Attribution Is a Lie You Tell Your Board. RevOps Can Make It Less of One

Somewhere in your company there is a slide that says marketing sourced 42% of pipeline. I would bet real money nobody in the room could say what would have to be true for that number to be wrong, and that everybody nodded at it anyway.

That is marketing attribution in most growth-stage companies. Not fraud. More like a shared agreement to stop asking questions once the number looks defensible. RevOps inherits the arrangement, then gets blamed for it. Teams almost always try to fix this by buying a better model. It rarely works, because the model is not the problem. You are asking it a question it structurally cannot answer, then treating the answer as an accounting fact rather than an opinion with a spreadsheet attached.

Why Marketing Attribution Is Always Wrong

Marketing attribution is always wrong because it can only see what it can track, and the things that move B2B deals are mostly untrackable.

Think about your own last significant purchase. Someone you trust mentioned a vendor in a Slack channel. You sat on it four months. You heard the name again on a podcast. Eventually you typed the company into Google and clicked the first result. Your attribution system recorded exactly one of those four events, and it was the least important one.

That is not a tracking gap better UTMs can close. It is a category error. Michael Kaminsky, co-founder of the measurement firm Recast, puts it bluntly: "The fundamental problem is that the thing we care about, the true incremental impact of an additional dollar spent on some marketing channel, is unknown and unknowable." No ledger anywhere records what would have happened in the world where you skipped the campaign.

The consequences show up as distrust. Forrester's Marketing Survey found 64% of B2B marketing leaders say their organization does not trust measurement for decision-making, and that 59% of CMO dashboards track sourcing metrics that do not reflect how demand actually originated. Forrester expects that to degrade by roughly 20% if nothing changes. Gartner, separately, found only 52% of senior marketing leaders succeed in proving marketing's value and getting credit for it.

Half the profession cannot prove its own worth. That is not bad luck repeated across thousands of companies. It is what happens when a discipline agrees to measure the wrong thing precisely.

The Models and What Each One Hides

Every attribution model is a hypothesis about what mattered, dressed up as arithmetic. Knowing which hypothesis you have adopted matters more than which model you picked.

First-touch assumes discovery deserves credit, which flatters top-of-funnel and makes brand and content look like heroes. Last-touch assumes the closing moment deserves it, flattering branded search and retargeting, two channels that mostly harvest demand somebody else created. If you have watched a paid search team take credit for a deal a conference conversation produced, you have watched last-touch work as designed.

Linear models spread credit evenly, which sounds fair and is mostly a way of declining to have an opinion. Time-decay favors recent touches and biases toward short sales cycles. Position-based models hard-code a guess that first touch, lead conversion, and opportunity creation are the three moments that count. Where did those weights come from? A vendor picked them. That is the whole provenance.

Multi-touch is what most teams aspire to, and it is better in that it admits journeys are plural. It is still blind to every touch that never produced a tracked event, which in B2B is most of them. Feeding a sophisticated model incomplete data does not produce a sophisticated answer. It produces a confident one, which is worse. We have written before about how dashboards on dirty data just make you confidently wrong, and attribution is that failure mode in its purest form.

Building a Marketing Attribution Model Honest Enough to Use

An honest marketing attribution model is one built for ranking and trend detection rather than truth, and RevOps should design it accordingly.

Start by writing down the decision the model exists to serve. It is almost always one of two: should we move budget between channels, or is a channel improving over time. Both are comparative. Neither requires knowing the true incremental value of a dollar, which is fortunate.

Then pick a simple model and freeze it. Choose W-shaped or first-touch or whatever survives an argument with your CMO, document the weights, and do not change them for four quarters. The most destructive habit I see in RevOps is re-tuning the model whenever a channel owner complains, which guarantees no two quarters are comparable. A stable wrong number is a measuring stick. A number that changes definition quarterly is noise with a logo on it.

Next, instrument self-reported attribution. Put a required "how did you hear about us" field on your demo form. It is unfashionable and qualitative, and it surfaces the dark-funnel channels your tracking cannot see. When self-reported and tracked data disagree loudly about a channel, that gap is the most interesting number in your business.

Finally, fix the inputs before blaming the model. Attribution sits on top of your CRM, and if opportunity stages are applied inconsistently or lead sources get overwritten on every touch, no model can rescue that. This is the unglamorous foundation work that RevOps exists to own.

