Your marketing attribution model shows that webinars drive 40% of closed revenue. So you double down on webinars. Six months later, pipeline hasn't grown.
What happened?
Attribution models promise to answer the most important question in B2B marketing: which channels actually drive revenue? But in practice, they answer a different question: which channels get credit for revenue under an arbitrary set of rules.
After implementing attribution models at three B2B companies and watching teams make million-dollar budget decisions based on flawed data, I've learned a hard truth: attribution models create false confidence more often than they create real insights.
Here's what to track instead.
The Fundamental Problem with Attribution
Attribution models try to assign credit for a conversion across multiple touchpoints. The prospect downloaded a whitepaper, attended a webinar, received three emails, visited the pricing page, and then signed up. Which channel "caused" the conversion?
The truth: nobody knows. Not you, not the prospect, not the attribution model.
Every attribution model—first-touch, last-touch, multi-touch, algorithmic—applies arbitrary rules to incomplete data and presents the results as fact.
First-touch gives all credit to awareness. This overvalues top-of-funnel activities and undervalues everything that actually drove the decision.
Last-touch gives all credit to conversion. This overvalues bottom-of-funnel activities and ignores months of nurture that made the decision possible.
Multi-touch distributes credit across touchpoints. But which touchpoints matter more? Equal weight is obviously wrong, so you pick a weighting model (W-shaped, time-decay, etc.) based on assumptions, not evidence.
Algorithmic attribution uses machine learning. Sounds sophisticated, but the algorithm is still guessing. It can identify correlation (webinars correlate with closes) but not causation (webinars caused closes).
All of these models share the same problem: they pretend to know things they can't possibly know.
What Attribution Models Can't See
The most important factors in B2B purchase decisions often aren't tracked in your attribution model:
- The LinkedIn post from your CEO that sparked initial awareness
- The conversation with a peer who recommended you
- The analyst report that validated you as a credible option
- The internal political dynamics that influenced vendor selection
- The timing of budget availability
- The competitor's misstep that created an opening
Your attribution model gives credit to the webinar they attended. It can't see that they attended the webinar because their colleague recommended you, and their colleague found you through an untracked dark social channel.
The data you have is incomplete. The model treats it as complete. Decisions based on incomplete data presented as complete lead to misallocated budgets.
The False Confidence Problem
Here's the real danger: attribution models create confidence in bad decisions.
A marketing team sees that "content downloads" account for 25% of attributed revenue. They decide to invest more in content. Seems logical.
But what if content downloads don't cause conversions? What if prospects who are already interested in your category download content as part of their research process, and they would have converted anyway?
The attribution model can't distinguish between correlation and causation. It just shows that content downloads happen before conversions and assigns credit accordingly.
This false confidence leads to doubling down on activities that don't actually drive results.
What to Track Instead: Leading Indicators
Instead of trying to attribute closed revenue backward, track leading indicators forward.
Track: Volume of high-intent actions per channel
Don't just count whitepaper downloads. Count how many downloads lead to demo requests within 30 days.
Don't just count webinar attendees. Count how many attendees take any meaningful action afterward (pricing page visit, demo request, sales conversation).
This doesn't tell you what "caused" conversions, but it tells you which activities correlate with buying behavior. That's actionable.
Track: Time-to-conversion by channel
Prospects from paid search convert in 14 days on average. Prospects from content convert in 90 days on average.
This doesn't mean paid search is "better." It means paid search attracts different intent levels than content. Understanding that helps you set appropriate expectations and metrics per channel, rather than trying to force every channel into the same attribution framework.
Track: Channel mix of your best customers
Don't ask "which channel drove the conversion?" Ask "which channels did our best customers engage with?"
If your highest-LTV customers consistently engaged with webinars, case studies, and product tours before buying, that's signal. It doesn't prove causation, but it suggests those channels play a role in quality customer acquisition.
What to Track Instead: Channel Contribution
Instead of attribution, track contribution.
Contribution question: If we stopped this channel, what would happen to pipeline?
This is testable. Pause the channel for 60-90 days and measure impact.
I know this sounds extreme. But if you're making budget decisions worth hundreds of thousands of dollars, running controlled tests is cheaper and more accurate than trusting an attribution model's guesses.
How to test contribution:
- Pick a channel that attribution says drives 15-25% of revenue
- Pause it completely for 60-90 days (don't just reduce it—pause creates clean data)
- Measure pipeline generated during the pause vs. the prior 60-90 days
- If pipeline drops significantly, the channel contributes. If it doesn't, the attribution model was wrong.
This method isn't perfect. External factors (seasonality, market conditions) affect results. But it's more reliable than any attribution model because it measures actual impact, not correlations.
What to Track Instead: Sales Team Insights
Your sales team knows which channels produce qualified pipeline. Ask them.
Questions to ask your sales team monthly:
- Which channel consistently produces deals that close?
- Which channel produces meetings that go nowhere?
- When prospects mention how they found us, what channels come up most?
- Which lead sources require more effort to close than others?
This is qualitative, not quantitative. It's messy. But it's often more accurate than attribution models because sales is measuring actual pipeline quality, not just touchpoint sequences.
When Attribution Models Do Work
Attribution isn't useless. It works when:
You're tracking short, simple funnels.
If most conversions happen within 7 days of first touch with 1-2 touchpoints, attribution is reasonably accurate. This is rare in B2B, common in e-commerce.
You're comparing channels at extremes.
Attribution might not tell you whether content or webinars drive more revenue. But it can tell you that paid search drives 10x more attributed revenue than display ads. At extremes, even flawed models show signal.
You're using attribution for hypotheses, not decisions.
"Attribution suggests webinars correlate with conversions. Let's test whether improving webinar quality increases conversion rates" is fine.
"Attribution shows webinars drive 40% of revenue, so we're shifting 40% of budget there" is not.
The Honest Alternative
Instead of pretending you know which channels drive revenue, be honest about uncertainty:
"We don't know exactly which channels cause conversions, but here's what we do know:
- Prospects who engage with [channels X, Y, Z] convert at higher rates
- When we pause [channel A], pipeline drops
- Sales says [channel B] produces the most qualified meetings
- Our best customers consistently engaged with [channel C] before buying"
That's not as neat as "webinars drive 40% of revenue." But it's honest. And honest data leads to better decisions than false precision.
Most B2B purchase decisions happen through complex, multi-touchpoint journeys that no attribution model can accurately track. Stop pretending otherwise, and start tracking the things you can actually measure: intent signals, contribution tests, and sales feedback.
That's not perfect. But it's better than false confidence in bad models.