How do you build multi-touch attribution for 18-month B2B enterprise sales cycles?

What multi-touch attribution actually means at enterprise cycle length

Most attribution tooling was designed for a 30-day e-commerce window, where a person clicks an ad, browses, and buys within a session or two. An 18-month B2B enterprise cycle breaks nearly every assumption baked into that design. Cookies expire. Prospects change employers. The person who downloaded your first whitepaper is frequently not the person who signs — and may have left the company entirely by the time procurement gets involved. Six to twelve humans touch the deal, each arriving through a different door, and the buying committee's internal Slack conversations, where the actual decision gets made, are permanently invisible to your instrumentation.

So the first thing to fix is the unit of measurement. In short-cycle attribution the unit is the person. In enterprise attribution the unit is the account, and every touch — regardless of who at that account performed it — rolls up to the account's opportunity timeline. This single reframe resolves most of the arguments that otherwise consume months. When the VP of Engineering attends a webinar in month three and the CFO signs in month sixteen, those are not two disconnected journeys. They are two touches on one account-level opportunity, and both deserve credit.

The second thing to accept is that attribution in this context is not an accounting exercise. Nobody is going to produce a defensible ledger of exactly which dollar of pipeline came from which webinar across an 18-month span. What attribution *can* do is answer directional questions that materially change budget: which channels reliably appear in the first 90 days of deals that eventually close, which content types show up disproportionately in the mid-cycle stall period, whether your field events actually accelerate deals or merely appear near ones that were already moving. Those are the questions worth building for. Treat the output as a directional signal with error bars, and the political temperature drops immediately.

The third reality is data decay. Over 18 months a single contact can change roles, email addresses, or companies two or three times. Their original first touch is tied to an email that no longer resolves. Marketing automation platforms commonly strip or rotate UTM parameters after a fixed retention window, and a well-meaning rep merging two lead records can silently delete fourteen months of campaign membership history. None of this is exotic — it is the normal operating condition of a long-cycle CRM, and any attribution build that does not plan for it will produce confident-looking charts assembled from wreckage.

Why does this matter to RevOps specifically? Because attribution is the mechanism by which marketing spend gets defended or cut in the annual planning cycle. If the model is naive, it systematically undercredits the slow-burn channels — analyst relations, community, long-form content, executive events — that do the heavy lifting in enterprise buying and overcredits the last-mile channels like branded search and demo request forms that merely catch demand someone else created. Cut the top of the funnel because a last-touch report told you to, and you will feel it eighteen months later, precisely when nobody remembers the decision that caused it.

There is also a downstream effect worth naming early: whatever model you build will eventually leak into compensation and territory conversations. The moment a channel owner's budget depends on an attribution number, that number stops being a measurement and starts being a target. Plan for that from day one by keeping attribution outputs separate from the systems that pay people. Attribution informs budget allocation; it should not directly determine variable compensation. Teams that skip this separation spend the following year arbitrating credit disputes instead of improving pipeline.

The step-by-step build sequence

Work in this order. The sequence matters more than the tooling, and skipping ahead to the modeling step is the single most common way these projects die.

Step one — baseline against reality. Pull your last 20 to 30 closed-won enterprise deals. For each, sit with the rep and the account executive and reconstruct the real story: how did this account first hear about us, who championed it internally, what triggered the evaluation, what nearly killed it, what unblocked it. Write those down as narratives before you touch a single field. This is your ground truth. Every model you build afterward gets scored against these narratives. If the model says the paid search campaign drove a deal that the rep says came from a customer referral at a dinner, the model is wrong — not the rep.

Step two — fix identity resolution. Before any modeling, ensure a person can be tracked across email changes and company moves. Implement a persistent contact identifier that survives the email address changing. Configure your CRM's deduplication rules to *archive* rather than delete campaign member history on merge — this is a settings change in most platforms and it is the difference between keeping and losing your longest-cycle data. Then create a write-once custom field that captures original source at first touch and is never overwritten by later form fills. That field alone recovers a large share of the attribution signal that otherwise evaporates.

