How does multi-touch attribution split credit?

Linear: credit per touch = 1 / number of touches. Position-based: 40% first, 40% last, 20% split across the middle; one touch gets 100%, and two touches get 50% each. Time decay: credit halves for each half-life before the conversion, with a default half-life of 7 days in Manalens.

Sources: Manalens attribution models documentation.

What does a worked example look like?

Example: one $20 purchase follows three touches, a creator video, a search and a Direct visit. The table shows the credit for each model.

Credit for one $20 purchase with three touches (example numbers)
ModelCreator videoSearchDirect
Linear$6.67$6.67$6.67
Position-based (40/20/40)$8.00$4.00$8.00
Last non-direct (single touch)$0$20.00$0

Sources: worked example; Manalens attribution models documentation.

Why does multi-touch attribution matter for indie PC and web games?

Players often meet a game several times before they buy: a showcase, a creator video, a newsletter, a sale email. A multi-touch model shows which channels take part, not only which one closed the sale.

It still sees only observed touches. Valve says it cannot connect a click in one browser to a conversion in the Steam client, and a player who saw an ad and did not click leaves no touch. So no model can share out credit for touches it never saw.

Sources: Steamworks UTM Analytics documentation; Manalens attribution models documentation.

How do I use multi-touch attribution in Manalens?

Choose linear, position_based or time_decay in the Attribution report, Ask AI or the get_attribution MCP tool. Set the time decay half-life from 1 to 30 days. The models do not use ad impressions, ad spend, probabilistic identity matching or causal uplift.

Sources: Manalens attribution models documentation.

Frequently asked questions

Which multi-touch model should I use for a game?

Use position-based to ask who opens and who closes, linear to see which channels take part most often, and time decay to see what worked close to the purchase. Compare them with last non-direct, the default, and keep any channel that is strong on one of them.

Does multi-touch attribution prove that a channel worked?

No. It shares out credit between touches that were observed before a conversion. It cannot tell you how many sales would have happened without a channel. For that, run an incrementality test, such as a baseline week or a country holdout.

How does time decay attribution give credit?

It gives more credit to touches close to the conversion. In Manalens, the credit halves for each half-life before the conversion. The default half-life is 7 days, and you can set it from 1 to 30 days. Use it to see what worked close to the purchase or wishlist.

Sources and implementation