Lecture 14: Attribution Models: Last-Click, Data-Driven, and Multi-Touch

Short answer: An attribution model is simply the rule your reporting uses to decide which click (or clicks) get credit when a conversion happens. Last-click gives 100% of the credit to the final touchpoint before conversion. First-click gives it all to the first. Linear and time-decay models split credit across every touchpoint in a journey. Data-driven attribution (DDA) — now Google's default and, since 2023, mandatory in most Google Ads accounts — uses machine learning to look at your account's actual converting and non-converting paths and assign credit based on which touchpoints statistically moved people toward conversion. None of these models is "the truth." They are lenses. Choosing the wrong lens for your business (for example, judging a 45-day B2B sales cycle with last-click) will systematically make your top-of-funnel keywords look worthless and your branded search look like a hero, when in reality the non-branded click did the hard work and the branded click just closed the deal.

What You'll Learn in This Lecture

• Why the same campaign can look "profitable" or "wasteful" depending only on which attribution model you view it through
• How last-click attribution works, why it dominated for a decade, and exactly where it breaks down
• First-click attribution and when it is useful for evaluating top-of-funnel and awareness campaigns
• Linear and time-decay models as practical middle grounds between the two extremes
• How Google's data-driven attribution (DDA) actually works under the hood, including converting vs. non-converting path analysis
• Why DDA is now the default — and in many accounts the only available — model in Google Ads
• What a real multi-touch SEM journey looks like, including branded search assisted by earlier non-branded clicks
• How cross-device paths (phone research, desktop purchase) get stitched together, and where that stitching fails
• How your attribution model choice directly changes Smart Bidding behavior and budget allocation
• Why attribution is getting harder in a cookieless, privacy-first world, and what "modeled conversions" actually means
• Why Google Ads, GA4, and your CRM will never show identical conversion numbers — and why that is normal, not a bug
• A practical framework for choosing the right model for your business type, sales cycle, and data volume
• Common attribution mistakes that lead advertisers to cut the wrong keywords and campaigns

Why Attribution Models Change How "Performance" Looks

Before touching any specific model, it helps to internalize one fact: a conversion is rarely caused by a single click. Google's own path-length data across advertiser accounts shows that a meaningful share of conversions — often 20% to 50% depending on vertical and average order value — involve two or more ad interactions before the purchase, and that is before you even count organic search, direct visits, email, and social touches in the same journey. When a customer searches "best project management software," clicks a non-branded ad, reads a comparison page, leaves, comes back three days later by searching your brand name, and converts, there were two SEM touchpoints in that single journey. The question an attribution model answers is: how much of the credit — and therefore how much of the reported ROI — belongs to each of those two clicks?

This matters enormously in practice because budget decisions are made from reports. If your reporting model assigns 100% of the credit to the branded click, the non-branded keyword that started the journey will show a low or negative ROAS quarter after quarter. A team optimizing purely on last-click data will eventually pause that non-branded keyword to "cut waste" — and then watch branded search volume and conversions quietly decline over the following months, because the keyword that was actually creating the demand is gone. The model you choose is not just a reporting preference; it is a lens that determines which parts of your account look like they are working and which look like they are failing, and those judgments directly steer where money goes next.

Last-Click Attribution: Simplicity and Its Blind Spots

Last-click (technically "last non-direct click" in most implementations) attribution assigns 100% of conversion credit to the final touchpoint before the conversion, ignoring everything that happened earlier in the journey. It was the default model in Google Ads and Google Analytics for well over a decade, and it remains the easiest model to explain to a client or a boss: "this ad caused this sale" is a simple, satisfying sentence.

Its strengths are real. Last-click is easy to audit, easy to reconcile with a single transaction record, resistant to modeling errors (there is no algorithm guessing at partial credit — the rule is deterministic), and perfectly adequate for short, simple purchase paths where most customers see one ad and buy immediately, such as impulse retail purchases or emergency local services ("locksmith near me").

Its blind spot is equally real: last-click systematically overvalues bottom-of-funnel, high-intent, often branded keywords and undervalues the awareness and consideration keywords that created the demand in the first place. A prospective SaaS buyer who clicks a non-branded ad for "CRM for small teams," downloads a comparison guide, and converts eleven days later after searching the brand name will have that entire conversion credited to the branded click. The non-branded click — arguably the harder and more valuable piece of marketing, since it introduced a new prospect to the brand — gets nothing. Run this pattern across an account for a year and last-click reporting will consistently recommend killing the very campaigns responsible for growth.

