Attribution Models Explained: Which One Fits Your Ads
. By Ads Daddy.
# Attribution Models Explained: Which One Fits Your Ads  An attribution model is a rulebook for handing out credit. A customer sees your Instagram ad, clicks a Google ad, opens an email, then buys. The model decides who gets the trophy. Pick the wrong rulebook and you'll cut the ads that started the sale and pour cash into the ones that merely finished it. Pick the right one and you'll know where the next dollar should go. Here's the plain-English version, with no theory for theory's sake. ## TL;DR - **Attribution models decide which touchpoint gets credit for a sale.** Single-touch models credit one step. Multi-touch models share the credit. - **Last click is the default in plenty of reports, and it's biased.** It flatters the final step and starves awareness ads. - **Google Ads now leans on data-driven attribution.** It has retired first click, linear, time decay and position-based for conversion actions. - **No model is the truth.** Treat it as a compass, then check it with incrementality tests. - **Fix your tracking before you pick a model.** Messy data makes every model wrong. ## Table of Contents - [What an attribution model actually does](#what-an-attribution-model-actually-does) - [The main attribution models, one by one](#the-main-attribution-models-one-by-one) - [Quick comparison](#quick-comparison) - [Single-touch vs multi-touch](#single-touch-vs-multi-touch) - [Which model should you use?](#which-model-should-you-use) - [The attribution window trap](#the-attribution-window-trap) - [MTA, MMM and why the ground is shifting](#mta-mmm-and-why-the-ground-is-shifting) - [Attribution is a compass, not a courtroom](#attribution-is-a-compass-not-a-courtroom) - [Fix your data before you pick a model](#fix-your-data-before-you-pick-a-model) - [Data-driven attribution: what it needs](#data-driven-attribution-what-it-needs) - [A worked example](#a-worked-example) - [Our perspective](#our-perspective) - [Want someone else to untangle it?](#want-someone-else-to-untangle-it) - [FAQ](#faq) ## What an attribution model actually does Think of a footy goal. The striker taps it in. But the midfielder won the ball and the winger crossed it. Last-click attribution gives the striker 100% of the glory. Everyone else wasted their afternoon, apparently. That's the whole idea. Customers rarely buy after one ad. They scroll past you on Facebook, search your brand name a week later, click a retargeting ad, then convert. Your reporting has to split that journey into credit somehow. The model is the split. Change the model and the same campaign can look like a hero or a dud. Same ads. Same sales. Different story. That's why this matters for your budget. The channel that looks best under one model can look mediocre under another, and you'll make spending calls off that picture. ## The main attribution models, one by one ### Last click All credit goes to the final touchpoint before the conversion. **Good for:** short buying cycles, simple setups, impulse purchases. **Watch out for:** it over-rewards branded search and retargeting. Those channels catch people who were already sold. Rely on it alone and you'll under-invest in the ads that created the interest in the first place. ### First click All credit goes to the first touchpoint. **Good for:** judging which channels bring in new people. **Watch out for:** it ignores everything that nudged the buyer over the line. It's the mirror image of last click, with the same blind spot. ### Linear Every touchpoint gets equal credit. Five touches, 20% each. **Good for:** a first look at long journeys when you don't want to play favourites. **Watch out for:** a throwaway banner view gets the same credit as the click that closed the deal. It's fair in a boring way. ### Time decay Touchpoints closer to the sale get more credit. Earlier ones get less. **Good for:** longer sales cycles where the final nudges really do matter, like a quote request or a high-ticket service. **Watch out for:** it still undervalues the opening move. ### Position-based (U-shaped) The first and last touches get the biggest slices, often 40% each, and the middle shares the remaining 20%. **Good for:** businesses that care most about how a lead arrived and what closed them. **Watch out for:** those percentages are a human guess, not a measurement. ### Data-driven This one uses machine learning. It compares the paths of people who converted with those who didn't, then assigns credit based on each touchpoint's estimated impact. Google Ads explains it in its attribution models help page, and notes that the first click, linear, time decay and position-based models are no longer supported, with affected conversion actions upgraded to data-driven.  **Good for:** accounts with enough conversion volume to feed the algorithm. **Watch out for:** it's a black box. You can't see the exact maths, and it only sees what the platform can track. ## Quick comparison | Model | Who gets the credit | Biggest bias | Best fit | |---|---|---|---| | Last click | Final touch | Favours bottom-funnel ads | Impulse buys, tiny budgets | | First click | First touch | Ignores closers | Prospecting analysis | | Linear | Everyone equally | Treats all touches as equal | Early exploration | | Time decay | Recent touches | Undervalues awareness | Long sales cycles | | Position-based | First and last | Arbitrary weights | Lead gen with clear stages | | Data-driven | Based on modelled impact | Black box, needs volume | Accounts with steady conversions | ## Single-touch vs multi-touch Single-touch models (first click, last click) give one touchpoint everything. They're easy to read and easy to be wrong with. Multi-touch models (linear, time decay, position-based, data-driven) spread credit across the journey. They're closer to how people actually buy. If you want to go deeper, we've written about [how to pick a multi-touch attribution model and the data it needs](https://adsdaddy.com/au/blog/what-is-multi-touch-attribution). A simple rule: if your customers buy in one sitting, single-touch is fine. If they research, compare, wander off and come back, you need multi-touch. ## Which model should you use? Don't start with the model. Start with the question you're trying to answer. 1. **"Where do new customers come from?"** Look at first-click views. 2. **"What closes the sale?"** Look at last-click views. 3. **"How should I split budget across channels?"** Use a multi-touch or data-driven view, then sanity-check it. 4. **"Would these sales have happened anyway?"** That's not an attribution question. It's an incrementality question. More on that below. Then match it to your business: - **E-commerce with cheap products and quick decisions:** last click plus a data-driven view is usually enough to start. - **Service businesses and lead gen with a longer chase:** position-based or data-driven, with lead quality tracked past the form fill. - **B2B on LinkedIn and Google:** multi-touch, because buyers take weeks and several people are involved. The best model isn't the fanciest. It's the one your team understands and will actually act on. A complex model nobody trusts is just an expensive screensaver. ## The attribution window trap A model decides how credit is shared. A window decides how far back to look. You'll see settings like "7-day click, 1-day view" in Meta reporting. That means a sale gets attributed to an ad if the person clicked it in the past seven days, or viewed it (without clicking) in the past day. Google Ads and Google Analytics have their own lookback settings too. Why it matters: - A short window undercounts channels that work slowly. - A long window can claim sales the ad had nothing to do with. - Two platforms with different windows will each claim the same sale. Add up their reports and you'll get more conversions than you actually had. That last one catches almost everyone. Facebook says it drove the sale. Google says it drove the sale. Your bank account says you made one sale. Everyone's reading from their own scoreboard. ## MTA, MMM and why the ground is shifting **Multi-touch attribution (MTA)** follows individual users across touchpoints. It's granular, but it relies on identifiers, and those keep disappearing as browsers and devices restrict tracking. The IAB has reported that 95% of advertising and data decision-makers expected continued signal loss as of 2024. That's why cross-device gaps and missing conversions are normal, not a sign you've broken something. We cover that problem in [fixing cross-device tracking](https://adsdaddy.com/au/blog/what-is-cross-device-tracking). **Marketing mix modelling (MMM)** works on aggregate data, such as weekly spend and weekly sales by channel. It doesn't need to track individuals, so it holds up better when signals vanish. The trade-off is less detail and more data needed over time. Most serious advertisers end up using both, plus experiments. MTA tells you what's moving day to day. MMM keeps the big picture honest. Experiments settle arguments.  ## Attribution is a compass, not a courtroom You will never reconcile every dollar between your ad platforms, your analytics and your bank. Cookie loss, cross-device journeys and walled gardens make sure of that. So stop trying to make it perfect. Use attribution to see direction: which channels are trending up, which are fading, where the relative winners are. Then do the one thing that separates grown-ups from spreadsheet tourists: test it. ### Incrementality testing Your dashboard says Meta drove 200 sales. Did it? Or did it just stand next to customers who were already buying? Run a holdout test. Hide your ads from a small random slice of your audience, keep showing them to the rest, then compare sales. If the holdout group buys almost as much, your ads were taking credit, not creating demand. It's the only reliable way to separate "driven by ads" from "attributed to ads". We break the method down in [our guide to ad incrementality](https://adsdaddy.com/au/blog/what-is-ad-incrementality). ## Fix your data before you pick a model Every model runs on your tracking. Rubbish in, rubbish out. Before you debate first click against data-driven, run through this list: 1. **Conversion tracking fires once, on the right action.** Not on page load. Not twice. 2. **UTM tagging is consistent across every channel.** "facebook", "Facebook" and "fb" show up as three sources. 3. **Your ad platforms and CRM talk to each other.** Otherwise you'll never see which clicks became paying customers. 4. **Lead quality is tracked, not just lead volume.** A cheap lead that never answers the phone isn't cheap. 5. **One person is responsible for the numbers.