How Should You Measure Marketing in B2B SaaS?

Why Does B2B Attribution Never Match Reality?

Because it can only see the end of the story. Attribution software measures identified sessions on your website. The B2B buying decision is made earlier, by more people, most of whom never identify themselves. The model is not broken. It is looking at the wrong part of the journey.

We build the tracking that feeds these reports, so we get the awkward conversation twice: once when a client asks why the numbers disagree, and again when a board asks which channel to defund based on them.

Here is what the software genuinely sees, what the buyer research says it misses, and the measurement framework we would build instead.

What Can Attribution Software Actually See?

The last stretch, and only the identified part of it. The 6sense B2B Buyer Experience Report for 2025 found the point of first contact with a seller sits at 61 percent of the way through the buying journey. Everything before that is happening without you in the room.

The scale of the invisible part is the important bit. 6sense reports buying groups averaging more than ten members on deals averaging 250,000 dollars, with buyers evaluating 5.1 vendors on average. Your form captured one person from that group, at the end.

The report is a serious sample, not a vendor anecdote. It surveyed nearly 4,000 respondents, with 49 percent at VP level or above and roughly half acting as ultimate decision makers, on purchases with median costs between 200,000 and 400,000 dollars. Those are the deals your attribution model is trying to explain with a single UTM parameter.

Which Models Does GA4 Still Offer?

Three, and fewer than most teams remember. Google's documentation lists data driven attribution, paid and organic last click, and Google paid channels last click. It also states that the first click, linear, time decay and position based models are no longer available as of November 2023.

That removal matters for anyone still describing a first touch strategy in a marketing plan. The model that plan depends on has not existed in Google Analytics for years, and reports built on the assumption it does are quietly using something else.

Data driven attribution is the interesting one. Google describes it as using machine learning across both converting and non converting paths, incorporating factors including time from the key event, device type, number of ad interactions, the order of ad exposure and the type of creative assets. It works counterfactually, comparing what happened against what might have.

What Is the Signal Loss Problem?

The inputs are getting thinner while the models get cleverer. Google announced in April 2025 that it would maintain its current approach to third party cookie choice in Chrome and would not roll out a new standalone prompt, citing an evolving global regulatory landscape among its reasons.

That is stability rather than reprieve. Google also pointed to further tracking protections in Incognito mode, including IP Protection. Any measurement approach that depends on recognising a person across sites and sessions is building on ground that keeps shifting.

The practical effect is that your attribution report is not just partial, it is inconsistently partial, and the gaps change over time. A channel that looks like it declined may simply have become harder to observe. We covered the collection side in our guide to server side tracking.

Should You Use First Touch or Last Touch?

Neither, as a budget decision. Both answer a question about the sequence of observed touches, and the 6sense findings say the decision was made before most of those touches existed. Optimising spend on either is optimising the visible tail.

Last touch has one honest use. It tells you which page or channel people arrive through when they are ready to act, which is genuinely useful for conversion work. Treat it as a report about your bottom of funnel plumbing, not about what created demand.

Data driven attribution is better than both within its own scope, and its scope is still identified digital touches. It cannot credit the conference talk, the peer recommendation or the podcast, and in B2B those are frequently the reason you were on the shortlist at all.

What Should You Measure Instead?

Four things, none of which is a channel report. Pipeline created against a target, by segment. Branded search volume over time, as a proxy for whether more of the market knows you exist. Win rate against named competitors. And self reported attribution collected at the point of contact.

Those four survive signal loss because none of them depends on tracking a person across sites. They are also the four a revenue leader actually cares about, which makes the conversation easier than defending a channel dashboard.

Add AI search visibility as a fifth if a meaningful share of your category researches that way, accepting that it is an impressions measure rather than a traffic one. We covered its limits in our piece on measuring AI search traffic.

How Do You Run a Holdout in B2B?

Geographically or by account list, never by user. B2B volumes are too low for a clean user level split, but you can hold out a region, an industry segment or a randomly selected half of your target account list, then compare pipeline creation over a full sales cycle.

The cycle length sets your patience. 6sense reports the average buying cycle at 10.1 months in 2025, compressed from 11.3 months in 2024. A holdout read at six weeks tells you nothing, because most of the effect has not happened yet.

This is the honest reason most B2B teams do not run holdouts, and it is worth saying out loud rather than pretending the test is quick. If you cannot commit to a full cycle, do not start, and rely on the directional measures instead.

What Is Self-Reported Attribution Good For?

Catching what the tracking cannot see. A single open field on your demo form asking how the person first heard about you will surface podcasts, conferences, communities and word of mouth that appear nowhere in analytics. It is the cheapest measurement upgrade available.

It is also unreliable in a specific, predictable way. People remember the last prompt rather than the first exposure, and they over report the obvious channels. Treat it as a discovery tool that tells you which channels exist, not as a weighting system.

Make it an open text field rather than a dropdown, at least at first. A dropdown can only return options you already thought of, which defeats the purpose. Read the answers monthly and let the recurring ones become your list later.

How Do You Present This to a CFO?

Separate what you can prove from what you can argue, on the same page. Proven: pipeline created, win rate, cost per opportunity in segments where the numbers are large enough to mean something. Argued: brand and demand work, justified by the market maths rather than by a channel report.

Name the measurement limits before you are asked. A marketing leader who explains that attribution sees the final 39 percent of the journey, and cites the buyer research for it, is more credible than one who presents a confident dashboard that later fails to predict anything.

Then propose the test rather than defending the belief. Offering a geographic holdout over a full sales cycle is a serious answer to a serious question, and it usually buys the patience the work needs.

What Would We Build First?

The self reported field and a branded search chart. Both take under a day, neither depends on cookies, and together they will tell you more about what is working than a quarter of dashboard tuning. Add them before you touch the attribution model.

Then fix your definitions so pipeline can be counted consistently by segment. Most attribution arguments we walk into are actually definition arguments in disguise, where two teams count an opportunity differently and blame the tracking.

Only after that is it worth investing in the tracking layer itself, and at that point the work is about resilience rather than precision. Our guide to third party cookies and your website covers what that resilience needs to account for.

If you want help building a measurement model your board will actually accept, we are happy to work through it with you. Find us at phoenix.studio.