B2B Marketing Attribution Doesn't Work for Agencies. Here's What to Measure Instead.
I sat in on a quarterly review with an agency founder a few weeks ago. Good shop. Around $3M, technical work, the kind of team that can build almost anything.
His co-founder pulled up the CRM on the big screen. Deal after deal, the source field said the same thing. Referral. Word of mouth. Referral again.
Then the co-founder asked the question. "We build attribution dashboards for our ecommerce clients every day. Why can't we measure our own marketing like that?"
Everyone looked at the head of marketing. She said she'd get better tracking in place. That promise is the trap, and I want to talk you out of it.
B2B marketing attribution doesn't work for an agency, and no tool you buy is going to change that. The co-founder's question has a real answer, and it has nothing to do with better tracking. We're trying to measure the wrong thing with the wrong instrument. Once you understand why, you stop apologizing for the fog and start measuring what actually tells you something.
Why ecommerce attribution works and yours doesn't
Start with the part nobody explains. People say "B2B isn't ecommerce" like it settles the argument. It doesn't. It just names the difference without explaining it.
The mechanism is simple. The number of touchpoints someone needs before they buy scales with how risky the purchase feels. Low risk, few touches. High risk, many touches. Price is a decent stand-in for risk, but the real driver is how much someone stands to lose if they're wrong, and how hard it is to undo.
Buying a $40 phone case is low risk. It's cheap, it ships back easily, and if you hate it you're out almost nothing.
So the journey is short. You see an ad, you click, you buy. The last click really is most of the story, which is why last-click attribution works fine for that phone case.
Hiring your agency is the opposite. It's expensive, it's hard to reverse, and if the buyer picks wrong they might lose their job over it. So the journey stretches out.
Nine months. A dozen or more touches. A podcast episode, a LinkedIn post someone forwarded, a conference talk, two calls, a proposal, a reference check.
I call this the Consideration Continuum. On one end, low-risk purchases with short journeys where last-touch tracking basically tells the truth. On the other end, high-risk purchases with long journeys where the last touch is almost meaningless.
Your agency lives at the far end. Your ecommerce clients live at the near end. Same tools, completely different reality.
The higher the risk, the longer the journey. Ecommerce sits where the last click still tells the truth. Your agency sits in the shaded zone, where the software goes blind.
So when your dashboard turns to fog, that isn't a discipline problem. You didn't fail to set up tracking correctly. You're aiming a short-journey tool at a long-journey decision. The fog is the honest output.
The Final Step Fallacy: Why You're Funding the Wrong Touchpoint
“The last touchpoint is the cashier, not the salesperson.”
This is where the wrong model starts costing real money.
When attribution credits the last touch, it credits whatever happened right before the deal closed. Usually that's a branded search, or a demo request, or a "let's talk" email. So the software points at those and says, that's what's working. Put more money there.
Think about what the last touch actually is. By the time a buyer types your name into Google, they've already decided. The branded search didn't persuade them. It's just the last step in a decision they'd already made months earlier.
The last touchpoint is the cashier, not the salesperson. The cashier rings up the sale, so if you only watch the register, the cashier looks like your best closer.
But the cashier didn't sell anything. The work happened out on the floor, long before anyone reached the counter. Paying the cashier a bigger commission won't sell one more unit.
That's the Final Step Fallacy. You fund the step that was always going to happen and starve the touches that did the persuading. The podcast, the writing, the talk, the slow work of becoming someone worth calling. Those get labeled "unmeasurable," and unmeasurable things get cut first when budgets tighten.
I've watched agencies defund the exact thing that was building their pipeline, because a dashboard told them it drove zero revenue. It drove almost all of it. The dashboard just couldn't see it.
The Final Step Fallacy hides the touches that build your pipeline before the last click. Signal Check reads the leading indicators of whether your positioning is landing with the buyers you want, so you can stop guessing which marketing to protect.
Run the free Signal CheckTwo things that make agency attribution even harder
The continuum is the main event. These two are secondary, so I'll keep them short.
First, sample size. Attribution models borrow their credibility from statistics, and statistics need volume. An ecommerce brand runs thousands of conversions a month, so patterns actually mean something.
