Multi-Touch Attribution for B2B: Honest Limits, Real Uses

Your twelve-month attribution report cannot tell the board which marketing created the deal. It can show which recorded interactions accompanied it, but missing conversations, identity breaks and sales-entered CRM fields narrow that claim sharply. B2B multi-touch attribution in a twelve-month cycle answers a narrower question than the one your board is asking, and the honest report says so on its first page.

What B2B multi-touch attribution actually answers

The precise question is: “Among the interactions our systems recorded for known opportunities, where did those interactions occur and how was credit distributed?”

The question routinely substituted for it is: “Which marketing caused this revenue, and what revenue would disappear without each channel?”

Those are not equivalent.

MTA can expose useful patterns. It can show that decision-stage visitors repeatedly reach a security page, that known opportunities commonly return through branded search, or that campaign identity vanishes between Google Ads, GA4 and the CRM. Those findings can guide instrumentation, content and journey design.

They cannot establish incrementality. A touch may appear because demand already existed. Branded search can receive credit for harvesting interest created elsewhere. A sales email may prompt the recorded website visit while remaining absent from the report.

Whether marketing is creating future demand is a separate management question, covered in our guidance on demand generation across long sales cycles. MTA should not be stretched until it claims to answer both.

Attribution maps recorded association; it does not identify the cause of revenue.

The recorded journey is not the buying journey

Take an illustrative B2B purchase with 14 actual touches: six identified browser interactions, two colleague forwards, three internal meetings, one peer recommendation, one trade-event conversation and one unlogged sales call.

The model sees six touches versus the 14 that occurred.

14 actual touches − 6 visible touches = 8 unrecorded touches

That gap is not merely a tracking inconvenience. It changes the apparent importance and order of everything the system can see.

A forwarded PDF may become an “organic direct” return. The same buyer on a work laptop and personal phone may become two users. Procurement can join after months of evaluation without inheriting the original contact’s history. Retargeting through a long buying cycle can also create numerous recorded impressions without proving that those impressions changed the purchase.

A longer window does not recover invisible evidence. In the illustrative journey, a 90-day view might contain four recorded touches versus six in a 365-day view. The extra two remain only a larger subset of an incomplete journey.

The honest limit here is that our interpretation is confounded by cycle-length drift, identity loss, consent changes, campaign renaming and CRM data entered by humans under quota pressure. What this cannot tell you is whether the eight missing touches reinforced, displaced or had no effect on the recorded six.

Comparisons also break when the buying cycle changes. A 12-month cohort should not be judged against a nine-month cohort as if only channel performance moved. The reasons and website implications sit in our analysis of why B2B sales cycles lengthen and what websites can do about it.

Invisible touches make precise credit look more certain than the evidence allows.

Without a holdout, credit is only correlation

Credit allocations can look exact to two decimal places while remaining causally empty. Without a holdout, credit allocation is a description of correlation.

A minimal operational holdout has four controls:

1. Define the eligible accounts before the campaign begins.
2. Randomly assign an exposed group and an untreated group.
3. Keep sales coverage, qualification rules and other planned activity consistent.
4. Run through one full median touch-to-qualified-opportunity lag, then compare opportunity rates per eligible account.

A 90% exposed versus 10% held-out split is operationally simple, but it is not automatically large enough. Sample size must be set before results are inspected. If the planned sample is not reached, the result remains directional rather than budget-authorising.

The holdout should remove the intervention being tested, not all marketing. For retargeting, eligible accounts in the control group receive no retargeting while both groups retain the same search, email and sales treatment.

Our falsifiable claim for that test is that retargeting increases qualified-opportunity creation; an adequately powered result showing equal or lower exposed-account rates would prove it wrong.

No holdout or credible natural experiment means MTA may describe history only.

Causal budget claims require a comparison the campaign did not touch.

When B2B multi-touch attribution may move budget

Use four budget gates before treating attribution as an instruction rather than a description:

1. CRM completeness: At least 90% of won opportunities must contain value, creation date, stage dates and first-known source. Below 90%, fix capture and freeze attribution-led transfers.
2. Journey reconciliation: Audit the latest 20 qualified opportunities. If 16 or more reconcile across website, advertising and CRM chronology, proceed; if 15 or fewer reconcile, report history only.
3. Cycle stability: The current cohort’s median cycle must sit within 20% of the comparison cohort. Beyond 20%, rebase the periods before judging channels.
4. Causal evidence: The holdout or natural experiment must reach its predeclared sample. An unfinished test cannot authorise movement.

Only when all four gates pass should the first transfer be capped at 10% of the channel budget. Smaller moves preserve the comparison and limit the cost of a false positive.

