DTCUP Attribution Models: How to Choose and Compare Channel Credit
Choose among eight DTCUP attribution models, understand how each allocates order revenue, and compare channels with consistent dates and custom categories.
Attribution answers a specific question: how should recorded marketing interactions share credit for an order? The answer changes with the model. Choosing a different model does not change what a customer paid, and it does not reveal a complete history when tracking evidence is missing.
This guide explains the eight models available in DTCUP, when each is useful, and how to compare them without mistaking attributed revenue for incremental revenue.
Choose a model for the question you are asking
Start with a decision, rather than looking for a model that makes a channel perform best. First Click helps examine introductions. Last Click and Last non-direct help examine closing interactions. Linear models help examine participation. Position based and Time decay apply explicit assumptions about where or when an interaction matters.
Consider a fictional $120 order with four qualifying touches: Paid Search, another Paid Search visit, Email, then Direct. All four occur inside the lookback window. The examples below allocate the same $120; they are illustrations, not DTCUP customer results.
First Click: credit the recorded introduction
DTCUP assigns all credit to the first qualifying acquisition touch recorded before the purchase, within the selected lookback window. In the example, Paid Search receives $120.
Best for: investigating which sources introduce customers in observed journeys. Use when: comparing discovery-oriented campaigns with a consistent time window. Limit: later interactions receive no credit. A first recorded click is not necessarily the first time someone encountered your brand; an earlier interaction may have happened outside the window or without usable tracking.
Last Click: credit the closing touch
All credit goes to the last qualifying touch before the purchase, including Direct. In the example, Direct receives $120.
Best for: understanding the final recorded acquisition step before a sale. Use when: inspecting closing journeys and the channels customers return through. Limit: this model gives no credit to the Paid Search and Email touches that came earlier. A strong Last Click result does not establish that the last touch caused the purchase.
Last non-direct: keep the latest identifiable marketing source
This model gives credit to the last qualifying non-direct touch. If the journey contains only Direct, DTCUP falls back to Direct. In the example, Email receives $120 because the subsequent Direct visit does not displace it.
Best for: reviewing marketing sources behind returning visits. Use when: you want to distinguish the latest recorded marketing interaction from a final direct return. Limit: it still concentrates all credit on one touch and can underrepresent the earlier acquisition channel.
Linear All: equal shares for distinct channels
DTCUP divides credit equally among the distinct observed channels in the qualifying journey. Each channel then divides its share among its touches. Paid Search, Email and Direct therefore receive $40 each in the example. The two Paid Search touches share that channel’s $40.
Best for: comparing participation without allowing repeated visits from one channel to increase that channel’s share. Use when: you want a channel-level view of multi-touch journeys. Limit: equal channel importance is an analytical assumption. It is not an estimate of the marginal effect of spending more money on each channel.
Linear Paid: equal shares among qualifying paid channels
Only qualifying paid channels participate in this allocation. Distinct paid channels receive equal shares, and each channel splits its share among its qualifying touches. Paid Search receives the full $120 in the example because it is the only paid channel.
Best for: examining how recorded paid channels share participation. Use when: comparing paid acquisition paths separately from organic and owned interactions. Limit: Email and Direct receive no credit here. Orders without a qualifying paid touch stay Unattributed; DTCUP does not invent a paid source for them.
Linear per touch: equal shares for recorded interactions
The legacy per-touch model divides credit equally among qualifying touches. Each of the four example touches receives $30: Paid Search totals $60, Email $30 and Direct $30.
Best for: examining interaction-level participation or comparing with earlier reporting conventions. Use when: the number of recorded interactions is part of your analysis. Limit: more touches mean more credit for a channel. Tracking frequency can therefore affect the result. This is different from Linear All, which first allocates by distinct channel.
Position based: emphasize the first and last touches
With three or more qualifying touches, the first receives 40%, the last 40%, and the middle touches share 20%. With fewer than three touches, credit is divided equally. For the four-touch example, the first Paid Search touch receives $48, the second $12, Email $12 and Direct $48. Paid Search totals $60.
Best for: an analysis that values introduction and closing interactions together. Use when: comparing early and late participation while retaining some middle-journey credit. Limit: 40/40/20 is a chosen allocation rule, not a learned causal relationship. Short journeys follow the equal-share rule rather than leaving credit unallocated.
Time decay: give more weight to recent touches
Credit decreases exponentially with the age of each qualifying touch and is normalized across the order’s eligible touches. DTCUP uses the configured half-life, with seven days as the default. With that default, a touch seven days older has half the unnormalized weight of an otherwise equivalent newer touch.
Best for: exploring whether recent engagement deserves more emphasis in your planning. Use when: comparing journeys where recency is a useful explicit assumption. Limit: dates matter, so the order of the example touches alone is insufficient to calculate their shares. Recency weighting does not prove that recent ads produced incremental sales.
Compare models with a consistent reporting setup
- Select the same website, purchase-date range, reporting timezone and currency.
- Keep the lookback window and journey context consistent. Purchase dates identify the orders in scope; the lookback window identifies eligible interactions before those purchases.
- Change one model at a time. DTCUP updates the current results without requiring a saved report.
- Inspect fractional purchase credit alongside revenue. An order can contribute to multiple channel rows, so distinct credited-order counts across rows are not necessarily additive.
- Use the comparison-period control to inspect change over time, while accounting for promotions, seasonality, product availability and data coverage.
- Open channel and campaign details to check the underlying evidence before changing a budget.
Read the column definitions. Attributed net revenue, gross revenue, fractional purchases and platform-reported conversions answer different questions. AOV based on fractional purchase credit will not always equal revenue divided by the displayed distinct-order count. Do not add platform-reported conversions to DTCUP-attributed orders.
Organize the same results with custom categories
Custom categories let you organize observed activity around business labels, such as a funnel-stage category containing Prospecting and Retargeting. Each subcategory can match channel, source, medium or campaign text using Equals, Contains or Starts with. Choose whether all or any of its rules must match.
Subcategories are checked from top to bottom and the first match wins. Rules are case-insensitive. Unmatched activity remains Unassigned; order revenue without linked qualifying touches remains Unattributed. Put narrower rules before broader ones and check campaign naming consistency before relying on the grouping.
Category rules are saved workspace configuration. They do not change the original touch evidence or the attribution model, and browsing a category does not save an analytical snapshot. Only workspace owners and administrators can edit these rules.
What attribution cannot tell you on its own
Missing consent, unlinked identities, incomplete history and unavailable provider data limit what can be observed. Unattributed revenue is an evidence gap, not proof that marketing played no part. Changing a model redistributes eligible credit; it cannot recover an interaction that was never collected.
DTCUP’s current model selector does not provide survey-blended attribution, deterministic ad-view attribution, or a proprietary causal model. Aggregate impressions are not evidence that a particular buyer saw an ad. These capabilities require additional verified data and a defined methodology.
For budget decisions, combine attribution with suitable experiments and business context. A model can help form a hypothesis about channel contribution; an experiment can help test whether changing spend changes outcomes.
Explore your own attribution
Open DTCUP Attribution to compare supported models, or manage custom categories in your workspace. Account access is required.
Further reading: Triple Whale’s overview of attribution models provides another product’s perspective. Model names and implementation details vary across platforms; the rules described above refer specifically to DTCUP.
Try a related toolDTCUP Attribution Models: How to Choose and Compare Channel Credit ↗
Help with implementationDTCUP Attribution Models: How to Choose and Compare Channel Credit ↗
Still have a question?Ask the DTCUP Community ↗


Leave a Reply
You must be logged in to post a comment.