Multi-touch attribution is the family of attribution models that distribute conversion credit across more than one touchpoint, instead of assigning all of it to the last one (last-click) or the first one (first-click). The point is to reflect the reality that most B2B and considered-purchase conversions involve several interactions over days or weeks, and a model that ignores all but one of them will mis-cost the channels that fed the funnel.
In practice, multi-touch usually means one of three approaches. Position-based (or U-shaped) attribution gives the first and last touchpoints extra weight, with the rest spread between them — useful when discovery and closing are the most important moments. Time-decay weights recent interactions more heavily, which fits sales cycles where recent contact is what tips the decision. Data-driven attribution uses observed conversion patterns to assign credit empirically; it's the most accurate when you have enough volume, and the hardest to explain when you don't.
I use multi-touch reporting alongside last-click rather than instead of it. Last-click is what most ad platforms optimise against, so it stays useful for in-platform decisions. Multi-touch is what I use when deciding where the marketing budget goes — which channels deserve more, which are coasting on credit they didn't earn.
The failure mode of multi-touch is over-investment in the model itself. A position-based or time-decay model run on clean data tells you most of what you need; a custom Markov-chain model run on dirty data tells you nothing useful. The leverage is in good source-tracking on every campaign, not in picking the most sophisticated math.