When Attribution Is the Wrong Metric

Q: How often should attribution models be updated?

Monthly is standard. Quarterly is acceptable if the business is stable. Do not update attribution models in response to a single week of data - wait for 4 weeks minimum. When you update, compare the new model to the old one. If attribution for a channel shifts by > 20%, investigate the cause before acting on it.

Why Attribution Fails for DTC

Attribution models rank channels by their apparent contribution to revenue. Last - click attribution gives 100% credit to the final touchpoint. Multi - touch models distribute credit across the journey. Neither solves the core problem: they measure correlation, not causation.

A customer who sees a Facebook ad, then searches your brand name, then clicks email, then buys is counted as an email conversion in last - click. But email didn't cause the purchase - the prior touchpoints did. Reallocating budget to email based on this data is a mistake.

The failure is structural. Attribution cannot answer: what would have happened without this touchpoint? It cannot isolate the incremental impact of a channel. It cannot account for brand lift, awareness decay, or competitive pressure. It reports what happened, not what caused it.

The Three Failure Modes

First - touch bias occurs when early awareness channels (organic search, social) are undervalued because they rarely close the sale. Operators cut awareness budgets and watch ROAS improve on last - click metrics, then watch revenue decline 6 - 8 weeks later as the funnel empties.

Second - click inflation happens when a channel becomes a "conversion sink." Email, SMS, and retargeting naturally sit late in the funnel. Attribution gives them outsized credit. Operators increase spend, the channel becomes saturated, and unit economics collapse.

Third - channel cannibalization occurs when attribution cannot distinguish between incremental and non - incremental conversions. A customer who would have bought anyway, but saw your retargeting ad first, is counted as a retargeting win. Budget shifts to retargeting. Incremental volume stays flat. ROAS improves while revenue stalls.

When Attribution Is Useful

Attribution works best as a diagnostic tool, not a budget allocation tool. Use it to spot anomalies: if email shows 40% of attributed revenue but represents 2% of spend, investigate whether email is truly driving incremental sales or capturing existing demand.

Attribution is also useful for understanding journey length and complexity. If the average customer touches 4 - 5 channels before converting, that's a signal to invest in retention and frequency, not just top - of - funnel volume.

Within a single channel, attribution can help optimize sequence. If customers who see email 1, then SMS, then email 2 convert at 3x the rate of those who see only email 1, that's actionable. The comparison is internal and relative.

What to Measure Instead

Incrementality testing is the gold standard. Run holdout tests: show ads to a test group, withhold them from a control group, measure the difference in conversion rate. This isolates causation. A 2 - 4 week test with 10,000+ users per group yields reliable results. Cost is 1 - 3% of channel spend.

Cohort analysis by first touch is more reliable than multi - touch attribution. Segment customers by the channel that first brought them to the site. Track their lifetime value. A cohort with 30% LTV from paid search and 15% from email suggests paid search is driving higher - quality customers, not that email is undervalued.

Unit economics by channel matter more than attributed revenue. Calculate: customer acquisition cost (CAC) = spend / new customers acquired. Payback period = CAC / (average order value - fulfillment cost). Channels with payback under 60 days are sustainable. Those over 120 days are not, regardless of attribution.

Benchmark against benchmarks, not attribution. If your email ROAS is 3:1 and industry average is 2.5:1, that's meaningful. If your email ROAS improved from 2.8:1 to 3.2:1 month - over - month, that's a signal to test, not to reallocate budget.

The Operator's Checklist

Before trusting an attribution report, run through this checklist:

Is this model measuring causation or correlation? (Correlation only - flag it.)
What percentage of conversions are attributed to the last touchpoint? (Above 60% - model is likely biased.)
Have I tested incrementality for this channel? (No - do not reallocate budget based on attribution alone.)
What is the actual CAC and payback period for this channel? (If payback > 90 days, attribution is irrelevant.)
Does this channel sit early or late in the funnel? (Late - stage channels will always appear overvalued.)
What would happen if I cut this channel by 50%? (If revenue drops more than 50%, it's driving incremental value.)

A Practical Decision Rule

Use attribution to identify which channels to test, not which channels to fund. If attribution shows email at 35% of revenue, that's a signal to run an incrementality test on email. If the test shows email is 80% incremental, increase spend. If it shows 20% incremental, cut spend. Attribution alone is not enough.

For mature channels with stable performance, use attribution as a sanity check. If email ROAS drops from 4:1 to 2.5:1 and attribution shows email's share of revenue fell from 30% to 18%, something changed. Investigate. But do not cut email budget based on attribution alone.

For new channels, ignore attribution for the first 60 days. Focus on CAC, payback period, and incrementality. Once the channel has 500+ conversions and 8+ weeks of data, attribution becomes a secondary metric.

Questions

FAQ

Should DTC brands use multi - touch attribution or last - click?

Neither should drive budget decisions. Multi - touch attribution is more honest about the customer journey, but it still cannot measure causation. Use it for diagnostics only. Last - click is simpler and equally flawed. If forced to choose, multi - touch is less likely to cause budget mistakes, but the difference is marginal.

What sample size do I need for an incrementality test?

Minimum 5,000 users per group (test and control) for statistical significance at 95% confidence. For channels with low conversion rates (< 2%), aim for 10,000 per group. Run for 2 - 4 weeks to account for day - of - week effects. Holdout tests are more reliable than matched cohorts.

How often should attribution models be updated?

Monthly is standard. Quarterly is acceptable if the business is stable. Do not update attribution models in response to a single week of data - wait for 4 weeks minimum. When you update, compare the new model to the old one. If attribution for a channel shifts by > 20%, investigate the cause before acting on it.

Can attribution work for subscription or membership businesses?

Attribution is even less reliable for subscriptions because the conversion event (first purchase) is not the revenue event (lifetime value). A channel that acquires low - LTV subscribers will appear valuable on first - purchase attribution but destroy unit economics. Always measure LTV by cohort and first - touch channel, not attributed revenue.

Want this on your account?

Thirty minutes. Bring the number that keeps you up.

Book a demo