Definition
Multi-touch attribution (MTA) is the practice of distributing conversion credit across every marketing touchpoint a customer interacted with on their path to purchase, rather than awarding 100% of the credit to a single touchpoint. If a customer reads a review, clicks a coupon site, and finally converts through a paid social ad, MTA credits all three — not just the ad they clicked last.
MTA exists because single-touch models (usually last-click) systematically misrepresent how modern customer journeys work. Most purchases involve 3–7 touches over days or weeks. Last-click attribution makes the closer look like a hero and hides every partner that drove awareness, consideration, and intent upstream.
Why it matters for affiliate programs
Last-click is still the payout truth in almost every affiliate program — you pay one partner per conversion, and it has to be decidable at conversion time. But using last-click for _understanding_ your program is a different problem. Multi-touch attribution gives you the full picture without changing how commissions are paid.
Consider a brand running both an SEO review site and a coupon affiliate. Under last-click:
- Coupon affiliate looks incredible (they're always the last touch right before checkout)
- SEO affiliate looks replaceable (they almost never close, but they introduce customers a week earlier)
Under multi-touch, both partners get proportional credit, and you can see that pausing the SEO affiliate would tank your conversion volume over the next 30 days because you'd be starving the top of the funnel.
The three main models
Linear attribution
Every touchpoint in the path gets an equal share of the credit. If a customer touched 4 affiliates, each gets 25%. Simple, easy to explain, and ignores recency + position.
Use linear when you care about "who was involved at all?" — it's the most neutral view and a good default for cross-partner comparison.
Position-based attribution (40/40/20)
The first touchpoint gets 40%, the last touchpoint gets 40%, and any middle touchpoints split the remaining 20% evenly.
Position-based is Google Analytics' default because it reflects a common product narrative: the first touch creates awareness, the last touch closes the deal, and middle touches play supporting roles. Good when those two moments matter more than what happened in between.
Time-decay attribution
Each touchpoint's weight decays exponentially with age. A typical half-life is 7 days — a touch from 7 days before conversion is worth half of a touch from the day of, and a touch from 14 days before is worth a quarter.
Mathematically: weight = 2^(−age_days / half_life_days).
Time-decay is useful for short-cycle products (subscriptions, impulse purchases) where recency matters more than order. For long-cycle B2B purchases, first-touch and linear are usually better signals.
Compared to Shapley and Markov models
Linear, position-based, and time-decay are all closed-form — you don't need to train anything, and the math fits on a napkin. They're also limited: they treat every touchpoint as equivalent evidence, which isn't true in the real world.
Shapley-value attribution borrows from cooperative game theory: for each partner, compute the marginal contribution they add to every possible subset of the conversion path, then average. Shapley respects how partners interact — a partner who's only useful when paired with another gets credit only when that pairing happens.
Markov-chain attribution models the customer journey as state transitions between partners and computes each partner's "removal effect" — the drop in conversion probability if you removed that partner entirely. This is the most causally-grounded classical MTA model.
Both require a trained path-to-conversion probability model fit on both converted and non-converted journeys. That's a full ML pipeline, not a spreadsheet.
The limitations of MTA
MTA is better than last-click for understanding, but it's still correlational. A partner can show up in every converting path and not actually _drive_ any of those conversions — they might be on the path because the customer would have converted anyway.
For the causal question, you need incrementality testing. Multi-touch attribution answers "who touched this conversion?" Incrementality answers "did touching it matter?"
Use MTA for allocation across partners who definitely contribute. Use incrementality to catch partners whose conversions were going to happen anyway.
How Trcker implements MTA
Trcker runs all three closed-form models nightly against the rolling 30-day window. For every approved conversion, we stitch together every click the same visitor made in the prior 30 days, dedupe by partner, and apply each model to the path. Results materialize in the attribution dashboard alongside the last-click baseline — so brand operators see how each model reallocates credit versus their payment truth.
Shapley and Markov are roadmap items — they require the probability-model infrastructure referenced above.
Related concepts
- Attribution — the general problem of assigning credit
- Incrementality — the causal alternative to attribution
- Cookie window — how long after a click a conversion can be credited
- EPC — earnings per click, the core affiliate economics metric