← Contents17 / 20

Module 17 — The Credit Wars: Attribution

"Attribution is the only field in science where every model is wrong, every team is certain, and the budget moves anyway. Pick the lie that funds the right things." — Dr. Tanya Vex, keynote, a conference you regret expensing

It is 4 a.m. and two grown adults with combined comp north of seven hundred grand are screaming at each other in a Slack thread about who sourced the Pendleton deal. Marketing has a webinar registration from fourteen months ago. Sales has a golf outing and a text message. The customer, asked directly, said they bought because "a guy I trust mentioned you on a podcast I can't remember." Nobody logged the fucking podcast. Nobody can log the podcast. The podcast is the truth and the podcast is gone, scattered into the digital ether like confetti at a funeral. Welcome to attribution: the most expensive, shit-soaked argument in the building, conducted entirely over data that does not exist, by people who are absolutely certain they're right.

I have watched attribution debates consume six figures of salary-per-hour in a single QBR. I have seen a VP of Marketing and a VP of Sales nearly come to actual physical blows over a Salesforce report that was wrong in four different directions simultaneously. Neither of them was right. The customer bought because their old vendor raised prices during a merger and their champion was scared. That information lives in zero dashboards. It will never live in zero dashboards. And yet here we are, screaming about the damn webinar registration.

THE JOB — What Attribution Actually Is (And Stop Bullshitting Yourself)

Attribution is the practice of assigning credit for a closed deal to the marketing and sales touches that influenced it. That's the official version. The real version: attribution is how you decide where the next dollar of budget goes, and who gets to feel important at the all-hands. It is a resource-allocation tool wearing a truth-seeking costume, and it fucking knows it.

Here is the thing nobody at the off-site will say out loud, so I will say it here, loudly, before the caffeine wears off: you cannot actually know why someone bought. The buyer doesn't know. It was a podcast, a Slack group, a former colleague, a billboard, a bad night, a competitor's pricing change, and one of your forty-one touchpoints, in some unknowable proportion, run through the wet meat of a human decision by a person who was also worried about their kid's soccer schedule during the discovery call. Attribution is not measurement. Attribution is modeling under permanent uncertainty. Anyone who tells you their attribution is "accurate" is selling a course, and the course is also wrong.

So why do it at all? Because the alternative — distributing six- and seven-figure budgets on vibes and seniority and whoever yelled loudest in the last leadership sync — is catastrophically worse. A wrong-but-consistent model that informs decisions beats a perfect model that doesn't exist. You're not solving for truth. You're solving for less-stupid budget allocation and less-insane internal politics. That's the goddamn job. Lower your expectations to "directional and defensible" and you will outperform every asshole in your peer group who thinks their model is gospel.

RULE No. 41: Attribution is politics with a math costume on. This is not a flaw. This is the whole sport. Understanding it is your competitive advantage over every person in the building who thinks they're doing science.

THE PLAYBOOK — Models, Math, and the Sourced/Influenced Split

1. Know the models and precisely what each one lies about

A touch is any logged interaction — ad click, form fill, email open, demo, SDR call, conference badge scan that one rep entered three months late. An attribution model decides how to split 100% of the credit across the touches on the path to close. Here are the ones that matter, and what each one is quietly lying to you about:

  • First-touch (FT): 100% credit to the first interaction. Lies toward: top-of-funnel / demand-gen channels. Good for "what creates awareness," absolutely fucking useless for "what closes deals." Your paid social team will love this model. That should tell you something important about who is motivated to advocate for which model.
  • Last-touch (LT): 100% credit to the final touch before the opportunity was created. Lies toward: bottom-funnel and direct/branded search — which is demand you already created through months of effort, and now you're crediting yourself twice for it. The default in most CRMs because it requires zero configuration and zero thought, and it is the laziest, most dishonest model in the entire canon. If your CMO is running last-touch attribution and reporting it as "how Marketing is performing," ask them why. Wait for the answer. Watch them sweat.
  • Linear: equal credit to every touch on the path. Fair, simple, and blandly wrong in a way that generates the least political heat. It pretends a throwaway email open that received three seconds of unfocused attention matters as much as the demo where the champion said "holy shit, this is exactly what we've been trying to solve." It doesn't. Nothing is equal in a B2B deal, and pretending otherwise is a form of cowardice dressed up as fairness.
  • Time-decay: touches closer to close get exponentially more credit, weighted by something like 2^(−Δt/halflife). Lies toward: late-stage and sales touches. Useful for short, transactional cycles. Dangerously misleading in long enterprise deals where the awareness-generation work that happened eighteen months ago was genuinely load-bearing and deserves real credit for it.
  • U-shaped (position-based): 40% to first touch, 40% to the lead-creation touch, 20% spread across the middle. Rewards demand creation and conversion — the two moments that actually matter — and mostly stops the bullshit fight over the middle.
  • W-shaped: 30% / 30% / 30% to first touch, lead-creation, and opportunity-creation; 10% spread across the rest. The honest workhorse for B2B because it credits the three real inflection points without pretending any single one of them closed the goddamn deal.
  • Full-path / Z-shaped: W-shaped plus a heavy weight on the closed-won touch. For long enterprise cycles where the late-stage motion is genuine work that moves paper through procurement hell and deserves real credit.
  • Data-driven (algorithmic): instead of you sitting in a conference room and decreeing weights based on feelings and seniority, the model learns weights from your actual conversion data. Two real methods worth knowing:
    • Markov chains: model the buyer journey as a sequence of states; compute each channel's removal effect — how much does overall conversion drop if you delete that channel from every path? That drop is its credit. Elegant, counterfactual, and the closest thing to asking "what would actually happen if we killed this channel?" that math currently provides.
    • Shapley value: borrowed from cooperative game theory, where it was invented to answer "how do you fairly split value among players who contributed differently to a collective outcome?" Credit equals a channel's average marginal contribution across every possible ordering of touches. Mathematically defensible, computationally brutal on large datasets, and the closest thing to a correct answer that exists in this whole sorry discipline. Your data science team will groan. The groan is appropriate. Run it quarterly, not in real-time.

