Attribution Models for B2B SaaS

• Use first touch to see where a recorded path began; HubSpot gives all credit to that first interaction under its First Touch rule✓ 3.
• Use last touch to inspect the final recorded interaction before a defined conversion; HubSpot assigns it all credit under Last Touch✓ 3.
• Compare models on the same event and cohort. GA4 has a Model comparison report for its paid and organic data-driven and last-click views✓ 1.

1. Define the event and unit before choosing a model

For one proposed B2B measurement design, maintain separate events for a form submission, qualified lead, accepted opportunity, closed deal and realized revenue. Each has a different owner and maturity period. A model that explains a website form submission should not quietly become a claim about deal revenue. Google describes attribution in terms of key events on user paths✓, while HubSpot separately identifies Deal Revenue Attribution for deals✓ 13. The event and object must travel with every chart.

Choose the reporting unit as well. A website visitor, CRM contact, buying account and opportunity are not interchangeable. Google's User acquisition report describes where new users came from; its Traffic acquisition report covers where new and returning users came from✓ 1. A B2B company may have several visitors and contacts from one buying account. If a team wants account-level revenue credit, it needs a documented way to join those identities to the opportunity. Report unmatched visitors and contacts instead of pretending that every website path becomes a known deal.

Set a frozen entry cohort and outcome cutoff. An account may research in one quarter, become a qualified opportunity in another, and buy later. Keep the source event date, conversion event date, campaign membership date and CRM stage history. A shorter attribution window can discard early interactions; a longer one may include touches with weak relevance. This window choice is an editorial reporting rule for the local analysis, not a universal software setting or a published benchmark.

2. Read the credit rules literally

HubSpot's First Touch assigns all credit to the first interaction, and Last Touch assigns all credit to the final interaction before conversion✓ 3. Those views answer different descriptive questions. First touch can help inspect how an account entered the measurable path. Last touch can help inspect what was recorded just before the chosen conversion. Neither rule proves that the selected touch alone produced the outcome. A buyer may have had sales conversations, word-of-mouth exposure or offline events that the chosen system did not record.

HubSpot's Linear model assigns equal credit across recorded interactions, while Time Decay gives more to interactions closer to conversion✓ 3. Linear is easy to explain but treats a brief visit and a substantial sales conversation alike if both are represented as one interaction. Time Decay explicitly favors recent events, which can be sensible for some operating questions but can understate early education in a long buying cycle. These are consequences of the documented credit rules, not measured evidence that either model increases pipeline.

HubSpot also documents an Empirical model that weights interaction types based on their frequency across conversion paths✓ 3. This is a platform-specific model. Its output depends on what interaction types the customer has captured and how those paths are represented. Do not call the result an independent causal estimate. Record which model was selected and retain the raw event history so a reviewer can understand why the channel totals moved.

3. Keep GA4 website credit separate from CRM opportunity credit

Google Analytics uses data-driven attribution by default and says its model distributes key-event credit using data specific to each key event✓ 1. The Model comparison report can show how its paid and organic data-driven and last-click approaches credit touchpoints differently✓ 1. For a website team, that comparison can reveal whether one reported channel total is unusually sensitive to the model. It does not mean GA4 sees every seller call, personal referral or later opportunity milestone.

Google also models online key events it cannot observe directly✓ 2. That matters when someone tries to reconcile a GA4 channel report with a CRM count: the two systems may differ in identity, event definition and observation. Google says attributed conversion data by channel may update for up to 12 days after the conversion was recorded✓ 2. Save an extraction date and allow a maturation interval before declaring that a recent week changed. A live dashboard number and a frozen board report can be different without either being an arithmetic error.

A practical comparison table should contain one row per event definition and model: eligible key events, measured period, model name, source system, extraction date, included channels and known missing paths. When joining to CRM deals, add the join key and unmatched count. If that join is unavailable, present website attribution and opportunity creation as separate reporting layers. Do not divide attributed website events by closed revenue and label the result channel ROI.

4. Understand Salesforce campaign influence before quoting revenue

Salesforce's Campaign Influence scans active campaigns for members who also hold a contact role on an open opportunity✓ 4. This is a specific relationship in CRM data, not a census of every way a buyer heard about the company. Missing campaign membership or missing contact roles can change the visible influence set. A team should audit those fields before treating a zero-influence report as proof that marketing had no role.

Salesforce says its Primary Campaign Source model assigns 100% influence to the campaign named in that opportunity field✓ 4. This is transparent and operationally simple, but it concentrates recorded credit by design. Salesforce also allows custom campaign influence models✓ 4. A custom allocation can reflect a documented business rule, but it cannot repair missing campaign members, incorrect opportunity links or inconsistent stage dates. Keep the primary source and custom view side by side during review so changes in allocation are visible.

Do not equate Salesforce campaign influence with Google key-event attribution or HubSpot interaction attribution merely because all use percentages. They have different underlying records, conversion objects and eligibility rules✓ 134. Reconcile their event dictionaries first. Use each system's output for the decision it can actually inform: website journey analysis, CRM campaign participation or opportunity reporting.

5. Use model disagreement as a diagnostic

For a local reporting exercise, calculate first-touch, last-touch and a multi-touch view over the same eligible opportunity cohort. Keep the raw opportunity and interaction counts fixed. When a channel's credited share changes across rules, investigate where it tends to appear in recorded paths and whether records are missing. The GA4 model-comparison feature and HubSpot's multiple credit rules show why this sensitivity view is possible✓ 13. Model disagreement is not by itself a defect. It shows that the credit question changes with the rule.

Make a budget decision only after adding cost, qualified outcomes and a separate assessment of incrementality. Descriptive credit can help a team find candidate channels to test or audit. It cannot establish that removing a channel would leave all other buyer behavior unchanged. If experimentation is not possible, state the uncertainty and keep attribution totals as associations under specified rules. Compare time periods only when event tracking, channel taxonomy, qualification and model settings remained stable, or annotate the change.

FAQ

How we researched this

We reviewed Google Analytics attribution and modeled-event documentation, HubSpot attribution-report documentation, and Salesforce Campaign Influence documentation on October 2, 2026. Fifteen ledger facts trace to exact excerpts in four dated page snapshots. Platform statements here describe only their own products. The proposed reporting and audit steps are editorial measurement guidance; no local model was run and no channel return or causal uplift is claimed.