** Shared ownership means nobody's checking. If you need a refresher on the basics, here's our walkthrough of [click attribution in five steps](https://adsdaddy.com/au/blog/what-is-click-attribution), and a general primer on [what ad attribution is](https://adsdaddy.com/au/blog/what-is-ad-attribution-a-2026-guide-for-marketers). ## Data-driven attribution: what it needs Data-driven attribution isn't magic. It learns from volume. Google has pointed to a baseline of around 200 conversions and 2,000 ad interactions over 30 days for the best results, as set out on its attribution models help page. It can work with less, but with thin data the picture gets shakier. If you're a small account with a handful of sales a month, don't beat yourself up. Use a simple model, watch trends rather than single numbers, and let volume build. Automation helps, too, but only when it's fed properly. See our take on the [role of machine learning in ads](https://adsdaddy.com/au/blog/role-of-machine-learning-in-ads-2026-guide). ## A worked example Say you run a Shopify store selling $90 skincare sets. Over a month, 100 customers buy. Their journeys look like this: - 40 saw a Meta video ad, later searched your brand on Google, and bought. - 30 clicked a Google Shopping ad and bought the same day. - 30 clicked an email and bought. Under last click, branded Google search and email take the credit. Meta shows almost nothing. A tired marketer cuts Meta. Weeks later, branded searches drop, because nobody's being introduced to you any more. Sales slide. Oops. Under a multi-touch view, Meta gets real credit for starting the 40 journeys. A holdout test would confirm whether those people would have turned up anyway. That's the difference between reading a report and reading the business. ## Our perspective We manage Meta, Google, YouTube, LinkedIn and Microsoft campaigns, and the same mistake shows up in nearly every account audit: someone treated the platform's dashboard as gospel. Here's how we think about it: - **Pick one primary view and stick with it.** Changing models every month just changes the story, not the results. - **Judge lead gen on what happens after the form.** SMS-verified, pre-qualified leads tell you more than a cheap cost per lead ever will. - **Use platform numbers for direction and your own numbers for money.** Your CRM and bank balance are the final scoreboard. - **Test before you cut.** If a channel looks weak in last click, hold out a segment before you switch it off. Attribution should make decisions easier. If your reporting is making them harder, the setup is the problem, not you. > *— Adrian* ## Want someone else to untangle it? Most business owners didn't start a company to argue with ad dashboards. That's fine. We run the campaigns, set up the tracking, and make sure leads and sales get credited properly across platforms. See what we do across [Meta, Google, YouTube, LinkedIn and Microsoft Ads](https://adsdaddy.com/au), or look at our [lead generation service](https://adsdaddy.com/au/lead-generation) if leads are your goal. Ready to talk? [Book a call](https://adsdaddy.com/au) and bring your current numbers. We'll tell you straight what they're really saying. For more on turning reports into decisions, read our piece on [using ad analytics to boost ROI](https://adsdaddy.com/au/blog/use-ad-analytics-to-boost-your-roi-and-campaign-results) and the broader guide to [attribution modelling for ROI](https://adsdaddy.com/au/blog/attribution-modelling-marketing-roi). ## FAQ ### What are marketing attribution models? They're sets of rules that assign credit for a sale or lead to the touchpoints a customer interacted with. Those touchpoints might be ads, emails, organic search or social posts. The rules can credit one step or share credit across the journey. ### When should I use MTA versus MMM? Use multi-touch attribution when you need user-level detail on digital journeys and your tracking is solid. Use marketing mix modelling when tracking is patchy, you spend across offline and online, or you want a big-picture view of channel effects. Many businesses use both and check the results with incrementality tests. ### What does "7-day click, 1-day view" attribution mean? It's a Meta-style attribution window. A conversion is credited to an ad if the person clicked it within the previous seven days, or saw it without clicking within the previous day. Shorter windows count fewer sales, longer ones count more, so compare results using the same window. ### Which attribution model is best for a small business? Start with whatever your platform reports reliably, usually last click or data-driven, and add a second view to cross-check it. Small accounts often lack the volume for data-driven models to work well. Track real revenue in your own system and test before cutting any channel. ### Why do Facebook and Google both claim the same sale? Each platform counts conversions using its own rules and windows, and neither sees the other's ads. If one customer touched both, both can claim the sale. Compare against your own sales data to see the true total. ## Recommended - [Attribution Modelling: Boosting ROI for Marketers](https://adsdaddy.com/au/blog/attribution-modelling-marketing-roi) - [What is ad attribution? a 2026 guide for marketers](https://adsdaddy.com/au/blog/what-is-ad-attribution-a-2026-guide-for-marketers) - [Pick a Multi Touch Attribution Model and the Data It Needs](https://adsdaddy.com/au/blog/what-is-multi-touch-attribution) - [What is ad incrementality: the marketer's real guide](https://adsdaddy.com/au/blog/what-is-ad-incrementality)