Flight log / Transmission 010

Cross-channel attribution: what to trust and what to ignore

Add up the conversions every ad platform claims and you'll often find you sold more than you sold. Even Google gave up on 4 of its own attribution models in 2023, noting that fewer than 3% of Google Ads web conversions still used them.

Attribution is a credit-assignment problem wearing a truth costume. Every method sees part of the buyer's path, and every method has a reason to see it a certain way. Often that reason is that it also sold you the ads.

So skip the hunt for the one true model. Sort the methods by the question each can actually answer, then build the smallest stack your stage needs. No vendor shootout, just the trade-offs.

What is cross-channel attribution?

Cross-channel attribution is assigning credit for a conversion across every channel a buyer touched: paid search, paid social, organic, email, events, sales outreach. The goal is to compare channels on one scoreboard instead of letting each one keep its own.

B2B makes it harder. Several people on a buying committee touch your marketing for one deal, and your tracking sees them as strangers. Sales cycles outlast most click windows. And the moment that actually moved the deal, a peer's recommendation or a podcast mention, usually leaves no trace a pixel can find.

Why can't you trust platform-reported conversions?

You can trust them for one job: optimizing inside that platform. They're the signal the bidding algorithm uses, and they're usually the fastest data you have.

What you can't do is add them up. Each platform counts conversions it touched using its own windows and rules, and some count people who only saw an ad. A buyer who clicks a search ad, sees a LinkedIn ad and later clicks a Meta retargeting ad can show up as a conversion in all three. The platforms aren't lying, exactly. They're each answering "did we touch this?" when you need "what caused this?"

• Trust: relative performance between campaigns, ads and audiences within the same platform.
• Ignore: the platform's total as a measure of its share of revenue, and any comparison of one platform's ROAS against another's.

What is data-driven attribution, and which models does Google still offer?

Data-driven attribution uses your account's conversion paths to estimate how much each touchpoint contributed, instead of applying a fixed rule like "last click gets everything." In Google Ads it's the default for most conversion actions.

Google Ads now offers three models: data-driven, last click and external attribution, which lets you import credit calculated elsewhere. Google stopped allowing new conversion actions to use first click, linear, time decay and position-based models in June 2023, and switched remaining actions to data-driven in September 2023. At the time, Google said fewer than 3% of Google Ads web conversions used those four rules-based models.

• Trust: data-driven attribution for comparing keywords, campaigns and Google channels against each other. It's better than last click at crediting upper-funnel search and video.
• Ignore: the idea that it's cross-channel. Google's model sees Google's touchpoints. It can't weigh a LinkedIn ad or a webinar it never saw.

What is marketing mix modeling (MMM), and when is it worth it?

Marketing mix modeling works from the top down. Instead of tracking individual users, it uses statistics to relate weekly or daily spend by channel to total outcomes over time, while accounting for seasonality, pricing and other factors. Because it needs no cookies or click IDs, it isn't affected by tracking loss.

MMM used to be an expensive consulting project. Two open-source options changed that:

• Meridian, Google's open-source MMM, became available to everyone on January 29, 2025. It uses Bayesian methods, can take incrementality experiment results as priors and accounts for reach and frequency.
• Robyn, from Meta Marketing Science, is an open-source MMM package available in R, with a Python version Meta labels as beta.

Free software isn't a free model, the same way a free telescope isn't a free astronomer. MMM needs a long history of spend and outcomes, enough variation in spend to learn from, and someone who can tell a sensible model from a confident wrong one. For most early-stage B2B teams with one or two channels and a few hundred deals, it's too early.

• Trust: directional budget splits across channels, diminishing returns curves, effects tracking can't see.
• Ignore: weekly campaign decisions. MMM is a quarterly planning tool, not a daily dashboard.

Where does incrementality testing fit?

Incrementality testing is the closest thing to ground truth in marketing measurement. You withhold ads from a control group, of people or regions, and measure the difference. It answers what no attribution model can: would this conversion have happened anyway?

Tests are slower and costlier than reading a dashboard, so you run them on the channels where the answer changes a big decision. They also calibrate everything else: a test result can tell you how much to discount a platform's self-reported numbers, and tools like Meridian can use test results to anchor the model. The full playbook is in incrementality testing for marketers who hate math.

• Trust: the causal effect of a specific channel or tactic, during the test period.
• Ignore: the urge to treat one test as permanent truth. Retest as spend, creative and seasons change.

Should B2B teams use self-reported attribution?

Yes. Self-reported attribution means asking buyers how they heard about you, usually with an open text field like "How did you hear about us?" on the demo or signup form. It's cheap, and it surfaces sources no pixel can see: podcasts, communities, newsletters, peer recommendations, a LinkedIn post someone screenshotted into Slack.

The pattern that shows up in most B2B accounts: software attribution over-credits the last trackable click, often branded search or retargeting, while self-reported answers point further upstream to the moment the buyer first remembered the name.

• Make it open text. Dropdowns anchor answers and hide the surprises.
• Make it required on high-intent forms only, such as demo requests, where the extra field costs little.
• Read it monthly, categorize answers and compare trends with your tracked data.

• Trust: which channels create awareness and memory.
• Ignore: precision. People misremember, and "Google" often means "I searched your name after hearing about you somewhere else."

Which attribution method answers which question?

What attribution stack fits your company stage?

You don't need everything on day one. Here's a practical stack by stage. Treat the stages as rough shapes, not rules.

The step most teams skip is the second one: sending CRM stages, like qualified lead and opportunity, back to the ad platforms. When platforms optimize for form fills, they find people who fill out forms. Feeding them pipeline signals changes who they find. For how attribution fits the wider program, see the B2B PPC playbook and our take on B2B paid media strategy.

Where this breaks

• Tracking loss keeps growing. Consent rules, browser restrictions and ad blockers mean every user-level method sees less of the path each year. That's part of why MMM and experiments are back in fashion.
• Long B2B cycles outrun windows. A deal that closes 6 months after the first click falls outside most platform attribution windows. Your CRM has to hold that history.
• Models disagree, and that's normal. If data-driven attribution, self-reported answers and a lift test point different ways, it usually means each sees a different slice. Use the disagreement to pick your next test.
• Small numbers lie. With a few dozen deals a quarter, any attribution split will swing on a handful of deals. Look at trends over quarters, not weeks.

FAQ