Home / Glossary / Attribution
Metrics · the contested topic · deep dive

Attribution, honestly.

The most-litigated topic in B2B marketing and RevOps — and the area where the most theater happens. Most attribution models can't tell a real cause from a vanity correlation. The only disciplines that actually inform budget decisions are incrementality testing and signal-level attribution. This essay walks the 5 canonical models, why each fails, what the dark funnel is and why it's bigger than your dashboard shows, and how signal-anchored selling moves attribution from "which channel" to "which signal" — the only level of analysis that tells you where to actually spend.

Category: Metrics · Read time: 14 min · Updated: 2026-05-24 · ATTR-1.0
TL;DR
Attribution is the practice of assigning credit to the marketing and sales touches that led to a closed deal. The 5 standard models (first-touch, last-touch, linear, U-shaped, W-shaped) all fail in different ways because none of them have a control group — they all assume the touches caused the deal rather than testing whether they did. The only honest disciplines are incrementality testing (compare sent vs holdout groups) and signal-level attribution (tag the triggering signal, not just the channel). Channel-level attribution tells you "email drove 40 meetings"; signal-level attribution tells you "funding-round triggers drove 28 of those 40, exec moves drove 8, tech changes drove 4 — funding is where your budget should compound." Most attribution dashboards are political artifacts, not decision tools. The honest ones reshape how a sales org spends.

01What attribution actually means

The shallow definition: attribution is the process of assigning credit to the marketing and sales touches that led to a closed deal. Did the LinkedIn ad get this customer? The conference booth? The cold email? Attribution models try to answer.

The shallow definition is also where most attribution conversations stop — and where most of the dysfunction begins. The honest definition is harder:

Attribution is the practice of estimating which marketing and sales activities caused revenue outcomes, in service of deciding where to invest the next dollar.

The word that does the load-bearing work is caused. Most attribution doesn't measure causation; it measures correlation between touches and conversions. The difference matters because budget allocation should be based on what causes revenue, not what's present near revenue. The two are different and the gap can be enormous.

What attribution is supposed to inform

Attribution exists to answer practical questions:

  • Where should we increase or decrease spend? If LinkedIn ads are driving meetings, raise the budget. If they're not, cut it.
  • Which channels work for which segments? Maybe paid search converts enterprise but cold email converts mid-market. Different motions for different deal sizes.
  • What's our customer acquisition cost by source? CAC per channel guides bidding strategy, comp plan design, and pricing decisions.
  • How long is our actual sales cycle? Time from first touch to closed deal tells you forecasting horizons and pipeline math.

Every one of these decisions should hinge on causal evidence. Most are made on correlative noise. The gap between the two is the central problem of B2B attribution.

02The 5 canonical models

Five attribution models dominate B2B marketing dashboards. Each was designed to solve a different perceived flaw in the previous one. Each has its own failure mode.

First-touch
100% credit to the first touch
The first known interaction (ad click, landing page visit, content download) gets all the credit for the eventual deal. Favored by demand-gen teams because it credits their top-of-funnel work.
The first touch happens 6 months before close. Crediting it 100% ignores everything that happened between.
Last-touch
100% credit to the final touch
The last interaction before conversion (the meeting-booked email, the demo request form) gets all the credit. Favored by SDR teams because it credits their closing-the-loop activity.
The last touch is often a calendar link click on a deal that was already won. Crediting it 100% rewards SDR theater.
Linear
Equal credit to all touches
If 8 touches led to the deal, each gets 12.5% credit. Treats all touches as equally important. Favored by marketing teams that want to show breadth of contribution.
A spam ad impression and a one-on-one exec dinner are not equally important. Linear pretends they are.
U-shaped
40% first + 40% last + 20% middle
Heavy weight to first and last touch, light weight spread across the middle. Acknowledges that bookends matter more than the journey middle. Popular in HubSpot's default model.
The 40/40/20 weights are arbitrary — no causal evidence justifies them. It's a vibes-based heuristic disguised as math.
W-shaped
30% first + 30% MQL + 30% opp + 10% middle
Adds the lead-to-MQL transition as a third anchor. Used at B2B SaaS orgs with formal MQL/SQL stages. Marketo and Bizible default to a variant of this.
The MQL stage transition is often a CRM-process artifact, not a real causal event. Crediting it inflates the activity that triggers MQL marking, not what actually moves the deal.
Data-driven / algorithmic
ML model assigns credit per touch
A machine-learning model (Markov chain, Shapley values) trained on historical conversion data assigns credit dynamically per channel. Sold by enterprise tools as "the answer to attribution debate."
The ML model is trained on the same observational data as the simpler models. Without a control group, the algorithm is sophisticated guessing, not measurement.

The honest summary: none of these models measure causation. All of them divide a fixed pie of "credit" across the touches that were present near the deal. None test whether the deal would have closed anyway without those touches — which is the only question that matters for budget allocation.