First-Click Attribution

First-click attribution is the mirror image: it assigns 100% of the credit to the very first touchpoint in the recorded journey, on the theory that the interaction that introduced the customer to your brand deserves the credit for eventually winning them. If a customer's first-ever interaction was a non-branded search ad, and everything afterward (branded searches, email clicks, direct visits) was just them "closing the loop" on a decision they'd already started making, first-click attribution tells that story.

First-click is genuinely useful for one specific job: evaluating top-of-funnel and awareness campaigns whose entire purpose is to generate new-to-brand demand rather than to close sales directly. If you are running a prospecting campaign and want to know which keywords, audiences, or ad creatives are best at pulling in customers who go on to convert eventually, first-click data (or a "new users" / "new-to-brand" report in GA4) tells you that far better than last-click does, because last-click will attribute all of that same prospecting campaign's downstream conversions to whatever closed the deal later.

The weakness of first-click is the same weakness as last-click, just flipped: it ignores every touchpoint in between and overcredits the earliest click even if that click had nothing to do with the final decision. A customer who clicked a display ad eight months ago, forgot about your brand entirely, and later converted after a completely unrelated branded search triggered by a friend's recommendation will still have that display ad credited with the win. First-click and last-click are useful as two ends of a spectrum for sanity-checking a keyword's real value — if a keyword looks great in first-click and terrible in last-click, that is a strong signal it is a top-of-funnel driver, not a closer, and should be judged and budgeted accordingly rather than paused.

Linear and Time-Decay Models

Between the two extremes sit two "rule-based" compromise models that split credit across every touchpoint in the journey instead of handing it all to one click.

Linear attribution divides credit evenly across every touchpoint. A four-click journey (non-branded search, retargeting display ad, email click, branded search) gives each of those four interactions 25% of the conversion credit. Linear is easy to explain and guarantees that assisting clicks are never reported as worthless, but it has its own distortion: it treats a passing display impression exactly the same as a high-intent search click, which usually overstates the value of low-effort, low-cost touchpoints like display and remarketing relative to search.

Time-decay attribution also spreads credit across all touchpoints, but weights the touchpoints closer to conversion more heavily, using an exponential decay curve (Google Ads historically used a 7-day half-life: a click 7 days before conversion got half the credit of a click on the day of conversion). This is a genuine improvement for most SEM funnels because it reflects a real behavioral truth — a click three months before purchase probably mattered less to the final decision than a click three days before purchase — while still acknowledging that earlier clicks contributed something rather than nothing. Time-decay was, for years, considered the best available "rule-based" compromise for accounts that weren't yet eligible for data-driven attribution, though it is now largely superseded.

Data-Driven Attribution: Google's Default Model

Data-driven attribution (DDA) is the model Google Ads and GA4 now use by default, and since 2023 Google has removed several of the older rule-based models (first-click, linear, time-decay, and position-based) from Google Ads conversion actions entirely for many accounts, leaving last-click and data-driven as the primary practical choices in that platform. Understanding how DDA actually works is essential, because it is no longer optional in the way it used to be — most advertisers are using it whether they consciously chose it or not.

DDA uses a machine-learning technique (Google has described it publicly as based on Shapley value game-theory concepts, adapted for advertising paths) that compares converting paths against non-converting paths across your account's own historical data. In plain terms: the model looks at thousands of customer journeys that ended in a conversion and thousands that did not, and it asks, statistically, "which touchpoints show up disproportionately more often in the paths that converted versus the paths that didn't?" A touchpoint that appears frequently in both converting and non-converting journeys in a similar ratio gets little credit, because its presence doesn't predict conversion. A touchpoint that appears far more often in converting journeys gets more credit, because its presence is a meaningful signal that the customer was more likely to buy.

This is a genuinely more honest way to measure contribution than any fixed rule, because it is derived from your actual account's behavior rather than an arbitrary formula applied uniformly to every advertiser. It also updates continuously as your account's conversion patterns change, rather than being locked to a static rule written years ago. The tradeoff is that DDA requires enough data to be statistically meaningful — Google's stated eligibility threshold has historically required a minimum volume of clicks and conversions (roughly 3,000 clicks and 300 conversions within a 30-day window at the conversion-action level, though thresholds and requirements evolve) — and for accounts below that threshold, DDA either isn't available or its output should be treated with appropriate caution, since ML models trained on thin data can behave unpredictably.