Step three — define what counts as a touch. Agree on a minimum engagement threshold and write it down. Email opens and ad impressions score zero — they are noise at this timescale. A meaningful touch requires active engagement: a reply, a form completion, meaningful time on a substantive page, attendance at a live event, a booked meeting. Getting sales and marketing to sign the same one-page definition here prevents roughly 80% of the credit fights that come later.

Step four — instrument the offline and the invisible. Field events, analyst briefings, dinners, and sales-led conversations are where enterprise deals actually turn, and none of them emit a click. Give every offline motion a unique tracking code and a consistent CRM activity naming convention. Accept that you will lose 10-20% of offline touches to human logging failure; that loss is tolerable as long as it is roughly uniform across channels, because uniform loss preserves relative comparisons even when it degrades absolute numbers.

Step five — model, then validate against step one. Only now do you assign weights. Start with a hybrid: first touch and last touch each take 20%, the middle 60% distributed with time decay so touches closer to close carry more weight. Run it against your 20-30 narrative deals. Where the credit distribution contradicts what the rep told you, investigate — sometimes the model has found something real that the rep did not perceive, and sometimes it has found a data artifact. Adjust, re-run, and stop when the two roughly agree.

Step six — publish narrowly, then widen. Ship the model to one segment or one pod for a full quarter. One saved report, one URL, same view every week. Only after two clean review cycles does it go company-wide.

Costs, timelines, and what to expect at each stage

Set expectations honestly with leadership before you start, because the timeline here is genuinely long and an unprepared executive will pull the plug at month four.

Time to first usable output: 8 to 12 weeks. That covers the baseline interviews, identity resolution cleanup, touch definition, and a first model validated against known deals. You will have something directionally useful at the end of this, but it will be built on partial history.

Time to trustworthy patterns: 3 to 6 months. This is when you have enough consistently-instrumented deals moving through stages that the mid-funnel patterns start to stabilize. Before this point, treat any channel-level conclusion as a hypothesis, not a finding.

Time to full-cycle validation: one complete cycle — 18 months or more. You cannot shortcut this. Until at least one cohort of deals has been instrumented correctly from first touch through close, you are extrapolating. Say this out loud in the kickoff meeting so nobody is surprised.

On cost, the largest line item is almost never software. It is the RevOps and marketing ops labor to clean identity data and enforce the touch definition, plus the recurring manager time to inspect the output weekly. A realistic staffing picture is one owner with genuine write access to CRM validation rules, part-time marketing ops support during the instrumentation phase, and a manager who will actually enforce the weekly inspection. Attribution projects run by someone without write access to the CRM fail — not slowly, but predictably, because every fix becomes a ticket in someone else's backlog.

On tooling: start with your CRM's native capabilities and custom fields. For most teams, campaign influence tracking, custom source fields, and a well-built report cover the majority of what is needed for the first two quarters. Dedicated attribution platforms add real value — particularly for identity resolution across anonymous web traffic and for account-level intent — but they add it *after* the manual workflow is proven. Buying the platform first is how teams end up with an expensive tool faithfully reporting on broken data.

Budget the ongoing maintenance too. Attribution is not a project that ends; it is a system that degrades. Every new campaign type, every acquired product line, every CRM migration introduces gaps. Expect a standing quarter-hour of weekly inspection and a half-day quarterly audit indefinitely. Teams that treat it as a build-and-forget deliverable find their model quietly wrong about a year later, usually discovered during the planning cycle when the numbers stop matching anyone's intuition.

One adjacent cost that surprises people: reporting the model's uncertainty is itself work. If you present a single number per channel with no error bars, someone will make a seven-figure decision on it. Producing honest ranges — this channel appears in 40-60% of closed-won journeys, with the wide band reflecting known logging gaps — takes extra effort and makes the output far more defensible under scrutiny.

