Compare first-touch, last-touch, linear, time-decay and position-based marketing attribution models on one checked customer journey. A business owner's guide to what can be measured from CRM and ad exports, how credits reconcile to revenue, and why attribution does not prove incremental return.
Marketing attribution models divide credit for an observed outcome among the interactions recorded before it. Multi-touch attribution shares the credit; first-touch and last-touch assign it to one point in the journey. For a business owner, the useful question is which view the evidence can support before it influences the budget.
Attribution is an allocation convention. A channel receiving more credit under a model has not thereby been shown to cause more sales. Keep the marketing ROI and ROAS calculation separate from the choice of credit rule, and label an ROI built from those credits as attribution-based.
For the underlying definition of attribution as assigning credit across a path, see the attribution overview, checked October 17, 2026. This guide compares conceptual rules rather than the currently available models in any particular product.
| Model | Credit rule | Business question it can help describe | Main sensitivity |
|---|---|---|---|
| First-touch | All credit to the first eligible recorded interaction | Where do the observed journeys begin? | Missing earlier interactions can change the apparent origin |
| Last-touch | All credit to the last eligible recorded interaction | What interaction precedes the observed outcome? | Late interactions receive all credit even if earlier work mattered |
| Linear | Equal credit to every eligible interaction | What does the journey look like with equal treatment of its touches? | Repeated or low-value logged touches receive equal shares |
| Time-decay | More credit to interactions nearer the outcome | What changes if recent interactions carry more weight? | The chosen decay rate and outcome date affect the split |
| Position-based | A chosen share to the first and last touches; the remainder to the middle | What changes if origin and completion carry extra weight? | The endpoint shares are a policy choice |
These are conceptual allocation rules, independent of which models an advertising or analytics platform currently offers. Define whether repeated interactions count individually or are consolidated, and whether the unit is a person, account, opportunity or purchase. Do not mix units within the same model run.
A synthetic B2B journey ends in $120,000 of net invoiced revenue. The eligible recorded interactions are paid search 21 days before the outcome, a webinar 14 days before, a sales meeting 7 days before, and an email interaction on the outcome date, before the outcome. All four are linked to this outcome under the approved account/opportunity mapping. There are no other eligible touches in this example.
Each row below is an alternative allocation of the same $120,000. Do not add the model rows together.
| Attribution model | Paid search, day −21 | Webinar, day −14 | Sales meeting, day −7 | Email, day 0 | Total credited revenue |
|---|---|---|---|---|---|
| First-touch | $120,000 | $0 | $0 | $0 | $120,000 |
| Last-touch | $0 | $0 | $0 | $120,000 | $120,000 |
| Linear | $30,000 | $30,000 | $30,000 | $30,000 | $120,000 |
| Time-decay, 7-day half-life | $8,000 | $16,000 | $32,000 | $64,000 | $120,000 |
| Position-based, 40% / 20% / 40% | $48,000 | $12,000 | $12,000 | $48,000 | $120,000 |
For time-decay, the relative weights are 1/8, 1/4, 1/2 and 1. Dividing each by their total, 15/8, produces shares of 1/15, 2/15, 4/15 and 8/15. Multiplying by $120,000 gives the four exact dollar credits shown above. Seven days is an illustrative half-life, not a recommended standard.
Time-decay weight = 2 raised to the power of (−days before outcome ÷ half-life in days)
Touch credit = outcome amount × touch weight ÷ sum of eligible touch weights
Reconciliation per outcome: sum of touch credits = outcome amount
The position-based example assigns 40% to the first interaction and 40% to the last. The remaining 20% is split equally between the two middle interactions, so each gets 10%. Write separate rules for one-touch and two-touch journeys; the four-touch convention does not settle those cases.
This table exposes the decision's sensitivity. Paid search receives the entire amount under first-touch and none under last-touch. That difference comes entirely from the credit rule; the business result has not changed.
Start with one outcome register: an outcome ID, account ID, outcome date, amount and status. State whether the amount represents an opportunity booking or realized net revenue. For realized revenue, retain the invoice mapping and reconcile to finance; an opportunity value must not silently stand in for an invoice.
Build a dated interaction register with an interaction ID, the usable account or opportunity key, channel, campaign and interaction type. Set the lookback window and eligibility rules before inspecting channel winners. A sent email and an engaged interaction are different events; specify what counts. Use account identifiers rather than personal details when account-level measurement answers the question.
Ad spend by campaign and day can support a spend comparison. It cannot identify which customers saw or clicked an ad unless an approved, reliable mapping is present. Do not manufacture a journey by joining every account in a region to every campaign run there.
Apply the selected rule to the eligible matched journeys, retain the credited rows, and compare channel totals under a second plausible rule. Show unmatched outcomes as unallocated, with their amount visible. Within each model, credited revenue plus unallocated revenue must equal the selected outcome register. The underlying outcome still counts once.
The revenue operations metrics guide owns funnel conversions and handoffs. Keep those stage measures beside attribution: allocating revenue to a webinar is a different question from whether webinar leads reached a meeting or closed.
A detailed split implies a complete journey. Offline conversations, untracked visits and earlier contacts can be absent. Describe the result as credit across recorded eligible interactions, and report coverage.
One account has overlapping opportunities. An account join can attach the same touch to several deals. Approve the mapping and treatment of overlapping journeys before allocating amounts.
A duplicated event earns more credit. Duplicate activity records can distort linear and time-decay models. Deduplicate by the agreed event key and show repeated legitimate interactions separately.
The model changes when the ranking disappoints. Keep the rule, version and window on the report. A sensitivity comparison is useful; quietly replacing the official rule destroys comparability.
Every platform claims the whole outcome. Platform reports can use different windows and eligibility rules. Reconcile the combined view to one outcome register rather than summing their headline revenue.
Credit becomes causal uplift. A rule allocates the observed revenue even if some would have arrived without any marketing. An incremental budget claim needs additional evidence; see association and cause.
Agree the journey scope and credit rules before asking for an explanation. Covirage works from business exports, applies agreed definitions and uses its tools to compute every figure; the external AI model explains the resulting figures and limits. A business team can begin with the files it already holds, without a data-team project.
See sales insights for the file-based workflow. Related: marketing ROI vs ROAS, revenue operations metrics and association and cause.
Multi-touch attribution divides credit for an observed outcome across more than one eligible interaction. Linear, time-decay and position-based attribution are rule-based examples. The result depends on the recorded journey, eligibility window and allocation rule.
First-touch assigns all credit to the first eligible recorded interaction. Last-touch assigns it to the last eligible interaction before the outcome. They describe the beginning and end of the observed journey and can produce very different channel totals from the same outcomes.
There is no single best model. Pick a rule for the decision, document the eligible touches and observation window, and compare how the channel ranking changes under another plausible rule. Sparse data can support a simple transparent model more honestly than a detailed split.
Yes, when they contain a usable outcome key, dated eligible interactions and a reliable mapping between them. Ad exports often contain only channel totals; without a valid join to the outcome, those totals cannot establish a customer journey. Preserve unknown and unmatched outcomes.