You close somewhere between fifteen and twenty-five deals a year. There is no model, single-touch or multi-touch, that produces a trustworthy pattern from twenty data points. The math simply isn't there, and no vendor selling you a model wants to say that out loud.
One honest exception. If you run a high-volume productized shop, dozens or hundreds of small, low-risk sales a year, this is less of your problem. You've got the volume, and you sit lower on the continuum, so more of your tracking tells the truth. This piece is for the agency selling a handful of big, considered engagements, where every deal is its own nine-month decision.
Second, the deals close where software can't look. A prospect hears about you on a podcast, mentions you in a Slack group, gets a "you should call these guys" from a peer at dinner. None of that shows up in your analytics.
Refine Labs studied this and found roughly a 90% gap between what software credited and what buyers said drove them. In one case a podcast that buyers named as the source of about half the revenue was credited with zero. [Cite and link Refine Labs "Attribution Mirage" at publish.] Across 70+ interviews with agency founders on my own podcast, the story is the same. The deal traces back to a conversation, and the conversation traces back to something the founder published months earlier.
What the software credited, next to what buyers actually said. Podcast and community score zero on the dashboard and carry most of the revenue in real life. Source: Refine Labs, "The Attribution Mirage."
Which brings up "referral." It's the most common source label in an agency CRM, and it's the least useful. Referral tells you who passed your name along. It doesn't tell you where that person first ran into you, and that's the part that actually matters. The label ends the inquiry right where it should start.
What to measure instead
“Models need thousands of conversions. Stories work at twenty.”
If the number is fake, chasing a better number is a dead end. What you need is a better instrument. For a long, high-consideration sale, the only instrument that captures the whole journey is the buyer's own story.
Ask people how they came to you, and actually listen to the answer. Skip the dropdown. A dropdown forces a nine-month journey into one word, that word is usually "referral," and you're right back in the junk drawer.
Use open text and real sentences in the buyer's own words. Then, on the first call, ask them to walk you through how they ended up talking to you. At close, write the source down in a sentence.
Now, if you've been around this problem, you might be thinking you already tried this. You added the "how did you hear about us" field months ago and it didn't help. Honestly, I made the same mistake. I treated the answers like a scoreboard to tally instead of stories to read.
That's the whole shift. You stop averaging the twenty answers and start reading them as twenty stories.
The value was never in counting the tag. It's in the pattern across the narratives. You're reading for what they encountered first, who earned their trust, what finally tipped them into reaching out, and how long the whole thing took.
Ten deals a year of that, gathered honestly, is a research corpus your competitors don't have. Models need thousands of conversions. Stories work at twenty.
This costs you something real. You give up the clean single number that makes a quarterly slide look finished, and you trade it for a stack of stories that takes discipline to collect at every close. It's slower to build and harder to hand off. It will never fit in one cell of a spreadsheet.
Then track leading indicators instead of per-channel ROI. Watch whether branded search is climbing. Count how often people say "I've been following your work" on a first call.
Track your inbound fit rate, whether the leads showing up look more like your best clients over time. Compare cohorts before and after a marketing push, and label that comparison what it is, which is directional, not proof.
One number you can't defend, or three instruments you can. The stack works at twenty deals a year. The single ROI figure never did.
This is exactly what we built Signal Check to surface. It's a free diagnostic that reads the leading indicators of whether your positioning is landing with the buyers you want. It won't hand you a per-channel ROI number, because that number would be fiction. It reads the real signals sitting underneath it.
Five questions, drawn straight from the piece above. Answer honestly about how your agency runs today.
Answer all five questions first.
Shorten the Journey With a Fixed-Price Intro Offer
There's a second move, and it follows straight from the continuum. If attribution gets easier as risk drops, then lowering the risk of the first purchase pulls that first engagement toward the easier end of the curve.
Give a prospect a fixed-scope, fixed-price way to work with you before the big commitment. A defined starting point with a clear deliverable and a set price. For the prospect, it's a low-risk way to find out if you're any good. For you, it's paid discovery, so you learn whether you fit and get paid to do it.
