

Take the example above with a 7-day window: only the newsletter and the direct visit are left. First click now credits the newsletter, and linear and position-based split 50/50 between newsletter and direct. Meta and Google Ads disappear from every model, even though both played a part. Last paid click finds no ad click at all; depending on the tool, the lead then goes to the last touchpoint or stays unattributed.

The platforms have their own defaults, which you need to know when reading their reports:

PlatformClick window (default)ViewsModel in the ad account
Google Ads30 days, adjustable from 1 to 901 day (engaged view: 3 days)Data-driven or last click
Meta7 days1 dayOwn attribution per ad set
LinkedIn90 days90 daysLast touch, click or view
TikTok1 or 7 daysoff or 1 dayOwn attribution per ad group

Sources: Google Ads conversion windows, Meta attribution setting, LinkedIn's conversion window and attribution model, TikTok attribution window. Since March 2026, Meta only counts link clicks as clicks; conversions after shares, saves and other interactions fall into a separate "engage-through" category (Meta for Business). These options change regularly, so check the current settings in your ad account.

For your own window, a rule of thumb: it should cover the time between first contact and conversion for most of your leads. For impulse offers 7 to 30 days is enough; for consulting, software and high-ticket services, think 90 days or more.

Why Meta and Google both claim the lead

Here's what happens in the example. Meta sees the view on day −1 and claims the lead through its 1-day view window. The click on day −21 falls outside its 7 days, but that doesn't matter. Google sees the click on day −14, which is inside its 30-day window, and counts the lead too. Both ad accounts report a conversion. Your CRM has one lead.

There are three reasons:

  • Each platform only sees itself. Meta knows nothing about the Google click, Google nothing about the Meta ad. Each gives itself 100% of the lead in its own model.
  • Views count. Meta and LinkedIn credit view-through conversions by default, and Google does for some campaigns. Buy a lot of reach and you get credited with many conversions, including from people who would have come anyway.
  • Different windows. With 7, 30 and 90 days side by side, older clicks count on one platform and not on another.

Say your CRM shows 100 leads for the month, Meta reports 70 and Google 60. That's 130 combined, 30 more than exist. Neither number is "wrong" by the platform's own rules, but you can't add them up and derive a cross-channel cost per lead from them. To see how big the difference is for you, use the attribution gap calculator.

Platform attribution vs. your own

The question isn't which number is right, but what you need it for.

Platform attributionYour own attribution
SeesIts own ads onlyAll channels, including organic
CountsClicks and often viewsUsually clicks and visits only
ModelFixed, partly opaqueYour choice, reproducible
Double countingYes, across platformsNo, each lead once
Use it forBidding and optimization inside the platformBudget across channels, cost per lead, ROAS

Both have their place. Meta's and Google's bidding algorithms learn from the conversions reported to them, so they need data that's as complete as possible, ideally sent through a Conversion API and including later steps such as qualified leads. The guide on Conversion API setup covers how. Deciding how much budget goes to Meta versus Google or LinkedIn, on the other hand, calls for a number that counts each lead only once.

Which model for which question

QuestionModel that fits
Which channels bring in new prospects?First click
What should I steer by day to day?Last non-direct click
Which ad was involved last?Last paid click
Which channels work together over a long journey?Linear or position-based
What works right before the decision, e.g. during a promotion?Time decay

How to settle on your model:

  1. Write down the question: spell out the decision the number will drive, for example "How do I split budget between Meta and Google?".
  2. Measure the journey: look at how many days and touchpoints typically sit between first contact and enquiry.
  3. Choose a window: pick one that covers most of those journeys. Too long beats too short.
  4. Set a default: one model for reports and budget, the same for everyone on the team. Last non-direct click is a good start if you have no specific question.
  5. Cross-check: once a month, compare with first click or position-based. If a channel loses in the default model but wins clearly in first click, it's an opener, not a candidate for switching off.
  6. Don't keep switching: a new model mid-quarter makes comparisons with last month meaningless. Switch deliberately and make a note of it.

Long cycles and B2B

In consulting, software or B2B services, weeks or months often pass between first contact and closed deal. That brings its own problems:

  • The first contact vanishes: Safari deletes script-set storage after 7 days without a visit. After that a visitor's history starts over and a later click looks like the first. Identity resolution helps as soon as someone identifies themselves with an email address or phone number.
  • Several devices and people: someone clicks the ad on their phone and enquires days later on a laptop. Without merging, those are two unconnected visitors. If an assistant clicks and the managing director enquires, even the best merging won't connect them, because they're two people.
  • The deal falls outside the window: if a deal closes four months after the first click and you attribute it with a 90-day window before the deal, that click drops out. Base the window on the gap from first contact to closed deal, not to the lead.
  • Leads aren't worth the same: a channel with lots of cheap enquiries looks good on leads and bad on deals. Judge channels on qualified leads and revenue as well. The guide on how to optimize for qualified leads shows how to get the platforms to do the same.