03Why most attribution is theater

The dirty secret of B2B attribution: most of it is political theater dressed in dashboard graphics. Three patterns make this true.

1. The model gets chosen to favor the team that picked it

Marketing teams pick first-touch (credits top-of-funnel work). SDR teams pick last-touch (credits booking). Demand gen picks linear (credits everything they do). The math is downstream of the political question: which team gets credit for revenue? The "right" model is the one that puts the team-running-attribution on the winning side of the quarterly forecast call.

2. The model never gets validated against reality

An attribution model claims "LinkedIn drove $400K of pipeline last quarter." Reality test: would you have closed that pipeline without LinkedIn? Nobody asks. There's no holdout group. There's no incrementality test. The attribution claim becomes the truth simply because it appeared on a dashboard.

3. The model conflates "was present" with "caused"

The buyer searched for your category, found you via organic search, clicked a paid ad, attended your webinar, talked to an SDR, and finally bought. Attribution distributes credit across those touches. But the buyer was probably going to buy something in this category regardless — they were already in-market. The touches funneled their existing intent toward you; they didn't create the intent. Attribution can't distinguish between "we caused the intent" and "we caught the intent" — and the budget implications are very different.

The political signal: attribution wars
When two teams at the same company report conflicting attribution numbers for the same deals ("Marketing says they sourced it; Sales says they sourced it"), that's not a data quality problem — it's a comp plan problem. The attribution model rewards whichever team's claim wins the political fight. Fixing the model doesn't fix the underlying incentive. Until comp plans share credit explicitly (e.g., split-credit attribution that doesn't require a winner), the wars continue regardless of which model is "right."

04Incrementality — the only real test

If attribution can't measure causation, what can? Incrementality testing. The discipline borrowed from consumer marketing (where Facebook and Google have run incrementality tests at scale for over a decade) and applied — slowly, painfully — to B2B.

The structure of an incrementality test

Take your target audience for a campaign. Randomly hold out 5-10% of the audience as a control group. Run the campaign to everyone else. Measure conversion rate in both groups. The difference between the sent and the holdout — net of statistical noise — is the incremental conversion the campaign caused.

Incrementality math · the formula
Incremental conversions = (Sent group conversion rate − Holdout group conversion rate) × Sent group size
Example: Send a campaign to 5,000 accounts; hold out 500. Sent group books 80 meetings (1.6%). Holdout group books 8 meetings (1.6% × 10% of size). Incremental: 80 − 8 = 72 meetings caused by the campaign. Compare this number to your standard attribution-model claim of "campaign generated 80 meetings" — and notice the gap. The 8 holdout meetings would have happened anyway. Your campaign didn't generate 80; it generated 72.

Why most B2B teams don't run incrementality tests

Three reasons, in order of frequency:

  1. Sample size. Incrementality testing requires statistical power. B2B audiences are often small (1K-10K accounts), conversion rates are low (0.5-3%), and the noise band swamps the signal unless you run tests for months. Many teams give up before they have enough data.
  2. Political risk. An honest incrementality test might reveal that your most-loved channel is doing nothing. Nobody wants to be the one who killed the LinkedIn ads program by running the test. Teams choose not to know.
  3. Operational complexity. Running a clean holdout requires synchronizing sequencer config, sales rep behavior (the holdout accounts can't be worked manually), and reporting. Most marketing ops teams don't have the discipline to maintain it.

Despite the difficulty, incrementality testing is the only attribution discipline that actually measures causation. Every team serious about budget allocation runs at least quarterly tests on their largest channels. Every team operating on attribution-model-without-incrementality is guessing with confidence.

The pragmatic adoption path
Most B2B teams can't run full incrementality on every campaign. Pick the 2-3 channels that consume the most budget and run incrementality on those first. Even a single quarter of holdout data on your biggest budget item often reveals you're spending 30-50% more than the incrementality justifies. That insight pays for the testing infrastructure many times over.

05Channel-level vs signal-level attribution

The other dimension where standard attribution falls short: it measures the wrong unit. Most attribution platforms tell you which channel drove a meeting. The more useful question is which signal drove the right cold email at the right moment.

Channel-level attribution
Signal-level attribution
What it credits
The medium of the touch (email, ad, call, LinkedIn)
The triggering event that made the touch relevant (funding, exec move, tech change)
Decision it informs
Which channel to spend more on
Which signal types are worth detecting + acting on
Granularity
6-10 channels per dashboard
8 signal types × 6-10 signal source combinations
Update cadence
Static — channel-mix decisions made quarterly
Dynamic — signal investment shifts as market patterns change
Available in 2026 tooling
Every major platform (HubSpot, Bizible, Salesforce)
Mama, Common Room, a few in-house builds

The argument for signal-level

"Email drove 40 meetings last quarter" tells you something — but not enough to act on. Of those 40 meetings, how many came from cold emails anchored on funding signals, how many on exec moves, how many on tech changes? If 28 came from funding-anchored emails and only 4 from generic email blasts, the lesson isn't "email works" — it's "funding-signal-anchored email works; other email doesn't." The budget implication is different. The team training implication is different. The infrastructure investment (signal mining) implication is different.