I want to be careful here, because this is where people overclaim. A smaller first purchase makes attribution a little easier, not solved. It's lower on the continuum, so the journey is shorter, so the last touch means slightly more.
It's still more art than science. Anyone who tells you a starter offer fixes attribution is selling something.
One honest side benefit. You'll close more of these small engagements than full ones, which means more buyers, which means more stories and faster feedback. That's better learning, still not statistics.
This is the work we do in the Bottleneck, our fixed-scope diagnostic engagement. It's how my own practice is built, on purpose, for exactly the reasons above. The lower-risk first step is better for the client and it teaches me more, faster. I'd rather take my own medicine than sell you a model I don't believe in.
The Commodity Pull: Why Easier-to-Close Work is the Wrong Goal
“The thing that makes premium work hard to attribute is the same thing that makes it valuable.”
The continuum has a second lesson in it, and it's easy to miss. Where you sit on it is not fixed. You choose it with your positioning.
Commodity work sits low. Buyers treat it as interchangeable and shop hard on price, so they decide fast. Premium, specialized work sits high, where buyers take months and decide on trust, with price the last thing they weigh.
Commodity work is also easier to attribute, and easier to close. The journey is short and rational, so the last touch means something and the dashboard looks clean. When leadership pushes for measurable marketing, the quiet answer the numbers keep suggesting is to sell simpler, cheaper work that tracks more cleanly.
That is the wrong pressure on your agency.
The unmeasurable end of the continuum is where your margins live. It's where you stop competing on price and a good reputation compounds. The thing that makes premium work hard to attribute is the same thing that makes it valuable. Trust gets built across a dozen touches no tool can see.
Margin lives on the premium plateau. The pull into the commodity canyon is strong because that work is easier to close and easier to measure. Easier is not better.
So a clean attribution number does more than report. It pulls you, slowly, toward commodity work, dressed up as discipline. The best-run agencies feel that pull and refuse it. They would rather carry a fuzzy number for premium work than a precise one for a race to the bottom.
This is also why the starter offer is a doorway, not a discount. You lower the risk of the first step so a good-fit buyer can say yes without betting the quarter. You don't lower the value of the work waiting on the other side. That's the difference between a low-risk entry and a commodity.
The conversation you actually need to win
“A made-up number is worse than no number.”
None of this matters if you can't hold the line in the room. So here's how I'd handle the quarterly meeting from the top of this piece.
Capture where every deal really came from, in the buyer's own words, at close. Not a dropdown. The story.
Report branded search, inbound fit rate, and "I've been following your work" mentions on a set schedule, so leadership sees movement even without a clean ROI line.
Once a year, line up your marketing eras against your pipeline eras and look honestly at what tracked with what.
It would be fake precision. The day a partner learns the number was invented, you lose the credibility you were trying to buy. That debt comes due at the worst possible time.
If I were in your shoes, here's roughly what I'd say out loud:
"I can tell you, deal by deal, where our clients say they actually came from, and I'll bring those stories to every review. I can show you branded search climbing, our fit rate improving, and how many buyers mention our content on the first call. What I won't do is hand you a dashboard that says LinkedIn drove 22% of revenue, because at our deal count that number would be fabricated, and I don't want us making budget decisions on a fabricated number."
That answer sounds more confident than "I'll get better tracking in place," because it is. It's the difference between apologizing for the fog and explaining it.
What to do this week
You don't need a project to start. You need three small changes.
Switch your "how did you hear about us" field to open text today. Kill the dropdown that manufactures the junk drawer.
Put one line in your first-call template: walk me through how you ended up talking to us.
Pull your last five closed deals and write the real source, in a sentence, from memory. That's your first five data points.
And stop shopping for the tool that will finally fix this. That's the most common move and the most expensive one, because the tool can't fix a problem that isn't about tooling.
Go back to that quarterly meeting. The next time your co-founder asks why you can't measure like your ecommerce clients, you won't promise better tracking. You'll walk them through the continuum, name the fallacy, show the real stories, and hold the line on the number you won't fake. Not a cleaner dashboard, a truer one.
