B2B Marketing Analytics

Measurement that changes a decision

B2B marketing analytics is the collection and interpretation of behavioural, campaign and revenue data about business buyers, in order to change what the programme does next. That second clause is the discipline. A report nobody acts on is bookkeeping with charts.

The test most reporting fails is easy to apply and uncomfortable to run. For every recurring number in the monthly pack, name the decision it informs and the level at which that decision changes. If a metric could move by a third in either direction and nothing about next month's plan would differ, it is being collected out of habit.

Three things get called analytics and are not: tag installation is implementation, a dashboard is presentation, a warehouse is plumbing. All are necessary; none is the discipline. Separate too b2b website analytics, meaning the behaviour of visitors on pages you control, from b2b analytics in the wider sense, which takes in pipeline, product usage and b2b customer analytics after the contract is signed. Most requests for better analytics are requests to join the two.

What makes business measurement structurally harder

Consumer analytics assumes a person, a device, a session and a transaction, roughly in that order. Business buying breaks all four assumptions at once.

• The buyer is plural and multi-device. Gartner's May 2025 press release, from a survey of 632 B2B buyers fielded in August and September 2024, describes buying teams ranging from five to 16 people across as many as four functions. In analytics each is a separate user, several of them more than once: laptop at the office, phone on the train, home machine in the evening.
• The decisive events happen offline. A call from a mobile, a conversation at a trade stand, a specification handed to a distributor, a signature inside a procurement portal. The site sees the form; it never sees the deal.
• Revenue is recognised in a different quarter. By the time the invoice exists, the campaign is closed and the budget reallocated.
• The numbers are small. At forty enquiries a quarter, every rate you quote is an estimate with a wide margin.

GA4, stated precisely rather than repeated

GA4 is event-based: every interaction is an event. Sessions still exist, timing out after 30 minutes of inactivity by default and producing a session_start event with a session ID and number, but they are derived from events rather than being the primary unit, and counts commonly differ from Universal Analytics. The pre-GA4 history is gone: standard UA properties stopped processing hits on 1 July 2023, and from the week of 1 July 2024 the data was deleted along with interface and API access.

Four documented limits matter more in B2B than anything in the interface.

• Retention. Standard properties cap user-level and key-event retention at 14 months, and the default is two months, not 14. What retention governs is narrower than commonly claimed: Google documents that the setting does not affect standard aggregated reports and only affects explorations and funnel reports. Google signals data is capped at 26 months regardless.
• Consent mode. Four parameters, ad_storage, ad_user_data, ad_personalization and analytics_storage, the middle two added in v2. Basic blocks tags until consent; advanced loads them with consent denied by default and sends cookieless pings.
• Data thresholding. System-defined, not adjustable, and Google does not publish the number. On a low-traffic property, demographic and query rows are suppressed routinely.
• Behavioural modelling. It requires at least 1,000 events per day with analytics_storage set to denied for at least seven days, and at least 1,000 daily users with consent granted for at least seven of the previous 28 days. Most B2B sites reach neither figure, so they absorb the full loss from consent refusal with none of the modelled recovery large consumer sites receive.

The CRM join is the programme

Everything useful here comes from one join: the website event and the CRM record for the same account. Without it you hold a pageviews report and a closed-won report with no arithmetic between them, and every budget argument becomes a matter of taste.

GA4 has no native lead-quality or CRM-revenue dimension. That is an assessment of the product's documented feature set rather than a statement from Google, but the consequence is not in doubt: joining site behaviour to revenue needs the BigQuery export, offline conversion import, or a warehouse holding both. The mechanics are unglamorous. Stamp every form submission with a durable identifier and write it to the CRM record. Carry the advertising click identifier through the form and store it on the lead, so a deal closing in November traces to a March click. Send closed-won outcomes back to the ad platforms; enhanced conversions for leads supplements the click identifier rather than replacing it, and from 15 June 2026 those uploads run through the Data Manager API rather than the Google Ads API.

Attribution, described honestly

Every attribution model is a rule for splitting credit, and every rule is wrong in a direction you can describe in advance.

• First touch over-credits discovery and ignores whatever got the deal over the line. It makes brand and content look infallible.
• Last touch over-credits the final click, usually a branded search by somebody already convinced. It flatters paid search and makes everything upstream look free.
• Multi-touch models are defensible at volume. On forty deals a year they produce differently shaped noise; splitting credit across six touchpoints on a base of nine closed deals is arithmetic, not insight.

The browser environment caps what any model can see. Chrome is not deprecating third-party cookies and has publicly reversed that plan, so cookie-based cross-site measurement still works on Chrome traffic. Safari has blocked cross-site cookies by default since Safari 13.1 and iOS 13.4, and Firefox partitions them per site rather than blocking them.

Which is why self-reported attribution deserves more respect than it gets. One open field asking how the buyer first came across you is imprecise and inconsistently answered, and it names channels no model can see: a podcast, a former colleague, a conference panel, an answer from an AI assistant.

Where the analytics budget gets wasted

• Buying a platform before naming a decision. A tool is easier to approve than an analysis. Write down the three decisions you want the data to settle, then buy the cheapest thing that settles them.
• Judging content by pageviews. b2b content analytics done properly asks which pages appear in the histories of accounts that became opportunities. A comparison page read by nine people from four target accounts beats a listicle read by nine hundred strangers.
• Predictive scoring on thin data. b2b predictive marketing needs training examples. A model fitted to sixty closed deals learns the sales team's calling habits and one year's seasonality. In double-digit win counts, a documented set of rules beats a black box and can be explained in a meeting.
• Leaving GA4 retention at the default. Two months of user-level data ruins year-on-year exploration work, and the change is not retroactive.
• Treating web analytics as the revenue system of record. The CRM and the ledger settle revenue; analytics explains behaviour.

A reporting set that survives a long cycle

Report leading and lagging indicators together, and cohort by the date of the touch rather than the date of the close. A campaign that ran in Q1 should be judged on the pipeline it created in Q1 and the revenue that pipeline eventually produced, wherever that revenue falls. Use rolling twelve-month windows for anything with a small monthly count, and quote ranges rather than single rates when the denominator is under a few hundred.

Name the traps in the report itself. Consent refusal reduces observed traffic without reducing actual traffic. Thresholding removes rows rather than flagging them, so an empty demographics table is not evidence of an empty audience. Self-reported source data will conflict with platform data; show both and explain the difference rather than picking the flattering one.

One new line item deserves a place. GA4 now has an AI Assistants channel in its default channel group, assigned where the medium exactly matches ai-assistant, covering referrers such as ChatGPT, Gemini, Deepseek, Copilot and Grok. Traffic from assistants is a real, separately identifiable channel. Its share is another matter: every circulating figure comes from a vendor panel, the estimates contradict each other, and no primary dataset exists.

Frequently Asked Questions