Multi-touch attribution: connect and allocate a customer journey
Follow a three-touch purchase, compare fractional credit, and check identity, eligibility, and revenue reconciliation before trusting a multi-touch report.
By Mika Garcia · Published
In this article
Multi-touch attribution shares credit across eligible interactions
Multi-touch attribution assigns portions of a conversion to more than one recorded interaction. It can show the role of earlier visits that disappear from a last-touch-only view. The result depends on both the allocation rule and the set of interactions the system can actually observe.
A useful report starts with an inspectable journey: which person, which eligible touches, which payment, and which amount. A sophisticated weighting rule cannot recover a campaign parameter that was never captured or connect two people merely because their journeys look similar.
This guide focuses on building and checking that journey. For the narrower comparison between two single-touch rules, use the first-touch versus last-touch guide.
Allocate one $240 purchase without creating extra revenue
In a hypothetical journey, a buyer clicks a social ad on Monday, returns through an email on Wednesday, and clicks a search ad on Friday before paying $240. Assume all three interactions are eligible and linked to the same buyer.
A linear model gives each touch $80. First-touch would give all $240 to social; last-touch would give all $240 to search. These are alternate views of one transaction. Adding their totals would triple the revenue without adding a single sale.
When multiple touches belong to the same channel, decide whether you allocate per touch and then aggregate, or collapse the path first. With two social touches and one search touch, equal credit per interaction gives social two-thirds. Equal credit per distinct channel gives it one-half. Write down which rule your report uses.
One $240 payment, three equal shares
1. Monday
Social: $80
First eligible interaction.
2. Wednesday
Email: $80
Middle interaction.
3. Friday
Search: $80
Last eligible interaction before payment.
Define the eligible journey before choosing weights
The lookback window determines how far before conversion you inspect. A touch outside that window cannot receive credit, even if the analyst knows it happened. Consent, cross-device identity and capture failures can further reduce the recorded journey.
Document how direct visits, repeated visits and post-purchase events are treated. If direct traffic is excluded in one model and included in another, you are changing eligibility as well as allocation. That makes the difference harder to interpret.
Keep a small journey sample beside the dashboard. For each example, write the conversion time, included touches, excluded touches and reason for exclusion. This makes a surprising channel result something you can investigate rather than debate.
Check the identity handoff before the model
A browser visit and a billing webhook arrive through different systems. They need an appropriate shared identifier to belong to one customer journey. Repeated payments also need transaction identifiers so a retry does not become another sale.
DATALYR documents capturing campaign parameters at arrival and linking identified activity across browser and server events. Its conversion-delivery lookup is separate from the reporting model you use to distribute revenue. Do not assume that selecting a reporting model changes every destination’s matching rule.
Check one known journey end to end before judging aggregate allocation. The payment can exist while campaign attribution is missing; that is an identity or capture investigation, not proof the channel produced no value.
Separate capture from allocation
1. Capture
Was the touch recorded?
Inspect campaign parameters and timestamps.
2. Identity
Does it belong to this buyer?
Check the supported visitor-to-customer link.
3. Allocation
Which rule assigns credit?
Apply eligibility and weights to the observed path.
Reconcile assigned and unassigned revenue
Define a revenue basis before comparing channels: successful gross payments, an amount adjusted for refunds, or another clearly specified measure. Then check that assigned plus unassigned revenue reconciles to that same source total.
For example, if the $240 purchase receives a $60 refund and your chosen policy allocates the adjusted $180 back across the original three touches, each would receive $60. This is a possible reporting policy, not a claim about automatic behavior in every product. A different refund policy must be described just as explicitly.
Leave genuinely unmatched revenue visible. Moving it into a convenient channel to make the report look complete removes a valuable diagnostic signal.
Use model differences to choose an investigation
Compare models on the same dates, transaction set and revenue basis. If social gains credit under linear allocation while search loses it, inspect representative journeys. The difference tells you the rule values their recorded positions differently; it does not prove the incremental effect of cutting either channel.
Use the result to identify a next question: are earlier touches missing, is a campaign mostly closing existing demand, or does the buying cycle exceed the window? A controlled experiment may be needed for the causal budget question.
Keep the model name, window and exclusions next to the result. A number called attributed revenue is not sufficiently defined until the reader knows how it was assigned.