Time-Decay Attribution
Time-decay attribution is a multi-touch model that gives more conversion credit to touchpoints that occur closer in time to the conversion.
Key takeaways
- Each touch is weighted by its age, then all weights are normalised to one conversion.
- The half-life sets how fast credit fades: one half-life halves a touchpoint's weight.
- Set the half-life against typical cycle length, never against a calendar month.
- Same-day paths collapse toward an equal split because every touch has similar age.
- Confirmation and reminder messages score highly despite having persuaded nobody.
In depth
Every touchpoint receives a weight computed from its age at the moment of conversion, usually through an exponential curve defined by a half-life. A touch that happened one half-life before the conversion is worth half as much as one at the conversion itself, a touch two half-lives earlier is worth a quarter, and so on down the path. Those raw weights are then divided by their own total, so the path still sums to exactly one conversion. Nothing but elapsed time enters the calculation.
The half-life is effectively the whole model. A seven-day half-life applied to a two-month cycle leaves the first month with almost no credit, while a thirty-day half-life on the same cycle produces something close to an equal split. Choose the value relative to the length of the journey rather than to a calendar convention. The trade-off is responsiveness against memory: short half-lives make the report react quickly to recent campaign changes but erase the contribution of anything that happened early.
Time-decay suits teams optimising the closing half of a long cycle, where the real question is which late-stage nudges shorten the gap to a decision. It also handles bursty journeys well, because a cluster of touches in the final week naturally outweighs a single visit from three months earlier. Where a scorecard quiz is used to re-engage a stalled lead rather than to acquire a new one, this model shows that role clearly, while first-touch reporting misses it entirely.
Recency is not causality. A confirmation email sent an hour before a purchase always scores well under time-decay even though it persuaded nobody, and any channel whose job is early education is penalised by construction. The model also behaves oddly on paths containing long gaps: a lead who researched in March and bought in September hands nearly all credit to September, hiding the fact that the decision was largely made six months before the money moved.
Example in practice
How to measure it
Report credited share by time bucket: touches in the final week, in the previous three weeks, and everything older. That distribution shows how much of the funnel your current half-life is effectively ignoring. Alongside it, track median days from first touch to conversion; when the cycle lengthens, the half-life needs revisiting or early touches will silently drop out of the picture.
For channel decisions, compare each channel's time-decay credit against its linear credit. Channels that gain are late-stage, channels that lose are early-stage, and the size of the shift measures where in the journey each one operates. Judge the late group on closing efficiency and the early group against origination goals, rather than holding both to the same cost target.
Common mistakes
Teams copy a seven-day half-life from an ecommerce example into a six-month enterprise cycle, then conclude that everything except the final sales email is worthless. Check the median time between first touch and conversion before setting the parameter, and re-run the report with two different half-lives. If the channel ranking changes between them, that ranking is a property of your chosen parameter rather than of your marketing.
The second failure is leaving automated messages inside the path. Order confirmations, meeting reminders and password resets fire immediately before or after the conversion and absorb the credit that a decay curve hands to the most recent touch. Exclude transactional and post-conversion events from the path definition explicitly. Otherwise the report slowly turns into a ranking of how close each system sits to the checkout button.
Frequently asked questions
How does the half-life work in time-decay attribution?
The half-life is the period over which a touchpoint's credit halves as it moves further from the conversion. A shorter half-life concentrates credit on the most recent touches, while a longer one spreads it more evenly.
What half-life should I use for time-decay attribution?
Anchor it to your median time from first touch to conversion, then pick a half-life somewhere around a quarter to a third of that span as a starting point. Shorter cycles need days, considered B2B cycles need weeks. Test two values and see whether your channel ranking survives both before committing the number to a dashboard.
How is time-decay attribution calculated?
Take the age of each touchpoint measured back from the conversion, convert it into a weight that halves once per half-life, then divide every weight by the sum of all weights on that path. The normalised values are the fractions of the conversion each touch receives. Roll those fractions up by channel to get credited conversions.
When is time-decay attribution the right model?
When the cycle is long enough for timing to carry information and the decision genuinely firms up near the end, which is common in considered B2B purchases. It is also the natural model when you are optimising nurture sequences or sales follow-up. Avoid it when your main question is which channels create new demand.
Does time-decay attribution undervalue brand and awareness campaigns?
Yes, structurally. Awareness touches sit furthest from the conversion, so the decay curve reduces them regardless of how much they contributed. Report those campaigns against first-touch or origination metrics instead, and reserve time-decay for the channels whose job is to move an already-interested buyer toward a decision.
What is the difference between time-decay and position-based attribution?
Time-decay weights by when a touch happened, so credit rises smoothly toward the conversion. Position-based weights by where a touch sits in the sequence, giving fixed shares to the first and last regardless of dates. On a path where the first touch happened yesterday, position-based still rewards it heavily while time-decay barely distinguishes it.
How does time-decay handle a long gap in the middle of a journey?
Badly, because it only sees age, not the shape of the gap. Everything before a dormant period is pushed toward zero even if that early research decided the purchase. When long gaps are common, add a rule that restarts the path after a defined period of inactivity, and report the two segments separately.