# What is marketing attribution? Detailed guide

> What is marketing attribution, in plain English: the five models, worked examples on one journey, what attribution cannot see, and how a small team starts.

Canonical page: https://statsy.co/blog/what-is-marketing-attribution  
Published: 18 September 2026  
Author: Manish Sharma, Founder, Statsy  
Category: Attribution  
Tags: attribution, analytics-basics

**In short:** Marketing attribution is the practice of deciding which marketing touches get credit for a conversion, and how much credit each one gets. There is no single correct split, so attribution is a set of conventions rather than a measurement. Five models are common: last touch, first touch, linear, time decay and position based. Each gives a different answer from the same journey.

## What is marketing attribution?

Marketing attribution is the practice of deciding which marketing touches get credit for a conversion, and how much credit each one gets. That is the whole idea. When someone reads your newsletter, clicks a search result a week later, and buys on a third visit, attribution is the rule you use to split the sale between those three moments.

The reason people ask what is marketing attribution is that the honest answer is uncomfortable: there is no single correct split. A sale is one event with several causes. Attribution is a set of agreed conventions for dividing one outcome among many inputs, the same way a company agrees on a revenue recognition policy. Different rules give different answers, and none of them is a measurement of truth.

That does not make attribution useless. It makes it a decision tool. You pick a rule, apply it consistently, and compare channels under the same rule. What you learn is relative: which channels look better or worse than they did last quarter, and where the next thousand dollars should go.

## Why does marketing attribution exist at all?

Attribution exists because channels report their own results, and every channel claims the same sale.

Run a Google Ads campaign and a Meta campaign in the same month, add an affiliate, and each platform will count conversions it believes it caused. Add those numbers up and you will often find you sold more than you actually did. Each platform sees only its own touches and credits itself for the ones it saw. Nobody is lying. They simply have partial views.

The second reason is that buying is rarely one click. In software and ecommerce, a first visit and a purchase are often weeks apart. If you only look at the last click, every channel that creates demand early looks worthless, and you will cut it.

The third reason is budget. Marketing attribution turns "we spent $8,000 across five channels" into "these two channels produced most of the revenue".

## What counts as a touch?

A touch is any recorded interaction where a visitor arrived at your site from an identifiable source. In practice that means a session with a referrer, a UTM-tagged link, or a paid click ID.

You get this data in three ways:

- **UTM parameters** on links you control. The five recognized parameters are `utm_source`, `utm_medium`, `utm_campaign`, `utm_term` and `utm_content` ([Google's reference](https://support.google.com/analytics/answer/10917952)). These are the only source data you fully control, so tag every link you publish.
- **The HTTP referrer**, sent by the browser when someone follows a link. Modern browsers default to a `strict-origin-when-cross-origin` referrer policy, so you usually receive the origin (`https://news.ycombinator.com/`) and not the full URL of the page that linked to you ([MDN](https://developer.mozilla.org/en-US/docs/Web/HTTP/Reference/Headers/Referrer-Policy)).
- **Click identifiers** such as `gclid` and similar parameters appended by ad platforms.

Direct traffic is what is left when none of those exist. It is a bucket of unknowns, not a channel. Treat a rising "direct" number as a tagging problem first and a brand-strength story second.

## What are the five attribution models?

The five common models are last touch, first touch, linear, time decay and position based. Each takes the same list of touches and splits the credit differently.

To keep the comparison honest, every example below uses the same journey and the same sale:

- **Day 1**: a visitor finds you through organic search.
- **Day 5**: they click a link in your email newsletter.
- **Day 12**: they type your domain directly and buy. The order is $600.

### Last touch

Last touch gives all the credit to the final touch before the sale. Here, direct gets $600 and the other two get nothing. It is the default in most tools because it is simple and unarguable: this is the click that preceded the money. It is also the model that most reliably undervalues everything that creates demand. Shopify's marketing reports default to a variant of this, last non-direct click, which skips direct visits and credits the last identifiable channel instead ([Shopify Help Center](https://help.shopify.com/en/manual/promoting-marketing/analyze-marketing/marketing-performance)).