RULE No. 42: First-touch and last-touch are not analysis. They are the two cheapest opinions available, and the CRM hands them to you for free precisely so you'll stop thinking. The CRM is not your friend.

2. Separate SOURCED from INFLUENCED — and never, ever let anyone blur them

This is the single most important distinction in attribution, and the one most aggressively weaponized by people who know exactly what they're doing. Burn it into your skull:

  • Pipeline/Revenue SOURCED: the channel or team that created the opportunity — the first-touch that generated the original lead and kicked off the whole chain. Single-credit. Winner-take-all. One owner per deal, full stop.
  • Pipeline/Revenue INFLUENCED: every channel and team that touched the deal anywhere on the path from unknown to closed. Multi-credit. A deal can be legitimately "influenced" by twelve different things.

The math you must never, under any circumstances, let slide: sourced numbers sum to 100% of pipeline. Influenced numbers sum to way more than 100% — because the same deal gets counted by every function that touched it anywhere on the path. When Marketing claims it "influenced 94% of closed-won revenue," that is technically true and strategically worthless bullshit — they influenced it by dropping a nurture email into an inbox attached to a deal that Sales hunted, cooked, dragged through legal, and closed over three goddamn quarters. Influenced is a coverage metric, not a credit metric. It tells you "were we present at some point?" — not "did we cause anything?" Report both. Label both precisely. Never add them together on the same slide. Never let anyone in the room merge them without correcting it loudly, in front of witnesses, with a Sharpie if necessary.

3. Self-reported attribution — the dumb-ass question that beats every smart model

Put a free-text or dropdown field on every demo request, signup, and contact form: "How did you hear about us?" Yes, it's unscientific. Yes, people lie, misremember, write "Google" when they mean "a colleague texted me at 9 p.m.," and pick the first option in the dropdown without reading. It is also the only instrument you have for capturing dark social — the podcasts nobody tracks, the Slack communities that live behind login walls, the Reddit threads at 11 p.m., the "a friend from my last company swore by it," the word-of-mouth that never fires a single tracking pixel and is therefore completely, utterly invisible to every sophisticated model in step 1. That dark social is real pipeline. It might be your best pipeline. You cannot see it with click-tracking. You have to ask.

Your attribution dashboard says "direct / organic." Self-reported says "your CEO's tweet in February." One of those is actionable intelligence. The other is your CRM lying to you with impressive confidence. Blend them: machine attribution for the trackable channels, self-reported for the dark matter. Triangulate obsessively. Never trust any single source of truth, because there is no single source of truth in attribution and anyone who tells you otherwise is selling something expensive and probably wrong.

4. Build the attribution playbook and then hold it like a hypothesis, not scripture

  1. Default to W-shaped for board-level reporting and budget decisions. It's defensible, it credits the three real inflection points, and it generates 70% fewer religious arguments than anything more exotic.
  2. Run Markov removal-effect analysis quarterly as a sanity check against W-shaped. Where they violently disagree — where one says channel X is critical and the other says it's noise — that gap is where your intuition is broken. Go investigate. Don't just pick the one that confirms what you already thought.
  3. Track all three lenses in parallel: sourced for accountability (whose job created this opportunity), influenced for coverage (what's in the mix and is anything suspiciously absent), self-reported for the dark social your tracking infrastructure can't see. Three lenses. Never merged. Never reported as if they're measuring the same thing.
  4. Report attribution as a directional input to a human decision, never as an automated verdict that automatically cuts someone's budget. The model proposes. The operator — meaning you, the tired person reading this at some ungodly hour — disposes. Any tool that makes the decision for you is a tool that will eventually destroy something that mattered and blame the algorithm.

