Stop Guessing on Attribution

Three Attribution Models and What They Actually Tell You

First - touch attribution credits the channel that initiated the customer journey. A user clicks a Facebook ad, leaves, returns two weeks later via organic search, and buys. First - touch gives 100% credit to Facebook.

Last - touch attribution credits the final channel before purchase. Same scenario: last - touch gives 100% credit to organic search.

Multi - touch attribution distributes credit across multiple touchpoints. Common rules: linear (equal credit to all touches), time - decay (more credit to recent touches), or custom weighted models.

The operational question is not which model is "correct" - it is which model answers the business question being asked. First - touch reveals which channels acquire new customer cohorts. Last - touch reveals which channels convert existing awareness into sales. Multi - touch obscures both by splitting credit.

Set Measurement Thresholds Before Choosing a Model

Attribution accuracy depends on data quality. Before selecting a model, establish thresholds for what counts as a valid touchpoint.

Threshold 1: Minimum time window. Define how long a customer journey can be. 7 days? 30 days? 90 days? Longer windows increase noise; shorter windows miss real influence. Most DTC brands use 30 days for paid channels and 7 days for retargeting.

Threshold 2: Minimum touchpoint count. Decide whether single - touch journeys (user sees one ad and buys) are attributed or excluded. Single - touch journeys are common in DTC and should not be discarded.

Threshold 3: Cross - device handling. A user clicks an Instagram ad on mobile, later searches on desktop, and buys. Can the platform track this? Google Analytics 4 uses cross - device modeling; Shopify's native attribution does not. Choose based on platform capability.

Threshold 4: Direct traffic classification. Decide what counts as "direct." Untagged traffic? Returning customers? Email clicks without UTM parameters? Misclassification here inflates direct and deflates paid channel credit.

Common Attribution Failure Modes

Failure mode 1: Platform bias. Facebook attributes revenue to Facebook. Google attributes revenue to Google. Each platform's attribution model favors its own channels. Reconcile across platforms monthly; flag discrepancies > 15% variance.

Failure mode 2: UTM parameter decay. UTM tags are lost when users click through multiple pages, use email, or switch devices. Audit UTM coverage quarterly. If < 85% of traffic is tagged, attribution is unreliable.

Failure mode 3: Attribution creep. As the model becomes more complex (adding custom rules, channel groupings, weighted touchpoints), it becomes harder to audit and easier to hide errors. Simpler models are more defensible.

Failure mode 4: Seasonal blindness. Attribution models trained on Q4 data (high volume, short windows) perform poorly in Q1 (low volume, longer windows). Recalibrate thresholds seasonally.

Failure mode 5: Offline - online gap. Phone orders, in - person referrals, and word - of - mouth are invisible to digital attribution. If offline revenue > 10% of total, digital attribution is incomplete.

Decision Rule: When to Use Which Model

Use first - touch attribution when optimizing customer acquisition cost (CAC) and channel mix. Question: "Which channels bring in new customers?" Answer: First - touch.

Use last - touch attribution when optimizing conversion rate and channel efficiency. Question: "Which channels close sales?" Answer: Last - touch.

Use multi - touch attribution only when you have > 10,000 monthly transactions and can afford to maintain the model. Below that threshold, the noise from multi - touch outweighs the insight.

Use a hybrid approach for mature brands: first - touch for acquisition budget allocation, last - touch for conversion optimization, and multi - touch for reporting to stakeholders (with caveats noted).

Audit Checklist for Attribution Integrity

Run this checklist monthly to catch drift:

Reconcile total attributed revenue against actual revenue. Variance should be < 5%. If higher, check for untagged traffic, bot traffic, or platform sync errors.
Audit UTM parameter coverage. Tag rate should be > 85% across paid channels.
Test a cohort manually. Pick 10 random orders from the past week. Trace the customer journey in your analytics platform. Does the attribution match the actual touchpoints?
Compare platform attribution (Facebook, Google, TikTok) against your single source of truth (Shopify, GA4). Flag discrepancies > 15%.
Review threshold changes. If time windows, device handling, or direct traffic rules changed, recalculate historical data to maintain consistency.
Check for bot and fraud traffic. Exclude IPs, user agents, and referrers known to be fraudulent before running attribution.

Building Attribution Into Workflow

Attribution is not a one - time setup. It requires monthly review and quarterly recalibration.

Assign ownership. One person (operator or analyst) owns the attribution model, thresholds, and audit checklist. Without ownership, drift accelerates.

Document decisions. Write down why you chose first - touch vs. last - touch, what your time window is, and how you handle direct traffic. This becomes the reference when questions arise.

Separate reporting from optimization. Use one model for internal optimization decisions (first - touch or last - touch). Use a different model (or disclaimer) for external reporting to stakeholders.

Test changes in isolation. If you adjust the time window or add a new channel, run the new model in parallel for 2 - 4 weeks before switching. Measure the impact on decision quality, not just on reported numbers.

Questions

FAQ

What's the difference between attribution and incrementality testing?

Attribution assigns credit to channels based on observed customer journeys. Incrementality testing measures the actual causal impact of a channel by running controlled experiments (e.g., holding back ad spend in one region and comparing results). Attribution is observational; incrementality is experimental. Both are useful. Attribution is faster and cheaper; incrementality is more accurate but requires volume and time.

Should we use Shopify's native attribution or a third - party tool?

Shopify's native attribution is single - device and last - touch only. It works for simple, fast - conversion DTC brands (e.g., impulse purchases, short customer journeys). If your customers use multiple devices, have longer journeys, or you need first - touch or multi - touch models, use Google Analytics 4 or a dedicated attribution platform. The trade - off is setup time and cost.

How do we handle attribution for repeat customers?

Repeat customers often have longer, more complex journeys. Decide upfront: do you attribute repeat purchases to the same model as first purchases, or separately? Most brands use the same model but segment the analysis (first purchase vs. repeat). This reveals whether channels are better at acquisition or retention.

What's a realistic attribution accuracy target?

Expect 80 - 90% accuracy on attributed revenue (the rest is untagged or bot traffic). Expect 10 - 20% variance between platform attribution and your single source of truth. If you're hitting > 90% accuracy, either your model is too simple or you're missing real complexity. Aim for consistency and auditability over false precision.

Want this on your account?

Thirty minutes. Bring the number that keeps you up.

Book a demo