### First touch

First touch gives all the credit to the touch that introduced the visitor. Organic search gets $600. This model answers a different question: where does demand come from? It is useful for judging top-of-funnel work such as SEO, podcasts and PR, and it is bad at judging anything designed to close a sale, because closing touches get zero.

### Linear

Linear splits the credit equally across every touch. Each of the three gets $200. Linear is the least opinionated multi-touch model. Its weakness is that it treats a ten-second bounce from a Reddit thread as equal to a thirty-minute pricing-page session, which is rarely how the sale actually happened.

### Time decay

Time decay gives more credit to touches closer to the sale, with credit falling off over time. With a seven-day half-life, a touch eleven days before the order carries roughly a third of the weight of a touch on the order day. Running that over our journey: organic search gets about $110, email about $163, direct about $327. Time decay suits long consideration cycles where recent activity genuinely matters more, and it is harsh on the first touch in a ninety-day journey.

### Position based

Position based, often called U-shaped, gives fixed weights to the ends and splits the rest across the middle. The common split is 40% first, 40% last, 20% shared among the rest: $240 to organic search, $240 to direct, $120 to email. It encodes a real belief, that discovery and the closing click matter most, and it is a reasonable default when you have no strong opinion.

### The same sale, five ways

| Model | Organic search (day 1) | Email (day 5) | Direct (day 12) |
| --- | --- | --- | --- |
| Last touch | $0 | $0 | $600 |
| First touch | $600 | $0 | $0 |
| Linear | $200 | $200 | $200 |
| Time decay | $110 | $163 | $327 |
| Position based | $240 | $120 | $240 |

One sale. Five defensible answers. Organic search is worth either nothing or the entire order depending on a dropdown.

## Single touch or multi touch: which should you use?

Use single touch when you need one number to act on quickly, and multi touch when you are deciding a budget across channels.

Single-touch models, first and last, are easy to explain and impossible to misread. They are fine when a channel mix is simple, when most journeys really are one visit, or when you need a fast read on a campaign that ran last week.

Multi-touch models, linear, time decay and position based, are the ones to use when you are moving money between channels. They are the only models that show a channel which reliably starts journeys it never finishes. In most small businesses that channel is content or organic search, and last touch quietly writes it off.

The practical approach is to run two models side by side and look at the gap. A channel that looks strong under first touch and weak under last touch is doing discovery work. A channel strong under last touch and weak under first touch is closing, not creating. Statsy supports all five models on the same data, so switching is a dropdown rather than a re-implementation.

## What can marketing attribution not see?

Attribution misses anything that happens outside a browser session it can tag. Three gaps matter most.

**Offline and word of mouth.** A recommendation in a Slack group, a conference conversation, a colleague forwarding a screenshot. These produce direct visits with no referrer. The sale is real and the cause is invisible.

**Dark social.** Links shared in messaging apps, private communities and email clients arrive with no referrer at all. Even public platforms obscure the source: X wraps links in `t.co` and LinkedIn in `lnkd.in`, so the referrer carries the wrapper domain, not the post that drove the click. You can tell social traffic came from LinkedIn; you generally cannot tell which post.

**Cross-device journeys.** Someone discovers you on a phone and buys on a laptop. Without a login to stitch the identities together, that is two separate visitors and the phone touch is lost. Privacy rules make this harder on purpose, and rightly so. Statsy applies cookieless mode automatically to visitors from the EU, EEA, UK and Switzerland and to browsers sending Global Privacy Control, so some journeys in those regions are measured as separate sessions rather than one stitched path.

The right response is not to buy a bigger tool. It is to size the gap: watch what share of converting journeys have no identifiable first touch, and treat that share as the error bar on every attribution number you report.