Where teams get it wrong

Automating before the manual process works. The most common failure, by a wide margin. A team buys a platform, connects it to a CRM full of duplicate contacts and stripped UTMs, and gets a beautiful dashboard reporting nonsense with great confidence. Run the process manually on one segment for two weeks first. Document the before and after on a single report. Only then automate. Automating a broken manual process makes the brokenness faster and harder to see.

Treating first-touch or last-touch as good enough. In an 18-month cycle, the majority of genuine influence happens in the middle — the invisible touches. A champion who attended a webinar fourteen months ago and now argues internally on your behalf. A peer review on a software marketplace that surfaced during procurement diligence. An analyst report your buyer's boss read. First-touch credits the whitepaper nobody remembers; last-touch credits the demo request form that the champion filled in as a formality. Both produce budget decisions that quietly starve the middle.

Letting attribution become a political weapon. The moment credit determines budget, every team develops a strongly-held methodological opinion that happens to favor their channel. Head this off structurally: agree on the touch definition and a contribution window before running any numbers. A workable convention is that touches within a defined window of a direct sales activity attribute to sales, and touches outside it attribute to marketing — which kills both the "I sent one email fourteen months ago" claim and the "our rep closed it single-handedly" claim in the same rule.

Changing the model every time someone dislikes the output. Freeze the weights for at least a full quarter. A model that changes monthly cannot be compared across periods, which means it cannot support any decision that matters. If the output is uncomfortable, investigate the underlying deals before touching the weights.

Optional fields. Anything not enforced at save time does not get filled under quarter-end pressure. If a field is genuinely required for attribution, make it block the save, and give managers a documented exception path with a reason field so waivers are visible. Review waivers monthly — a recurring waiver pattern means the rule is wrong, not that the reps are.

Company-wide rollout before the pilot proves out. Widening the blast radius before fill rates hold is how you get an org-wide reputation for a broken system that then takes two years to rebuild trust in.

Ignoring the deals that did not close. Attribution built only on closed-won produces survivorship bias. The channels that appear in every won deal may also appear in every lost deal, in which case they are not differentiating anything. Run the same model against closed-lost and compare — the channels whose presence differs meaningfully between the two populations are the ones actually doing work.

Choosing your model: a decision framework

There is no universally correct attribution model. There is a model that fits your cycle length, deal count, and data maturity. Use this framework rather than copying whatever a vendor recommends.

If you close fewer than about 50 enterprise deals a year, statistical models are not available to you — the sample is too thin for any algorithmic approach to produce stable weights. Use a rules-based hybrid and lean heavily on the qualitative narrative from reps. Your ground truth is human, and that is fine.

If your cycle is under six months and mostly digital, standard time-decay works well and you can skip much of the identity-resolution work described above, because your data does not have time to decay.

If your cycle runs 12 months or longer with a multi-person buying committee — the case in the question — you need account-level rollup, persistent identity, and a hybrid position-based plus time-decay model. This is the most labor-intensive path and the only one that produces honest answers at this cycle length.

If you have thousands of deals a year and a mature data warehouse, algorithmic or data-driven attribution becomes viable. It requires substantial volume to converge and it produces a model nobody can fully explain, which is a real organizational cost — you trade interpretability for accuracy. Make that trade knowingly.

A useful adjacent tool at any volume: incrementality testing. Rather than asking which touches were present, hold a channel out from a matched segment and measure the difference in pipeline creation. This answers the causal question that attribution can only approximate. It is harder to run and slower to read at enterprise cycle length, but a single well-designed holdout can settle a debate that three quarters of attribution reporting could not.

Related questions

Can I retrofit attribution onto deals already in flight?

Partially. You can reconstruct recent touches from CRM activity history and email logs, but touches older than your marketing platform's retention window are usually unrecoverable. Retrofit what you can, mark those deals as partially-instrumented, and exclude them from channel-level conclusions.

Should attribution feed sales compensation?