Take an illustrative UK consultancy spending £8,000 per month on paid media. The five labelled inputs are:

The arithmetic is reproducible:

£8,000 × 12 = £96,000 annual media spend

£240,000 ÷ £96,000 = 2.50 attributed pipeline per £1 spent

£120,000 ÷ £96,000 = 1.25 attributed won revenue per £1 spent

£8,000 × 10% = £800 proposed monthly test transfer

The £240,000 attributed pipeline versus £120,000 attributed revenue describes the system’s allocation. It does not show what would have happened without the channel.

If any gate fails, the authorised transfer is £0 versus the proposed £800. If all four pass and the holdout favours the alternative, move £800 and retain the control.

That separation between reporting and experimentation is central to measurement-led growth marketing.

Budget follows tested incrementality, while attribution explains recorded history.

If your website and campaign evidence cannot explain where committee-led deals slow down, map the buying journey with Actualyse — book a call

What did not work: four fixes that preserve false certainty

The version reaching us in audits usually contains one of four attempted fixes.

Buying another attribution platform

Connecting Google Ads, GA4 and HubSpot can improve joins and expose broken tracking. It cannot reconstruct an undocumented boardroom discussion, a forwarded screenshot or an anonymous visit.

MTA software processes the available evidence; it does not resolve measurement uncertainty.

Expanding every window from 90 to 365 days

A 365-day window versus a 90-day window includes more old correlations. It does not reveal whether the longer history caused the deal, and campaign structure may have changed during the added period.

Longer visibility is useful only when the report also shows cohort and cycle drift.

Making sales complete more CRM fields

Mandatory fields often produce “unknown”, convenient defaults or retrospective guesses. Humans working under quota pressure optimise for progressing the opportunity, not preserving a clean research record.

Audit field accuracy against source evidence before treating completion as truth.

Debating increasingly clever touch weights

Reallocating credit among recorded touches cannot create a control group. The argument changes who receives credit without testing whether any touch changed the outcome.

A weighting debate is often polished avoidance of the causal question.

Better tooling cannot manufacture evidence the business never recorded.

Build the report around buying routes and decisions

Website measurement becomes more useful when routes reflect how different people evaluate the offer. Flattening every stakeholder, industry and intent into one “website visit” destroys that context.

Four examples from our work show the architecture problem. Lanteria required a broad HR software capability to serve multiple stakeholder audiences. AfriCap Hub organised an executive-events catalogue around discovery, filtering and registration. Savgen needed a technical, multi-industry offer structured coherently. Lake Erie Shores separated stay and ownership audiences under one site.

Those were architecture and design decisions, not attribution experiments. The project pages publish no performance figures, so they cannot prove that a route caused more revenue.

They do show why meaningful route labels matter. A finance visitor reading implementation evidence should not be treated as interchangeable with an operational user comparing features. Role-based website personalisation can change those journeys, while case-study storytelling for complex services can provide evidence for later-stage stakeholders. Attribution should record those distinctions without claiming they caused the deal.

The report’s first page needs six statements:

1. Measured question: The exact association the report calculates.
2. Population: Included opportunities, channels, dates and markets.
3. Coverage: Recorded versus expected sources, including known identity gaps.
4. Confounds: Cycle-length drift, consent changes and human-entered CRM data.
5. Causal status: Descriptive, directional or supported by a completed holdout.
6. Permitted decision: Diagnose tracking, change content or move a capped budget amount.

If “permitted decision” says “describe recorded history”, no channel should be cut from that page.

An honest first page protects the board from false precision.

FAQ

Should GA4 or the CRM be the source of truth?

Give GA4 authority over observed website behaviour and the CRM authority over opportunity stage and value. If more than 5% of closed-won records disagree after joining the systems, publish both totals and block channel-level ROI rankings until reconciled.

How should consent loss appear in the report?

Record consent state through Google Tag Manager and Consent Mode, using the GDPR/PECR basis agreed with counsel. Our reporting rule is that consented-session coverage below 80% triggers a “partial journey” label; cross-period comparisons are blocked if consent rates differ by more than five percentage points.

Can Performance Max receive MTA credit?

Include the campaign only at the granularity Google Ads can substantiate. If one Performance Max campaign spans more than two offers, restructure it before assigning offer-level credit; a blended campaign cannot support a precise offer decision.

When should qualified opportunities replace won deals as the optimisation outcome?

If the business records fewer than 20 wins per quarter, optimise against sales-accepted opportunities and retain won revenue as the quarterly validation measure. Do not lower the qualification standard to manufacture volume.

Operational definitions prevent attribution debates from becoming budget theatre.

Summary

Apply these five rules:

• Label MTA descriptive until a holdout or natural experiment reaches its predeclared sample.
• Below 90% CRM completeness, fix capture and block channel ROI claims.
• If 15 or fewer of 20 sampled journeys reconcile, freeze budget movement.
• Rebase comparisons when median cycle length moves by more than 20%.
• Cap the first budget transfer at 10% and keep the holdout intact.

Actualyse builds websites and campaigns designed for long, committee-driven B2B sales cycles. Book a call to talk through where yours stands.