For long cycles, linear or position-based make good cross-checks, because they don't drop the early touchpoints. Time decay with a short half-life is a poor fit here: with a 7-day half-life, a click eight weeks back counts for less than one percent of a click on the day of the enquiry.

Clicks only, or views too?

View-through means someone sees an ad, doesn't click, and converts later. The platform still claims the conversion.

The case for it: video and reach campaigns have an effect without a click. Count only clicks and you undervalue them.

The case against it: a view proves little. Someone who already meant to enquire and scrolls through their feed will often see an ad along the way. With retargeting to past site visitors, it's almost guaranteed. View-through inflates the numbers of exactly those campaigns that reach people who are already close to deciding.

For comparing channels, click-only attribution is the sturdier basis: a click is an action, and every channel is treated the same way. If you want to know what reach campaigns achieve without clicks, a test with a holdout group (part of the audience doesn't see the ads) tells you more than view-through numbers.

Common problems and solutions

ProblemCauseSolution
Platforms together report more leads than the CRMEach platform claims the lead, partly through viewsDon't add platform numbers; split budget by your own attribution
Direct traffic gets lots of leadsLast click counts typed-in addresses and bookmarksUse last non-direct click as the default
Awareness campaign looks uselessJudged by last click onlyCross-check with first click or position-based
First contact missing on long journeysWindow too short or visitor ID deletedLonger window; use identifiers like email to merge visitors
Numbers jump after a changeModel or window changed mid-periodDocument the change; only compare periods with the same setting
Many leads unattributedMissing UTM parameters, blocked scripts, leads from other sourcesTag every ad with UTMs; check tracking; show unattributed leads openly
Google no longer offers first clickGoogle dropped rule-based models in 2023Use your own attribution tool for rule-based models

How to do it with TSO

TSO is a tool for your own attribution: it connects clicks on every platform with leads, deals and revenue. For the steps in this guide, that means:

  • All seven models from the table: first click, last click, last non-direct click, last paid click, linear, position-based (40/20/40) and time decay with a 7-day half-life. Windows: 1, 7, 30 or 90 days, or unlimited.
  • Switch without re-tracking: TSO stores every touchpoint, so you change model and window per dashboard, retroactively too, and see how the numbers shift. The default is last non-direct click with 90 days.
  • Clicks only, each lead once: TSO attributes conversions to clicks only, not views. Anything that can't be matched to a click appears as its own "unattributed" row in every report.
  • Merging devices: when someone identifies themselves with an email, phone number or external ID, TSO merges their visitors into one person, so the first contact isn't lost.
  • Deals, not just leads: deals and payments come in via webhook, API or Stripe and are attributed like any conversion, over the clicks in the window before them. For long cycles, choose a long window.

In parallel, TSO feeds the platforms through their Conversion APIs so their bidding keeps learning. More on models and windows on the attribution page.

Sources

How TSO helps

Frequently asked questions

Which attribution model is best?

None of them. Each model answers a different question. Last non-direct click is a solid default for day-to-day decisions, first click shows where new prospects come from, and linear or position-based work as a cross-check for long decisions.

Why do Meta and Google together report more leads than my CRM?

Each platform only sees its own ad interactions and claims a conversion as soon as one of them falls inside its window. Meta also counts conversions after a mere view by default. So both claim the same lead.

What's the difference between last click and last non-direct click?

Last click gives the conversion to the very last visit, even if someone just typed in your address. Last non-direct click skips those direct visits and credits the last channel before them, such as the ad or the newsletter.

What attribution window should I use for B2B?

Measure how many days pass between first contact and enquiry or closed deal, then pick a window that covers most cases. For consulting and software that's often 90 days or more; 7 days usually cuts off the first contact.

Does Google Ads still offer first click and linear attribution?

No. Since 2023, Google Ads and Google Analytics 4 only offer last click and data-driven attribution. If you want rule-based models such as first click or time decay, you need your own attribution tool.

Terms in this guide

Written by
Daniel KalcherCTO

Daniel is CTO and leads the technology behind tracking, attribution and conversion APIs at TSO.

More guides

All guides →

From the guide to clean measurement.

Set up TSO with us in a demo or request beta access.