Multi-Touch Journeys in SEM

Example: A prospective customer searches "project management tools for agencies" and clicks a non-branded Search ad (Touch 1). She browses the pricing page, leaves without converting, and is served a remarketing display ad two days later, which she notices but does not click (a view-through impression, Touch 2 — not a click, but part of the journey). Five days after that, she searches "Acme Software" by name, clicks the branded Search ad (Touch 3), and signs up for a trial, which converts to a paid plan eleven days later (the actual conversion event). Under last-click, Touch 3 (the branded click) gets 100% of the credit. Under first-click, Touch 1 (the non-branded click) gets 100%. Under linear, credit is split between the two clicks (view-through impressions are typically reported separately). Under data-driven attribution, the model might assign roughly 65% credit to the non-branded click that introduced her to the category and brand, and 35% to the branded click that closed the loop — because DDA has learned, from thousands of similar paths in the account, that non-branded discovery clicks in this pattern are strongly associated with eventual conversion, even though the branded click was the technical last touch.

This kind of journey is completely ordinary in SEM, and it is precisely why "branded search" needs careful interpretation. Branded keywords very often show the best last-click ROAS in an account simply because they capture the final step of journeys that non-branded keywords started. That does not mean branded search is your best marketing — it may mean your non-branded campaigns are doing the expensive work of demand generation and your branded campaigns are collecting a easy layup. Cross-device paths add another wrinkle: the same customer may research on mobile during a commute, compare options on a work desktop, and finally convert on a home laptop. Google's cross-device reporting (built on signed-in Google account data) can stitch some of this together, but a meaningful share of cross-device journeys are invisible to any single platform, meaning even data-driven attribution is working with a necessarily incomplete map of the full customer journey, not the complete truth.

How Attribution Model Choice Affects Bidding Strategy Decisions

Attribution isn't just a reporting exercise — it feeds directly into Smart Bidding. Target CPA, Target ROAS, Maximize Conversions, and Maximize Conversion Value all optimize toward the conversion data reported under whatever attribution model is active on the relevant conversion action. If your conversion action uses last-click attribution, Smart Bidding will systematically bid up on keywords and audiences that tend to be the last touch — often branded terms and high-intent bottom-funnel queries — and bid down on keywords that tend to assist rather than close, because the algorithm literally cannot see the value those assisting clicks created.

Switching a conversion action from last-click to data-driven attribution can visibly shift bidding behavior within days to weeks: non-branded, top-of-funnel keywords that were being bid down under last-click often see increased bid pressure and impression share once DDA recognizes their real contribution to conversions, while some previously "top performing" branded exact-match terms may see bids ease slightly, since DDA correctly identifies that a portion of that credit belongs upstream. This is one of the most concrete, measurable ways attribution model choice changes actual account performance and spend allocation, not just how a report reads — it changes what the algorithm is being told to chase. For this reason, any advertiser evaluating "why did this campaign's CPA change" after a bidding strategy change should first check whether the underlying conversion action's attribution model also changed at the same time; the two are frequently intertwined and get conflated.

Attribution in a Privacy-Restricted, Cookieless World

Attribution has become measurably harder over the past several years, and the cause is structural, not a Google product decision alone: Safari's Intelligent Tracking Prevention, Firefox's Enhanced Tracking Protection, iOS App Tracking Transparency, and the broader decline of third-party cookies all reduce the ability of any single platform to observe a customer across sessions, devices, and browsers. A journey that would have been fully visible in 2017 — non-branded click, retargeting impression, branded click, purchase, all tied to one persistent identifier — may today have gaps where the platform simply loses the thread between touchpoints.

Google's response, alongside the broader industry's, has been "modeled conversions": when a portion of conversions cannot be observed directly (for example, because consent was declined, or a cookie expired, or cross-device stitching failed), Google's systems use statistical modeling — trained on the patterns of conversions it can observe, applied to the population of users where signals are missing — to estimate how many additional conversions likely occurred and attribute them proportionally back to the campaigns and keywords that most plausibly drove them. Consent Mode (in its basic and advanced forms) is the mechanism that keeps this modeling honest: when a user declines analytics or ad-personalization cookies, Consent Mode still allows anonymous, aggregated signals to reach Google, which feeds the conversion modeling without violating that user's choice. Advertisers who have not implemented Consent Mode correctly in the EU/UK will typically see reported conversions understate reality, sometimes significantly, purely because the modeling has too little signal to work with — this is one of the most common causes of an advertiser mistakenly believing performance has collapsed when in fact only visibility has.

Reconciling Ad Platform Numbers vs GA4 Numbers

Nearly every advertiser eventually asks: "Google Ads says 340 conversions this month, but GA4 says 290, and our CRM says 260 closed deals — which one is right?" The honest answer is that all three can be correct simultaneously, because they are measuring different things with different rules, and expecting them to match exactly is the actual mistake, not the discrepancy itself.

