Most explanations of attribution models define each one in isolation, which is why the differences never quite stick. This guide does it differently: one customer journey, run through every model, so you can see exactly how the same $2,000 sale gets carved up differently depending on the rule you choose.
The example journey we’ll use throughout
Meet Dana, who needs her roof repaired. Over eleven days, Dana touches this business five times:
- Day 1 — Facebook ad. Dana sees a storm-damage awareness ad and clicks through to a blog post. Reads, leaves.
- Day 4 — Organic search. Dana searches “how much does roof repair cost,” finds the company’s pricing guide, reads it, leaves.
- Day 7 — Retargeting display ad. A banner follows Dana around the internet. She clicks, browses the reviews page, leaves.
- Day 10 — Branded search (paid). Dana searches the company by name, clicks the branded search ad, and requests a quote form.
- Day 11 — Phone call. Dana calls from the confirmation page to schedule, and books a $2,000 job.
Five touches: Facebook → Organic → Display → Branded Paid → Phone Call. One conversion, worth $2,000. Every model below answers the same question: who gets the credit?
(Notice, before we start, that if this business doesn’t track calls, the journey appears to end at a form-fill — and if Dana had skipped the form and just called, the entire journey would be invisible. Models divide the credit; call tracking decides whether there’s a conversion to divide at all.)
First-touch
Rule: 100% of credit to the first touchpoint.
Dana’s journey: Facebook gets the full $2,000. Everything after day 1 gets nothing.
First-touch answers “what starts journeys that end in revenue?” It flatters demand-creation channels — awareness ads, content, top-of-funnel campaigns — and completely ignores everything that closed the deal. Use it when you’re evaluating how customers discover you; never use it alone to judge closing channels.
Last-touch and last non-direct
Rule: 100% of credit to the final touchpoint. The common “last non-direct” variant skips direct visits and credits the last marketing touch.
Dana’s journey: the phone call — or, one step upstream, the branded paid search click that led to it — gets the full $2,000. The Facebook ad that started everything gets nothing.
Last-touch answers “what finishes journeys?” It flatters demand-capture channels — branded search, retargeting, anything close to the money — and systematically starves the top of the funnel. It’s also the most common default in analytics tools, which quietly shapes how the whole industry thinks. The full argument about when each single-touch lens is right lives in First-Touch vs. Last-Touch Attribution.
Linear
Rule: equal credit to every touchpoint.
Dana’s journey: five touches, $400 each. Facebook, organic, display, branded paid, and the call all get 20%.
Linear is the diplomatic model — nobody’s ignored, nobody’s favored. That’s also its weakness: it asserts that the drive-by display click mattered exactly as much as the branded search that closed. Linear is most useful as a neutral baseline when you suspect your current model is distorting things and want a sanity check.
Time-decay
Rule: credit weighted toward recency — the closer a touch is to the conversion, the more it earns (typically on a half-life curve).
Dana’s journey: the day-11 call and day-10 branded search take the largest shares; the day-1 Facebook ad gets the smallest sliver.
Time-decay encodes a defensible intuition: recent touches are more plausibly connected to the decision. It suits shorter sales cycles and promotion-driven businesses. Its bias is the same as last-touch’s, just softened — demand creation is still structurally underpaid.
Position-based (U-shaped and W-shaped)
Rule (U-shaped): heavy credit to the first and last touches — commonly 40% each — with the remaining 20% spread across the middle.
Dana’s journey: Facebook gets $800, the closing touch gets $800, and organic, display, and the middle touches split $400.
U-shaped encodes another intuition: discovering the customer and closing the customer are the two moments that matter most, and the middle is supporting cast. W-shaped extends this for longer B2B-style funnels, adding a third peak at a mid-journey milestone (like a lead-created event).
These models are popular because they feel balanced. Just stay honest about what the 40/20/40 split is: a reasonable-sounding convention, not a discovered truth.
Data-driven / algorithmic
Rule: no fixed rule. An algorithm compares thousands of converting and non-converting journeys and infers how much each touchpoint type actually shifts conversion probability, assigning fractional credit accordingly.
Dana’s journey: the split depends entirely on what the algorithm learned from everyone else’s journeys. Perhaps it noticed that people who read the pricing guide convert at triple the rate — organic gets a big share. Perhaps branded search clicks happen in nearly every journey, converting or not — branded gets less than last-touch would give it.
Data-driven attribution is the most sophisticated approach and the least inspectable. It needs meaningful conversion volume to learn from, and when the model belongs to the platform selling you ads, its incentives deserve scrutiny — both issues examined in Data-Driven Attribution: Google’s Model vs. Independent Tools. One more dependency worth underlining: an algorithm can only learn from conversions it can see. Feed it web conversions but not calls, and it will confidently optimize toward form-fillers.
Side-by-side credit table
The same $2,000 job, five ways:
| Touchpoint | First-touch | Last-touch | Linear | Time-decay* | U-shaped |
|---|---|---|---|---|---|
| Facebook ad (day 1) | $2,000 | $0 | $400 | ~$130 | $800 |
| Organic search (day 4) | $0 | $0 | $400 | ~$230 | ~$130 |
| Display retargeting (day 7) | $0 | $0 | $400 | ~$400 | ~$130 |
| Branded paid (day 10) | $0 | $0 | $400 | ~$570 | ~$130 |
| Phone call (day 11) | $0 | $2,000 | $400 | ~$670 | $800 |
Time-decay figures illustrative, using a typical recency curve.
Read that table as a budget meeting. Under first-touch, Facebook looks like a hero and branded search looks worthless. Under last-touch, the reverse. Same customer, same revenue, opposite conclusions. The model you choose is a decision about which channels get funded — which is why choosing consciously beats inheriting a tool’s default.
How to choose for your business
Skip the search for the “accurate” model — ground truth doesn’t exist. Instead:
Match the model to the decision. Judging demand capture? Last-touch or time-decay. Judging demand creation? First-touch alongside it. Reporting to a board that wants one number? Position-based is a defensible compromise. High conversion volume and complex journeys? Data-driven earns a look.
Use two lenses, not one. The cheapest sophistication available: run last-touch and first-touch side by side. Channels that look strong in one and absent in the other are telling you what role they play — that’s more insight than any single model provides.
Fix your data before your model. Every model above produced nonsense for businesses that can’t see phone calls, because Dana’s day-11 conversion simply wouldn’t exist in the dataset. Complete conversion capture — including calls fed into your ad platforms — improves every model simultaneously. Model selection is a second-order optimization; data completeness is first-order.
Keep windows in mind. A model can only divide credit among touches inside the attribution window — a mismatch between window and sales cycle distorts every model equally.
Frequently asked questions
Which attribution model does GA4 use?
GA4 defaults to data-driven attribution for most reporting, with last-click variants available in comparison tools. Defaults change over time, so verify in your property — and remember that whatever the model, GA4 divides credit only among the conversions and touchpoints it can see.
What is U-shaped attribution?
A position-based model that gives the first and last touchpoints the largest shares of credit (commonly 40% each), splitting the remainder across middle touches. It encodes the assumption that discovery and closing are a journey’s two most important moments.
When should I use time-decay?
When recency plausibly matters most: short sales cycles, promotional or seasonal businesses, and re-engagement-heavy funnels. For long consideration cycles where early education does the heavy lifting, time-decay will systematically underpay your top-of-funnel work.
Deeper on the two most consequential choices: First-Touch vs. Last-Touch and Data-Driven Attribution. Or zoom out to the complete attribution guide.