Attribution Models Explained: Last-Click to Multi-Touch

What a Marketing Attribution Model Actually Does

A marketing attribution model is a rule for splitting conversion credit across the touchpoints a customer passed through before converting. That's the entire job. The model doesn't discover new touchpoints, doesn't verify the ones it has, and doesn't know whether the journey it's scoring is complete. It takes an ordered list of recorded interactions and distributes a conversion's value across them according to a formula.

This matters because attribution debates usually focus on the formula — last-click versus linear versus time-decay — when the formula is the easy part. Every model in this guide can be computed in a spreadsheet. The hard part is the input: a journey timeline that actually contains all the touchpoints, connected to the same person, across sessions and devices. We'll cover the models first, then the data layer they all quietly depend on.

One journey to score five ways

Throughout this guide we'll use a single hypothetical journey. The numbers are illustrative — chosen for arithmetic clarity, not drawn from any benchmark:

• Day 1: Clicks a LinkedIn ad (li_fat_id captured) and reads a product page
• Day 8: Returns via organic search to a blog article
• Day 12: Clicks a paid brand-search ad on Google (gclid captured)
• Day 14: Types the URL directly and converts — a $1,000 purchase

Four touchpoints, one conversion, $1,000 of credit to distribute. Each model gives a different answer to the same question: who earned it?

The Five Classic Marketing Attribution Models

1. Last-click attribution

The rule: 100% of credit goes to the final touchpoint before conversion. In our journey, the Day-14 direct visit takes all $1,000.

What it rewards: closers — brand search, retargeting, direct visits, anything positioned at the end of the journey. It's simple, universally supported, and requires the least data: even single-session analytics can compute it.

What it punishes: everything that created the demand. The LinkedIn ad that introduced the buyer gets zero. Worse, "direct" often wins by default because people bookmark or retype URLs for their final visit — which tells you nothing about what marketing did. Many platforms compensate with a last non-direct click variant that skips direct visits; in our journey that reassigns the full $1,000 to the Day-12 brand-search click, which flatters paid search instead.

2. First-click attribution

The rule: 100% of credit goes to the first recorded touchpoint. The LinkedIn ad takes all $1,000.

What it rewards: discovery — the channels that introduce new prospects. Useful when your question is "where do customers first hear about us?" and for justifying top-of-funnel budget.

What it punishes: everything that converted the interest. It also has a subtle dependency: "first" means first recorded. If your identity data only persists a few days, the earliest touch you can see is not the actual first touch — first-click attribution silently becomes "first-click-we-still-remember" attribution.

3. Linear attribution

The rule: credit is split equally across all touchpoints. Four touches, $250 each.

What it rewards: participation. Every channel that showed up gets paid, which makes linear a fair-minded baseline and a good first multi-touch model to adopt.

What it punishes: nothing — and that's the weakness. A glancing blog visit earns the same credit as the ad click that started everything. Linear answers "who was present?" rather than "who mattered?"

4. Time-decay attribution

The rule: touchpoints closer to the conversion earn more, typically via an exponential half-life (7 days is a common default: a touch 7 days before conversion earns half the weight of one on conversion day).

What it rewards: recency — the nurturing and closing interactions in the final stretch. It fits short buying cycles, e-commerce, and promotions where the last week genuinely decides the sale.

What it punishes: patient demand creation. In a long B2B cycle, the touch that mattered most may be the earliest one — precisely the touch time-decay pays least.

5. Position-based (U-shaped) attribution

The rule: 40% to the first touch, 40% to the last, and the remaining 20% split across the middle.

What it rewards: openers and closers equally — a sensible compromise for considered purchases where discovery and conversion are both decisive moments.

What it punishes: the middle of the journey — the comparison content, email nurture, and repeat research visits that keep a long deal alive. And the 40/20/40 split is convention, not evidence; nothing in your data says first and last are worth exactly 40% each.