Channel-level attribution lets a team conclude "we need to do more email." Signal-level attribution lets them conclude "we need to invest in earlier detection of funding signals because that's where the meetings come from." The second insight produces 5-10x ROI on the same budget; the first insight produces marginal improvement at best.

How signal-level attribution gets built

Three components needed: (a) the signal-mining layer that tags each cold outbound send with the triggering signal at time of contact, (b) the reply/meeting ingestion that captures outcomes (Reply Loop), (c) the dashboard that aggregates signal-type-by-signal-type outcomes. Most existing attribution platforms have (b) and (c) for channels but lack (a) for signals. The signal layer is the missing infrastructure.

06The dark funnel problem

Even with perfect attribution and incrementality testing, you can only measure what you can see. The "dark funnel" is the term for everything attribution platforms cannot capture — and in B2B, the dark funnel is bigger than the visible funnel.

Self-reported attribution
When a buyer fills in "how did you hear about us?" on the demo form, their answer almost never matches what your tracking captured. The buyer says "podcast"; your data says "paid search." Both can be true; only one shows up in your dashboard.
Peer recommendations
A trusted peer in the buyer's Slack community recommends your product. The buyer Googles you, lands on your site, fills the form. Attribution credits "organic search"; the actual cause was the peer recommendation that triggered the Google search.
Podcast + content consumption
The buyer listens to a podcast where your CEO appeared. Three weeks later they search for you and buy. The podcast appears nowhere in your tracking but caused the conversion. Hard to measure, real to revenue.
Repeat exposure compounding
The buyer saw your LinkedIn ad 47 times across 6 months. None of the individual impressions converted; the cumulative effect did. Attribution credits the final touch; the actual cause is the 47-impression buildup.
Email forwards
A champion forwards your cold email to the actual decision-maker. The forward lands in a different inbox, a different person clicks, but your tracking sees only one open. The downstream effect of the forward is invisible to standard email-tracking pixels.
Brand search after exposure
The buyer saw your booth at a conference, didn't engage, later Googles your brand name and lands on the homepage. Attribution credits "direct" or "branded search"; the actual cause was the conference booth that happened 4 weeks earlier.

How to think about the dark funnel

The dark funnel is real and large — likely 30-60% of B2B influence happens outside trackable channels. The dangerous response is to invest heavily in untrackable activities because "we can't measure them but they must be working." The honest response is two-pronged:

  1. Use self-reported attribution as a corrective. Adding "how did you hear about us?" to every demo form gives you a parallel attribution signal. When self-reported says "podcast" and your platform says "paid search," the gap is the dark funnel surfacing itself. Pay attention to the gap, not just to the platform claim.
  2. Run incrementality tests on dark-funnel-suspected activities. If you suspect a podcast tour is driving brand search but can't directly measure it, run an incrementality test: in two parallel cities, pause podcast investment for 60 days in one, maintain in the other. Compare brand-search volume + branded conversion. The delta is the podcast's incremental effect — measurable even though the touches are not.

Mature B2B marketing teams accept that ~30-50% of their attribution will always be guesswork because the dark funnel is real. They focus their measurement discipline on the half they CAN measure and use incrementality tests to estimate the half they can't.

07Building an honest attribution model

If you're starting an attribution function (or fixing a broken one), here's the practical playbook. Skip steps at your own risk.

  1. Pick one attribution model and stop debating. All five canonical models are wrong in different ways. Pick the one that least mis-aligns your incentives (most B2B SaaS settles on W-shaped or linear) and commit. The debate over which model is "right" is unwinnable; the cost of having no model is worse than the cost of an imperfect one.
  2. Add a "how did you hear about us?" field to every demo form. Free-text, optional, no autofill. The self-reported answers become the parallel signal that catches what the tracking misses. Aggregate quarterly; compare against your platform's attribution claims.
  3. Implement incrementality on your top 3 budget channels. Hold out 5-10% from each channel quarterly. Measure incremental conversions. Adjust budget based on the incremental number, not the attribution number. This single discipline change reshapes spending allocations more than any model choice.
  4. Tag every outbound send with its triggering signal. Build the signal-level attribution layer alongside the channel-level layer. Eventually, the signal layer will replace the channel layer as the operating metric for budget allocation; in the meantime, run both.
  5. Build a quarterly "model audit" cadence. Once a quarter, compare the attribution platform's claims against (a) self-reported attribution, (b) incrementality test results, (c) sales-rep qualitative input on which deals actually came from what. Where they disagree, investigate. Where they agree, increase confidence.
  6. Tie attribution to budget decisions explicitly. Attribution that doesn't change spending is decoration. Every quarterly attribution review should produce a budget reallocation proposal: "based on the data, we're moving $X from Y to Z." If nothing changes, the attribution function isn't doing its job.
  7. Communicate attribution as ranges, not point estimates. "LinkedIn drove between $300K and $500K of incremental pipeline last quarter" is honest. "LinkedIn drove $412K" is fake precision. The leadership team needs to develop tolerance for attribution ranges; otherwise they'll keep demanding spurious accuracy.

