How Operators Think About Attribution

Why Attribution Matters to Operators

Attribution determines budget allocation. If a channel appears to drive 30% of revenue, it typically receives 30% of spend. If the model is wrong, capital flows to the wrong lever.

The gap between platform attribution and actual margin is where operators find leverage. Meta reports a $100K attributed order. Shopify reports the same order. But if COGS is 40%, the true margin is $60K - and if the channel's CAC is $45K, the unit economics are sound. If CAC is $55K, they're broken. Attribution alone doesn't tell you which.

Operators use attribution to answer: which channels are worth scaling? Which are declining? Which need creative refresh? The framework must be consistent enough to spot trends, not so rigid it ignores reality.

The Three Attribution Models and Their Thresholds

Last-click attribution assigns 100% credit to the final touchpoint before purchase. Threshold for use: brands with short consideration cycles (< 7 days), high impulse categories, or heavy paid search. Failure mode: undervalues awareness and top-funnel work; overstates bottom-funnel channel value.

First-click attribution assigns 100% credit to the first touchpoint. Threshold for use: awareness-heavy campaigns, brand-building phases, or when testing new channels. Failure mode: overstates top-funnel impact; ignores the work of conversion channels.

Linear attribution splits credit equally across all touchpoints. Threshold for use: mid-funnel brands with 3 - 5 typical touchpoints per customer. Failure mode: assumes all touchpoints are equally valuable (rarely true); dilutes signal from high-impact moments.

Time-decay models weight recent touchpoints higher. Use when consideration cycles are 14 - 30 days.
Multi-touch models (Shapley value, incrementality testing) are gold standard but require 6+ months of data and statistical rigor. Threshold: $500K+ monthly ad spend.
Platform attribution (Meta, Google, TikTok) is biased toward that platform's touchpoints. Discount by 15 - 25% when comparing across channels.

The Operator's Attribution Checklist

Before trusting any attribution model, operators verify three things: data completeness, model consistency, and margin alignment.

Data completeness: Are all touchpoints tracked? (Check: UTM parameters on all paid ads, organic search tagged, email platform integrated, SMS integrated, direct traffic isolated.) Missing data = biased model.
Model consistency: Is the same model applied to all channels? (Check: last-click applied uniformly, or linear applied uniformly. Mixing models across channels creates false comparisons.)
Margin alignment: Does attributed revenue match P&L revenue? (Check: attributed revenue within 5% of Shopify reported revenue. If gap > 10%, investigate: duplicate orders, refunds, discounts, or tracking gaps.)
Holdout test: Does the channel still drive orders if you pause it? (Check: pause channel for 1 week, measure order lift/decline. If orders don't drop, attribution is overstating impact.)

Common Attribution Failure Modes

Correlation mistaken for causation. A customer sees a retargeting ad and buys the next day. Attribution credits retargeting. But the customer may have bought anyway - the ad was the last touchpoint, not the cause. Mitigation: run incrementality tests (holdout groups) quarterly.

Cross-device blindness. Customer browses on mobile, buys on desktop. If tracking doesn't stitch devices, the purchase appears organic. Mitigation: implement first-party data layer (Segment, mParticle) and cross-reference CRM.

Attribution creep. Platform reports improve over time. Meta's attributed revenue grows 20% YoY while actual revenue grows 5%. The platform is capturing more credit, not the channel improving. Mitigation: lock attribution model and review changes quarterly.

Seasonality distortion. Q4 attribution looks different than Q2 - consideration cycles shorten, impulse increases, paid search dominates. Comparing Q4 to Q2 without adjustment is invalid. Mitigation: compare same quarter year-over-year.

Decision Rules: When to Act on Attribution Data

Attribution data should trigger action only when it meets three criteria: statistical significance (sample size > 100 orders), consistency (trend holds for 2+ weeks), and margin validation (attributed channel's CAC < LTV).

Scale a channel: attributed ROAS > 3:1 AND margin-adjusted CAC < 30% of LTV AND trend holds for 3+ weeks.
Pause a channel: attributed ROAS < 1.5:1 AND margin-adjusted CAC > 50% of LTV AND trend holds for 2+ weeks.
Investigate a channel: attributed ROAS changed > 30% week-over-week. (Check: creative fatigue, audience drift, tracking error, or market shift.)
Remodel attribution: platform-reported revenue diverges > 10% from P&L for 2+ consecutive weeks.

Attribution and the Full Funnel

Attribution often focuses on last-click because it's easiest to measure. But operators know awareness, consideration, and conversion are separate problems with separate metrics.

Awareness channels (brand search, YouTube, podcasts) rarely show last-click attribution but drive consideration. Measure via brand lift studies or incrementality tests, not last-click.

Consideration channels (retargeting, email, SMS) show high last-click attribution but depend on prior awareness. Measure via holdout tests to isolate true impact.

Conversion channels (paid search, affiliate, direct) show strong last-click attribution and respond to holdout tests. These are the easiest to optimize.

Operators allocate budget across all three, but measure each differently. Conflating all three under one attribution model is the root of most budget misallocation.

Building an Attribution System

Start with last-click attribution on a single model (e.g., 30-day window). Lock it for 90 days. Measure consistency, not perfection.

Add margin data: COGS, CAC, LTV. Recalculate attributed revenue as attributed margin. This is the true signal.

Run one holdout test per quarter on your highest-spend channel. Compare order volume and margin in holdout vs. control. This is your ground truth.

Review attribution model quarterly. If platform data improves or business model shifts, update the model and document the change.

Never trust attribution alone. Always cross-check with P&L, holdout tests, and customer surveys.

Questions

FAQ

Should we use multi-touch attribution?

Multi-touch (Shapley value, incrementality) is more accurate but requires 6+ months of clean data and statistical expertise. Start with last-click or linear. Upgrade to multi-touch only if: monthly ad spend > $500K, consideration cycle > 14 days, or budget allocation decisions are worth the complexity cost.

How do we handle direct traffic in attribution?

Direct traffic is usually repeat customers or organic search without UTM tags. Assign it to a 'repeat' or 'organic' bucket, not to paid channels. If direct traffic is > 30% of orders, audit your UTM implementation - you're likely losing data. Direct traffic should not be credited to paid channels retroactively.

What's the right attribution window?

Use 30 days for most DTC brands. If consideration cycle is < 7 days (impulse, flash sales), use 7 days. If > 30 days (high-ticket, B2B), use 60 days. Lock the window and don't change it mid-quarter. Changing windows mid-analysis invalidates comparisons.

How do we know if our attribution model is wrong?

Three red flags: (1) attributed revenue diverges > 10% from P&L revenue, (2) pausing a high-attributed channel doesn't reduce orders, (3) attributed ROAS improves while actual margin declines. Run a holdout test to validate. If holdout results contradict attribution, the model is wrong.

Want this on your account?

Thirty minutes. Bring the number that keeps you up.

Book a demo

Attribution as a Measurement System