No. Once attribution credit determines pay, the number becomes a target and the methodology becomes a negotiation. Keep attribution for budget allocation and channel strategy. Compensation should run on booked revenue and quota, which are unambiguous.

How do I handle multi-product or land-and-expand accounts?

Attribute the initial land and each expansion as separate opportunities on the same account, with a shared account-level touch history. Expansion deals typically show a much shorter effective cycle because the relationship touches predate them — do not average the two together.

What about anonymous traffic before a form fill?

Account-level identification tooling can resolve some anonymous sessions to companies, which is genuinely useful at enterprise scale. Treat it as a soft signal for pipeline timing rather than a hard touch, and never let it outweigh a confirmed human interaction.

Does this apply to partner-sourced pipeline?

Yes, with an added rule. Partner-sourced deals need an explicit sourcing flag set at creation and protected from later overwrite, or your model will credit whichever marketing touch happened to land after the partner introduction.

FAQ

How long before attribution data is reliable enough to act on?

Expect three to six months of consistent instrumentation before channel patterns stabilize, and one full 18-month cycle before you can validate end-to-end. Early signals from shorter segments and from stage-progression data can guide adjustments in the meantime, but no tool shortens the time required for actual deals to progress.

Do I need a dedicated attribution platform, or is my CRM enough?

Start with the CRM. Custom source fields, campaign influence tracking, enforced required fields, and one well-built saved report cover most of the need for the first two quarters. Dedicated platforms earn their cost mainly through identity resolution across anonymous traffic and account-level intent — value that only materializes once the manual workflow is already proven.

How do we capture offline touchpoints like field events and executive dinners?

Give every offline motion a unique tracking code and log it as a CRM activity under a consistent naming convention. For events, use a dedicated landing page or registration code. Assume 10-20% logging loss as normal; as long as that loss is roughly uniform across channels, relative comparisons between channels remain usable.

What weighting should we start with?

A hybrid: 20% to first touch, 20% to last touch, and the remaining 60% distributed across middle touches with time decay so later touches carry more weight. Validate against 10-15 deals whose real story you know from the reps, adjust where the model contradicts reality, then freeze for a quarter.

How do we stop attribution from turning into a sales-versus-marketing fight?

Agree on the definition of a meaningful touch and on a contribution window before anyone sees a number. Run a 30-day joint review with a handful of reps and a marketing ops lead, asking one question each week: does this output match what we know about these deals? Alignment on the definition is worth more than any technical refinement.

Should we model closed-lost deals too?

Yes. Attribution built only on wins suffers from survivorship bias — a channel present in every won deal that is equally present in every lost deal is not differentiating anything. Running the model against closed-lost and comparing distributions is one of the cheapest, highest-signal analyses available.

Sources

• https://hbr.org/2015/03/making-the-consensus-sale
• https://business.linkedin.com/marketing-solutions/b2b-institute
• https://www.gartner.com/en/sales/topics/b2b-buying-journey
• https://support.google.com/analytics/answer/10596866
• https://developers.google.com/analytics/devguides/collection/ga4/attribution
• https://www.salesforce.com/products/marketing-cloud/best-practices/marketing-attribution/
• https://knowledge.hubspot.com/reports/analyze-your-attribution-reports
• https://www.thinkwithgoogle.com/marketing-strategies/data-and-measurement/

Related on PULSE

• [What attribution model works for a multi-touch enterprise sales motion?](/knowledge/q112)
• [How do you build a multi-touch attribution dashboard after cookie deprecation in 2027?](/knowledge/q16193)
• [How do we map multi-touch attribution to our sales compensation plan without over-crediting pipeline touches?](/knowledge/q699)
• [How do you build a multi-year account expansion plan when buyers want to see 18-month value but you have 3-year thinking?](/knowledge/q628)
• [How does the 2027 consolidation of analytics tools impact the accuracy of multi-touch attribution in long-cycle deals?](/knowledge/q16317)