Reporting Uncertainty to the Board

Report attribution as a range with a stated method, because a false decimal point costs you more credibility than an honest interval ever will.

Here is the move that changes the conversation. Instead of "marketing sourced 42% of pipeline," try "under our W-shaped model, marketing influenced between 35% and 50% of pipeline this quarter, up from last quarter on the same model, and about a fifth of closed-won deals report a source our tracking never saw." Harder to write, much harder to attack, because you named the uncertainty before anyone else could.

Boards do not punish uncertainty. They punish surprise. A director told for six quarters that attribution is directional will absorb a miss. A director shown two decimal places all year reads the same miss as a competence failure, and is right to.

Give the board three things: the direction, the confidence, and the decision it supports. Skip the mechanics unless asked. The distinction between a number that describes and a number that drives an action is the entire game, and it is worth being deliberate about which one you are actually presenting.

Spending Against Directional Truth

Once you accept that attribution ranks rather than proves, the way you allocate budget changes, and holdout testing becomes the thing that earns its keep.

A holdout is simple. Turn a channel off in a matched region or segment for six to eight weeks and watch pipeline. Your model says that channel produced 12% of pipeline. The holdout tells you whether the pipeline disappears when the channel does. Those numbers are often not close, and the gap is worth more than any model refinement you will buy. Measurement specialist Andrew Covato describes the mature version as triangulation: "You're triangulating the ground truth experiment, the MMM, and the attribution data, and putting that together in a unified package which gives you the best of everything."

You do not need media mix modeling at growth stage. You need one honest experiment a quarter. Pick the channel with the biggest spend or the loudest internal advocate, run a real holdout, and let it settle an argument attribution has failed to settle for a year.

Budget pressure makes this urgent. Gartner's 2025 CMO Spend Survey found marketing budgets flatlined at 7.7% of company revenue, so the fight is about reallocation, not growth. Every dollar you move comes from a colleague, and "the model says so" will not win that fight twice. An experiment will. The same logic applies to the predictive layer everyone is bolting onto their pipeline right now, where the honest question is always whether the signal survives contact with reality.

Attribution will still be a lie you tell your board. Told carefully, with its error bars showing and an experiment behind the decisions that matter, it becomes a useful one.

Frequently Asked Questions

What Is Marketing Attribution and Why Is It Important?

Marketing attribution assigns credit for a conversion or revenue to the touchpoints that influenced the buyer. It matters because it is the primary input most teams use to decide where the next dollar goes, which makes its flaws expensive rather than academic.

How Accurate Is Marketing Attribution?

Not very, and honest practitioners say so. Forrester found 64% of B2B marketing leaders say their organization does not trust its own measurement, and no model observes the private conversations and community recommendations where much of B2B buying happens.

What Are the Main Types of Marketing Attribution Models?

Single-touch (first-touch and last-touch), linear, time-decay, and position-based models such as U-shaped and W-shaped, plus multi-touch attribution spreading credit across the journey. Each encodes a different assumption about what mattered, and none is neutral.

What Are the Limitations of Marketing Attribution?

It only sees what it can track. Offline conversations, private communities, forwarded links, word of mouth, and long-run brand effects are invisible, which biases credit toward whichever channel happened to be last and clickable.

What Should RevOps Use Instead of Attribution?

Not instead, alongside. Pair one frozen attribution model with self-reported source data and a quarterly holdout test, then report results as a range. Attribution ranks your channels; experiments tell you what is actually incremental.

References

• Forrester, B2B Marketing Leaders Don't Trust Their Measurement
• Forrester via Forbes, B2B Marketing Measurement Isn't Trusted, And It's About To Get Worse
• Gartner, Only 52% of Senior Marketing Leaders Can Prove Marketing's Value
• Gartner 2025 CMO Spend Survey
• Recast, What Does Incrementality Mean?
• Supermetrics, Marketing Attribution in a Privacy-First World with Andrew Covato
• Automata Intel, CRM Data Hygiene
• Automata Intel, What RevOps Actually Is
• Automata Intel, The Difference Between a Dashboard and a Decision
• Automata Intel, AI Pipeline Forecasting: Useful Signal or Expensive Guesswork?