Google Ads counts a conversion the moment its own attribution model and tracking (often a data-driven model, sometimes with a broader attribution/lookback window) determines an ad-influenced conversion occurred, including some conversions modeled from incomplete signal. GA4 uses its own attribution model (also data-driven by default, but computed independently from Ads, and inclusive of all channels, not just paid search), its own session and user-identity logic, and its own conversion counting rules, which can differ from Google Ads even for the exact same website event. Your CRM, meanwhile, is typically tracking a business-defined event — a closed-won deal, a paid invoice, a completed onboarding — which happens further down the funnel than a form-fill or an "add to cart," and which may be recorded well after the ad click that originally drove it, sometimes outside any lookback window the ad platforms use at all.

Layer on top of this: cross-domain tracking gaps, ad blockers, users declining cookie consent, bot traffic filtering differences, and the simple fact that Google Ads and GA4 sometimes attribute the "same" conversion to different channels because their models weigh the same touchpoints differently. A 10-20% variance between Google Ads and GA4 conversion counts is common and not, by itself, a sign of broken tracking. What should trigger investigation is a sudden, unexplained shift in that variance (for example, GA4 tracking dropped 40% overnight while Ads stayed flat), or a directional mismatch (Ads shows conversions rising while the CRM shows revenue falling), because those patterns suggest a tracking break rather than normal model disagreement. The practical habit to build is to pick one system as your primary decision-making source of truth for a given decision (usually the CRM for revenue and deal quality, Google Ads for bid and budget optimization, GA4 for cross-channel behavior analysis) rather than trying to force three systems into agreement.

Choosing the Right Model for Your Business Type

There is no universally correct attribution model — the right choice depends on sales cycle length, average order value, data volume, and how many channels typically appear in a customer's path to purchase.

High-volume e-commerce with short, simple paths (most customers see one ad and buy same-day): last-click or data-driven attribution both work reasonably well here, since journeys are short enough that the difference between models is often small. If conversion volume comfortably clears Google's DDA eligibility thresholds, data-driven is still preferable for its accuracy on the multi-touch minority of journeys, but the practical stakes of the choice are lower than in longer-cycle businesses.

Considered-purchase e-commerce and DTC with higher price points (furniture, appliances, luxury goods): expect meaningfully longer paths with retargeting and branded search playing a real assisting/closing role. Data-driven attribution is strongly preferable here, since last-click will understate the demand-generation campaigns and misallocate budget toward remarketing and branded search that are simply harvesting demand created earlier.

B2B and long sales-cycle businesses (SaaS, enterprise services, anything with a multi-week or multi-month consideration phase and often a human sales process after the initial lead): this is where attribution model choice matters most and where last-click is most dangerous. A lead form-fill is rarely the true conversion event; the real business outcome (a closed deal) may occur months after the ad click and be recorded in a CRM, not GA4 or Google Ads. These businesses benefit most from offline conversion imports feeding CRM deal data back into Google Ads (covered in Lecture 13), combined with data-driven attribution, so that the model can credit early-funnel keywords for the deals they actually influenced rather than only crediting whatever touchpoint happened to be logged last before the lead form was submitted.

Local service businesses (plumbers, dentists, local law firms) with typically short, single-touch paths and phone-call conversions: last-click is usually adequate here because the journey is genuinely simple, though call tracking accuracy matters more than attribution model sophistication in this segment.

As a general operating principle: default to data-driven attribution wherever it is available and your data volume supports it, since it is derived from your own account's actual behavior rather than an arbitrary fixed rule. Use first-click or top-of-funnel reporting as a supplementary lens specifically when evaluating prospecting and awareness campaigns, since even DDA can undervalue a channel whose entire job is introducing new customers rather than closing them. And whatever model you use, revisit the choice whenever your sales cycle, product mix, or channel mix changes meaningfully, since a model that fit your business two years ago may be quietly misleading you today.

Related Lessons Across SEO, AEO, GEO, SEM, and PPC

Use these connected lessons to move through organic search, answer engines, generative AI visibility, paid search, and PPC without losing the bigger strategy.

• Lecture 1: What Is SEM? How Search Engine Marketing Works in 2026 (SEM) - return to the course foundation when you need the big picture.
• Lecture 1: What Is PPC? How Pay-Per-Click Advertising Works (PPC) - separate SEM strategy from PPC execution.
• Lecture 38: PPC Strategy Roadmap: Bringing Search, Social, Retail, and AI Together (PPC) - connect SEM with the full PPC channel roadmap.
• Lecture 4: Keyword Research Fundamentals (SEO) - connect organic keyword research with the same demand signals.
• Lecture - 7: Prompt Research: Finding What People Ask AI Tools About Your Topic (GEO) - turn search queries into AI prompt research.