# Multi-touch vs single-touch attribution

> The difference between single-touch (first-click / last-click) and multi-touch (linear, time-decay) models, and when each framing tells you something useful.

Source: https://help.tryordinary.com/concepts/multi-touch-vs-single-touch

---

Attribution models fall into two families:

- **Single-touch** — all credit to one touchpoint (either the first
  or the last).
- **Multi-touch** — credit split across multiple touchpoints.

This article explains when to reach for each.

## Single-touch models

**First-click** and **last-click** are both single-touch. They're
simple, intuitive, and famously wrong in opposite ways.

- **First-click** over-credits the channel that opens the door.
- **Last-click** over-credits the channel that happens to be the
  last URL the customer clicked before buying.

### When single-touch is fine

- You spend 90%+ of your budget on one channel (e.g. all Meta). No
  multi-channel story to tell.
- Your typical customer has 1-2 touches before converting. Short
  funnels don't need fancy allocation.
- You need a simple, explainable number for a stakeholder who doesn't
  want to hear about half-credits.

## Multi-touch models

**Linear** splits credit evenly across every touch. **Time-decay**
weights recent touches higher.

### When multi-touch is worth it

- You run **multiple channels** (Meta + Google + email + influencer)
  and want to see how they stack.
- Customer journeys are **long** — your typical buyer touches 4-6
  UTMs before converting, often across weeks.
- You're trying to **discover undervalued channels**. A channel that
  rarely appears as first or last click can still show meaningful
  linear-attributed revenue if it shows up mid-funnel frequently.

## Worked comparison

Pretend a customer had this journey:

1. Podcast sponsorship link (utm_medium=podcast) — day -14
2. Meta retargeting ad (utm_source=facebook, medium=paid) — day -3
3. Email with discount (utm_source=email) — day 0 (ordered $100)

Single-touch:

- First-click: **Podcast $100**, Meta $0, Email $0
- Last-click: Podcast $0, Meta $0, **Email $100**

Multi-touch:

- Linear: Podcast $33, Meta $33, Email $34
- Time-decay: Podcast $10, Meta $25, Email $65

The podcast gets zero credit in last-click, 100% in first-click, and
a meaningful share in linear. If you're deciding whether to renew
the podcast sponsorship, single-touch gives you a badly distorted
answer; multi-touch at least acknowledges the channel mattered.

## Which should I use for day-to-day reporting?

No single answer. Most teams we see settle into one of these patterns:

- **Last-click for reporting, first-click for a sanity check.** Pick
  last-click as your canonical model (it matches what most ad
  platforms report), and occasionally switch to first-click when
  evaluating awareness channels.
- **Linear all the time.** Treats every channel fairly. Good for
  teams who don't want to argue about model choice — pick one and
  move on.
- **Time-decay for performance, first-click for brand.** Time-decay
  rewards channels that drove urgency; first-click rewards channels
  that introduced the customer. Report on both.

Ordinary lets you switch models per-report, so you can run the same
report under different models to see the deltas.

## The model is not ground truth

Important caveat: **no attribution model is correct**. They're
heuristics for dividing up credit among channels that all contributed
somehow. The "truth" — did the customer buy because of channel X? —
is unknowable without randomized experiments.

Use attribution models for relative channel comparisons and directional
planning. Don't treat them as proof that channel X generated exactly
$Y — that's not what they measure.

For causal claims, run an experiment — Ordinary's built-in A/B testing
(variants with a randomized control) measures the real lift attribution
can't. See [A/B testing and experiments](https://help.tryordinary.com/features/experiments).

## Related articles

- [Attribution models explained](https://help.tryordinary.com/concepts/attribution-models) — per-model
  definitions and examples.
- [Attribution reports](https://help.tryordinary.com/features/attribution) — where you pick
  the model.
- [Why channel and campaign totals don't add up](https://help.tryordinary.com/concepts/channel-vs-campaign-attribution)
  — the one source of confusion that surprises every operator the
  first time they see it under linear or time-decay.