One journey, five answers

Here's the same $1,000 conversion scored by all five models (time-decay uses a 7-day half-life; figures rounded):

Same journey, same conversion — and the LinkedIn ad is worth anywhere from $0 to $1,000 depending on which formula you chose. That spread is the point: an attribution model is a lens, not a measurement. Channels whose value swings wildly across models are the ones to investigate, because the swing means their contribution is concentrated at one end of the journey.

A sixth family worth knowing: data-driven attribution, where an algorithm estimates each touchpoint's contribution by comparing converting and non-converting paths instead of applying a fixed rule. Several large analytics and ad platforms now default to it. It can outperform fixed rules — but it is even hungrier for complete journey data, and it inherits every gap in your identity layer.

Every Model Is Only as Good as Its Identity Spine

Look back at the worked example. All five models assumed the four touchpoints were connected to one person. That assumption is doing more work than the formulas are.

If the Day-1 LinkedIn click happened on a phone and the Day-14 conversion on a laptop, a tool without cross-device identity resolution sees two unrelated visitors. If the analytics cookie was set by JavaScript, Safari's tracking prevention caps it at about 7 days — so by Day 12 the Day-1 visit belongs to a stranger. Either way, the journey your attribution model actually scores is one or two touchpoints long, and every model collapses toward the same answer: credit the final session. You can select linear or position-based in a dashboard and still be running last-click in practice, because last-click is what a one-touch journey produces under any formula.

Truncated journeys don't fail loudly. The report still renders, the percentages still sum to 100 — the credit is just quietly reassigned to whichever channel touched the visitor last.

This is why identity resolution comes first, before any modeling choice. The spine of a trustworthy attribution setup is a persistent identifier that survives weeks-long journeys: a server-set first-party cookie (ClickStream's persist for roughly 400 days — the practical browser maximum), plus an identity graph that links the identifiers a visitor accumulates across devices into one profile, so an authentication on any one device retroactively unifies the journey. We've written a full deep dive on this dependency in Multi-Touch Attribution Requires Multi-Session Identity.

The 13 click-ID types that anchor paid attribution

For paid channels, the spine has a second component: click IDs. When someone clicks an ad, the platform appends an identifier to the landing URL. These parameters are deterministic proof of which ad drove which visit — and they appear exactly once, on the click-through pageview. Capture them then or lose them.

That's 13 click-ID types across the major ad platforms. The ClickStream SDK captures all 13 automatically on every pageview — along with the five UTM parameters — and stores them against the visitor's persistent identity, so the Day-1 li_fat_id is still attached to the profile when the Day-14 conversion arrives.

Choosing an Attribution Model (and Keeping It Honest)

There is no correct model — only a model whose bias matches the question you're asking:

• Short buying cycles (e-commerce, self-serve): time-decay fits journeys where the final days decide the sale.
• Long, considered cycles (B2B, high-ticket): position-based respects both discovery and close; first-click is useful specifically for sizing top-of-funnel channels.
• Getting started: linear is the gentlest step up from last-click — it's hard to game and exposes how much of your journey data was invisible before.
• Always: run at least two models side by side. Model disagreement isn't noise; it's a map of where in the journey each channel does its work.

Whichever model you choose, three hygiene practices protect the input data. Keep campaign tagging consistent — a disciplined UTM taxonomy is what makes touchpoints classifiable in the first place. Handle consent correctly, since unconsented sessions become journey gaps. And exclude automated traffic before modeling: bots don't convert, so bot-polluted touchpoint counts dilute the denominator of every channel they touch.

The Bottom Line

• An attribution model is a credit-splitting rule, not a measurement. Each one encodes an opinion about where journeys are won.
• Last-click rewards closers; first-click rewards discovery; linear rewards presence; time-decay rewards recency; position-based rewards the bookends.
• All five assume a complete, person-level journey. Broken identity truncates journeys, and truncated journeys make every model behave like last-click.
• Fix the spine first — persistent first-party identity, cross-device resolution, and click-ID capture — then let the models disagree productively.

Choose your attribution model second. Choose your identity foundation first — it decides how much journey there is to attribute.