How Should a Small B2B Team Handle Marketing Attribution?

How should a small B2B team handle marketing attribution?

Pick one model, understand exactly what it ignores, and add a self-reported question on your forms to catch what it cannot see. For most small B2B teams that is last click plus asking people how they heard about you, and the second half matters more than the first.

The reason is structural rather than a criticism of the tools. Attribution models were built to allocate credit across trackable click paths. A B2B purchase involves a group of people, several months, conversations in private channels, and increasingly an AI answer that never sends a referrer. A large share of the influence happens where no model can look.

So the useful goal is not accuracy. It is a consistent, well understood signal you can compare over time, plus a separate mechanism for the parts it misses.

What models can you actually choose from now?

Fewer than most guides assume. Google's documentation lists three reporting attribution models for Google Analytics 4 properties: data-driven attribution, paid and organic last click, and Google paid channels last click. Google also notes that the first click, linear, time decay, and position-based models are no longer available as of November 2023.

That deprecation quietly ended a lot of common advice. If a blog post tells you to compare first click against last click to see which channels start versus finish deals, that comparison is not available in GA4 any more, regardless of how sensible it sounded.

Google defines paid and organic last click as ignoring direct traffic and attributing 100 percent of the key event value to the last channel the customer clicked through before converting. Its own example is a path of display, social, paid search, then organic search, which assigns 100 percent to organic search.

What does direct traffic do to your numbers?

It disappears, mostly. Google states that all attribution models exclude direct visits from receiving attribution credit, unless the path to the key event consists entirely of direct visits. That single rule reshapes how a B2B report looks, because direct is where a lot of B2B intent lives.

Think about how your buyers actually arrive. Somebody reads your post on a phone, mentions you in a private channel, a colleague types your domain in a week later, and a form gets filled. The visit that converts is direct, the influence was your content, and the model assigns credit to whichever earlier click it can find, if any.

This is not a bug in Google's logic. Excluding direct is reasonable, because direct is mostly an absence of information rather than a channel. But it means your report is a report about clickable paths, and you should read it that way rather than as a report about what caused revenue.

Is data-driven attribution worth using at low volume?

Only if you have enough conversion paths for a model to learn from. Google describes data-driven attribution as using machine learning to evaluate both converting and non-converting paths, taking a counterfactual approach that contrasts what happened with what could have occurred, and comparing users exposed to an ad against similar users in a holdback group.

That is a genuinely sophisticated method and it needs data to work. Google says each data-driven model is specific to each advertiser and each key event, which means a small B2B site with a handful of monthly conversions is asking a model to learn a pattern from very few examples.

There is also a reporting wrinkle worth knowing. Google notes that with data-driven attribution you may see decimals, or fractional credit, in metrics like key events and revenue, because credit is distributed across contributing interactions. Fractional numbers in a board report invite questions you may not want to spend the meeting on.

What should you set your lookback window to?

Longer than the default if your sales cycle is long, and deliberately rather than by accident. Google says the key event lookback window determines how far back a touchpoint is eligible for credit, that acquisition key events default to 30 days with a 7 day alternative, and that all other key events default to 90 days with 30 or 60 also available.

For B2B software, 90 days is often the honest minimum and sometimes still too short. If your average cycle is four months, a 90 day window structurally under-credits the content that started the relationship, and no amount of model choice fixes a window that ends before your buyers decide.

Google also notes the lookback window applies to all attribution models and all key event types, and that changes apply going forward. So this is a setting to get right early rather than one to tune repeatedly.

What happens when you change the model?

Your history changes with it, which is a detail people discover at the worst moment. Google states that changing the reporting attribution model applies to historical and future data. Your last quarter's channel mix can look different on Tuesday than it did on Monday, without anyone having changed a campaign.

There is a second scoping subtlety that causes arguments. Google says changing the reporting model is reflected in reports using event-scoped traffic dimensions such as source, medium, campaign and default channel group, while user and session scoped dimensions like session source or first user medium are unaffected.

So two reports in the same property can legitimately disagree, and the explanation is dimension scope rather than a data problem. Knowing that saves a lot of time when somebody says the numbers do not match.

What should a small team actually measure instead?

Self-reported attribution, plus a small number of stable directional metrics. A required open text or dropdown field asking how the person heard about you collects the influence that models structurally cannot see, and it costs one form field.

Treat those answers as qualitative signal rather than data to average. If eleven people this quarter mention a specific post or podcast, that is a stronger instruction than a percentage in a dashboard, and it is immune to the direct traffic problem entirely.

Then keep three boring numbers that do not depend on path reconstruction: qualified conversations per month, where the first meaningful touch was according to the person themselves, and time from first known touch to closed deal. Those survive tool changes and model deprecations, which is more than most dashboards manage.

How do you track campaigns without over-engineering it?

Consistent tagging on the things you control, and nothing more elaborate than that. UTM parameters on your own outbound links are the only part of the picture you can make reliable, so make them reliable and accept the rest is estimated. I covered the mechanics in setting up UTM tracking for Webflow campaigns.

The failure mode here is a naming convention nobody follows. Three variants of the same campaign name split one channel into three rows and produce a report that understates everything. A short written convention and one person who owns it beats any tooling decision at small scale.

Where extra plumbing genuinely pays is getting conversion data back into a place you can read alongside behaviour, which I went through in wiring Webflow Analyze conversions into Looker Studio. Reporting infrastructure is worth building once. Attribution precision mostly is not.

What about traffic from AI answer engines?

Assume it is under-counted and design around that. Referral data from AI surfaces is inconsistent, and a reader who gets your argument summarised in an answer and types your name later arrives as direct, which every model excludes from credit by the rule Google states plainly.

The practical consequence is that your best performing content can look flat in a channel report while clearly working in conversations. This is exactly the case the self-reported question catches, because people will name the article even when the analytics cannot.

It also argues against putting your most persuasive material behind a form, since gated content cannot be read by the systems doing the summarising. I made that case in why gating content hides it from AI engines.

What should you do next?

Check two settings today: which reporting attribution model your property is using, and what your key event lookback window is set to. If the window is shorter than your sales cycle, you have been reading a report that was never able to see the beginning of your deals.

Then add the how did you hear about us field to your main form this week. It is the highest value attribution work available to a small B2B team, it takes ten minutes, and it collects the exact information no model in the list can produce. If you want a second opinion on what your current setup is actually measuring, reach out and I will take a look.