## How long should the lookback window be?

The lookback window is how far back attribution looks for touches before a conversion. Set it to slightly longer than your real sales cycle.

If your median time from first visit to purchase is three weeks, a seven-day window will discard most of the journey and hand almost everything to last touch by default. If you sell a $12,000 annual contract with a four-month evaluation, a 30-day window is worse than useless. Statsy's window runs from 1 to 365 days with a default of 90, which covers most software and ecommerce cycles.

Two things to know. First, changing the window changes history, so decide it once and note the date if you change it. Second, a longer window increases the number of touches per journey, which mechanically dilutes each touch under linear and position based while barely moving last touch. If a channel's numbers shift after a window change, check the window before you check the channel. Google Analytics 4 exposes its own lookback settings separately from the reporting model ([GA4 attribution settings](https://support.google.com/analytics/answer/10597962)), so check both before comparing two tools.

## How does a small team start with attribution?

Start with tagging discipline, not with a model. The model is a dropdown; the data quality underneath it is the actual work.

1. **Tag every link you control.** Newsletters, social posts, partner placements, sponsorships, QR codes. Use a fixed convention for `utm_source` and `utm_medium` and write it down, because `Newsletter`, `newsletter` and `email-newsletter` become three channels in every report you will ever run.
2. **Define the conversion that matters.** For most businesses it is a paid order, not a signup. If you attribute signups you will optimize for signups, and cheap signups are easy to buy.
3. **Set the lookback window to match your sales cycle**, then leave it alone.
4. **Pick two models and keep both.** Last touch and position based is a good pair: one tells you what closed, the other tells you what contributed.
5. **Review monthly, not daily.** Attribution data is noisy at small volumes. A week of data from thirty orders will send you chasing ghosts.

Google Analytics 4 is a reasonable free starting point, but note that it retired first click, linear, time decay and position based as reporting models in 2023, leaving data-driven and last click ([Google's attribution documentation](https://support.google.com/analytics/answer/10596866)). If you specifically want to compare the five classic models, you will need a tool that still offers them.

## What should you do next?

1. Open your analytics and find the share of conversions attributed to "direct" or "(none)". Anything above a quarter usually means untagged links, not brand demand. Fix the tagging first.
2. Write your UTM convention into a shared doc, one line per parameter, and use it for every link you publish from now on.
3. Set your lookback window to cover your real sales cycle, measured from first visit to first payment, not guessed.
4. Run the same month through two models, one single touch and one multi touch, and list the channels where the two disagree most. Those are the channels you currently misjudge.
5. If revenue rather than signups is the number you care about, connect attribution to your payment provider so the credit is attached to money received. Statsy does this with signed webhooks from Stripe, Lemon Squeezy, Paddle, Polar and Razorpay; the [marketing attribution software](/marketing-attribution-software) page covers how the models are applied.

## Questions people ask

### What is marketing attribution in simple terms?

It is a rule for splitting credit for a sale across the marketing touches that came before it. A visitor arrives from search, returns from email, then buys. Attribution decides how much of that sale each touch earns.

### Which attribution model should I use?

Run two: one single touch and one multi touch. Last touch shows what closed the sale, position based shows what contributed. The channels where the two models disagree are the ones your current reporting misprices.

### What is the difference between single-touch and multi-touch attribution?

Single-touch gives all credit to one touch, either the first or the last. Multi-touch splits credit across every touch in the journey. Use single-touch for quick campaign reads, multi-touch for budget decisions across channels.

### Why is so much of my traffic showing as direct?

Direct is what remains when there is no referrer, UTM or click ID. It usually means untagged links, messaging apps and email clients rather than brand demand. Audit your own links before crediting brand awareness.

### How long should an attribution lookback window be?

Slightly longer than your real sales cycle, measured from first visit to first payment. Too short and everything collapses to last touch. Statsy allows 1 to 365 days and defaults to 90, which covers most software and ecommerce cycles.

---

Statsy (https://statsy.co) · More posts: https://statsy.co/blog