08The tools landscape (2026)

The attribution tools market segments roughly into three tiers, each with different strengths and pricing.

Channel-level attribution (the mainstream)

HubSpot Attribution, Salesforce Campaign Influence, Bizible (now Adobe Marketo Measure) — these are the dominant tools. They handle the basic touch tracking, model selection, and dashboard reporting. They do not handle incrementality testing natively (you bolt it on with custom holdout logic) and they do not handle signal-level attribution at all. Price: $0-15K/year embedded in your existing CRM/marketing-automation contract for most teams.

Multi-touch + dark-funnel (the premium tier)

Dreamdata, HockeyStack, CaliberMind — newer entrants focused on better multi-touch modeling, server-side tracking (to survive iOS/cookie restrictions), and self-reported attribution integration. They're a meaningful upgrade over the mainstream tools but still channel-centric. Price: $20-100K/year depending on data volume.

Signal-level attribution (the emerging tier)

Mama, Common Room, Champify (limited) — tools that capture the triggering signal at time of contact and tie it to downstream outcomes. This is the newest tier and the most strategically valuable for signal-anchored outbound teams. Most installs are still custom; productized signal-level attribution dashboards are still maturing. Price varies — typically bundled with signal-mining + brief generation rather than sold as standalone attribution.

What to actually buy

Honest advice for most B2B SaaS teams:

  • $5-50M ARR: Use your CRM's built-in attribution. Add self-reported attribution via forms. Run quarterly incrementality on your top 3 channels manually. Don't overspend on attribution tooling.
  • $50-200M ARR: Upgrade to a premium attribution tool (Dreamdata or HockeyStack). Make incrementality testing a standard quarterly process owned by RevOps. Start tagging outbound with signal data.
  • $200M+ ARR with serious outbound motion: Add signal-level attribution via Mama or equivalent. The marginal insight from understanding which signals drive replies is worth more than the marginal accuracy improvement of better multi-touch modeling.

09Common mistakes

Mistake 1 · Treating attribution as a math problem
Attribution is a causal inference problem, not a credit-distribution problem. Better math (Shapley values, Markov chains, ML) applied to observational data without a control group is sophisticated guessing, not measurement. The only honest improvements come from adding experimental discipline (holdouts, incrementality), not from picking a fancier model.
Mistake 2 · Picking the model that makes your team look good
If Marketing picks first-touch and Sales picks last-touch and both insist the other is wrong, the underlying problem isn't the model — it's the comp plan. Until shared-credit comp is implemented, the attribution wars are a symptom, not a cause. Fix the incentive before fixing the math.
Mistake 3 · Ignoring self-reported attribution
The "how did you hear about us?" answer often disagrees with your tracking platform — and the buyer's answer is often closer to truth. Self-reported attribution is the cheapest way to surface dark-funnel activity that's actually driving revenue. Teams that skip this field on demo forms are flying blind on 30-50% of their pipeline causation.
Mistake 4 · Reporting attribution to 3 decimal places
"LinkedIn drove $412,371 of pipeline last quarter" is fake precision dressed as data. The honest version: "LinkedIn drove somewhere between $300K and $500K of incremental pipeline, with confidence based on our holdout test." Leadership teams that demand fake precision are training their analytics team to lie smoothly.
Mistake 5 · Confusing channel attribution for signal attribution
"Cold email drove 40 meetings" is true and useless. "Funding-signal-anchored cold email drove 28 of those 40 meetings; the other 12 were generic outbound" is true and changes the budget. Channel-level attribution is necessary but not sufficient for budget allocation in 2026.
Mistake 6 · Attribution without budget action
If quarterly attribution review never changes how money is spent, attribution is decoration. The discipline of "based on this data, we're shifting $X from Y to Z" is what makes attribution worth running. Without budget action, the dashboards are political artifacts — pretty, defensible, useless.
Try Mama free

Channel attribution tells you what's busy. Signal attribution tells you what's working.

Mama tags every outbound send with its triggering signal at time of contact, then ingests reply outcomes via Reply Loop. The result: signal-level attribution that shows which signal types drove which meetings — the actionable layer most attribution platforms can't produce. Start the trial and add the missing dimension to your dashboards.