HOW IT GOES TO HELL (And It Always Goes To Hell)

The VP of Vibes discovers influenced pipeline as a metric and never psychologically recovers from it. Every board deck henceforth reads "Marketing influenced 94% of closed-won revenue." He influenced it the way a weatherman influences the weather by standing near it and wearing a jacket. The CRO knows the number is shit. The board doesn't know yet. This works until the budget review, when someone divides influenced pipeline by marketing spend and asks why the hell the ratio keeps changing. That's a bad meeting. That's a meeting where careers develop cracks.

The Brilliant Jerk in Sales — the one closing 40% of the number while refusing to log a single touch in the CRM — watches attribution models credit Marketing with deals he ran solo from cold email to signed contract. He is correct that the model is lying. He is wrong that the solution is to poison the data by logging garbage, submitting fake activities, and screaming in the channel. Now attribution is a grudge with a dashboard. The meetings are a fucking warzone and the model is wrong in a new and worse direction.

The Guru — Lance "The Closer" DiMarVo — sells a $6,000 cohort called "Last-Touch Truth: Own Your Close" because last-touch attribution makes his late-stage tactics look like the only thing that ever mattered in the history of B2B sales. Last-touch credits the closer and erases eighteen months of demand generation, content, events, and SDR effort that loaded the gun he pulled the trigger on. Lance loves last-touch. Lance is the last touch. Lance is the world's most expensive weather vane pointing at himself.

The RevOps Martyr spends eight weeks building a mathematically gorgeous Shapley value model, documents it thoroughly, presents it in a deck that is frankly a piece of work, and watches leadership ignore it entirely in favor of the last-touch report from the standard Salesforce dashboard that takes three goddamn clicks to open and has been wrong since it was configured. Correct and unused is just expensive. The model lives in a browser tab that hasn't been opened since the Q3 presentation. The Martyr knows. The Martyr has moved on, internally, to somewhere quieter where they don't have to watch good work die in front of people who won't read it.

The Founder demands "one number for how much Marketing is worth" because the board meeting is in nine days and he needs a clean story. There is no one number. There will never be one number. The Founder gets a number anyway — whichever model flatters the prior he walked in with, selected with the kind of confidence that only comes from not understanding the methodology — and reorgs the budget around a lie with a decimal point and a percentage sign. This is precisely how you defund demand generation for two quarters and then stand in a September pipeline review asking why the hell top-of-funnel collapsed. I have watched this exact sequence play out four times at four different companies. It gets more depressing, not less, with each repetition.

FIELD RULES

  1. Attribution is politics with a math costume. Pick your model before you know the answer, write down why you picked it, document it somewhere people can actually find it, and you will fight 90% fewer bullshit battles than everyone who picks the model after seeing what it says.
  2. Sourced sums to 100%. Influenced doesn't. Self-reported catches the dark. Use all three lenses. Label all three, clearly, every time. Never add them together in the same chart unless your goal is to lie to powerful people, which, my attorney notes, has consequences.
  3. First-touch and last-touch are the lazy defaults. B2B has long, multi-touch, multi-stakeholder journeys that span months. Model them with W-shaped or data-driven, or stop pretending you measured anything meaningful and go home.
  4. Self-reported "how did you hear about us?" is unscientific and absolutely indispensable. It is the only window into word-of-mouth and dark social you will ever have. Do not overthink it. Just ask. Log the answer. It will surprise you.
  5. The buyer doesn't fully know why they bought. Neither do you. Neither does the model. Use attribution to allocate budget less stupidly, generate fewer insane arguments, and stop having the same goddamn blood feud about the Pendleton deal in every QBR — not to crown a hero or settle scores that should have been buried in Q1.
  6. A model nobody trusts is worth less than no model at all. If your teams are gaming the data, fighting the methodology in Slack at midnight, or ignoring the output entirely in favor of their feelings, you don't have an attribution system. You have an expensive, politically-charged spreadsheet that everyone pretends to believe and nobody actually uses. Kill it, rebuild the process, and start with a model people agree to before they know which one benefits them. That conversation is uncomfortable. Have it anyway.

From the field: It is 4 a.m. and I am staring at four attribution reports on the same closed-won quarter. They credit Marketing with 71%, 44%, 38%, and 19% of revenue, respectively. All four are running on the identical underlying data. All four are "correct" by their own internal logic. On Tuesday, someone in a Patagonia vest will pick one of these numbers, put it in a deck, and make a headcount decision based on it. I am going to recommend W-shaped, annotate the slide "directional, not definitive, do not fire anyone based solely on this," and expense the bourbon as professional development. The bourbon is the most defensible line on the report.