An attribution model is a set of rules for how you assign credit to the marketing touchpoints that led to a conversion. The same underlying data (the same sequence of clicks, views, and ultimately a purchase) can produce very different numbers depending on which model you apply.
None of these models is objectively correct. Each is a lens that illuminates certain parts of your customer journey while obscuring others. The goal is to understand what each lens shows you, and what it hides.
First-touch attribution
How it assigns credit: 100% of the conversion credit goes to the very first touchpoint the customer had with your brand.
Example: A customer finds you through a Google search ad, then a week later clicks a Meta retargeting ad and buys. Under first-touch, the Google ad gets 100% of the credit.
When it's useful: Understanding which channels are best at bringing new people into your funnel. If you want to know where your top-of-funnel awareness is coming from, first-touch gives you a clear signal.
What it hides: Everything that happened after the initial contact. The channels that nurture and convert get no credit, so first-touch models consistently undervalue retargeting, email, and bottom-of-funnel campaigns.
Last-touch attribution
How it assigns credit: 100% of the conversion credit goes to the final touchpoint immediately before the purchase.
Example: Same customer, same journey. Under last-touch, the Meta retargeting ad gets 100% of the credit.
When it's useful: Measuring which channels or ads are most effective at closing sales. Last-touch is the default model for most ad platforms, which is why it's the most familiar.
What it hides: Everything that built awareness and intent before the final click. Last-touch models systematically overvalue retargeting and brand campaigns that capture people already in the market, while undervaluing the awareness channels that put them there in the first place.
Linear attribution
How it assigns credit: Credit is divided equally across every touchpoint in the customer's journey.
Example: A customer touched four channels before buying: a podcast ad, an organic search, a Meta ad, and a direct visit. Linear gives each touchpoint 25% of the conversion credit.
When it's useful: Getting a balanced, even-handed view across a long customer journey where multiple channels contribute roughly equally. It's a solid default for businesses that know customers go through a multi-step process but don't have enough data to weight touchpoints more precisely.
What it hides: The reality that not all touchpoints are equally valuable. Treating a display impression that lasted one second the same as a clicked ad and a webinar attendance misrepresents the relative impact of your channels.
Time-decay attribution
How it assigns credit: Touchpoints closer in time to the conversion receive more credit than earlier ones. Credit decays exponentially as you go further back.
Example: A customer first encountered the brand via a YouTube ad 30 days ago, clicked a Google search ad 10 days ago, and converted after clicking a Meta ad today. Time-decay gives the most credit to the Meta ad, less to the Google ad, and very little to the YouTube ad.
When it's useful: Short sales cycles where the most recent interactions genuinely drove the decision, and for businesses where brand awareness is relatively stable and the marginal impact of a touchpoint diminishes quickly over time.
What it hides: The value of early-funnel awareness that may have been essential to the purchase, even if it happened weeks before conversion. If you rely heavily on top-of-funnel content or brand building, time-decay will systematically undervalue it.
Position-based (U-shaped) attribution
How it assigns credit: The first and last touchpoints each receive the most credit (typically 40% each), with the remaining 20% split across any touchpoints in the middle.
Example: Four touchpoints in a journey: first-touch gets 40%, last-touch gets 40%, and the two middle touchpoints each get 10%.
When it's useful: Businesses where both acquisition (first touch) and conversion (last touch) are key decisions, and where the middle of the funnel is more about nurturing than persuasion. This model is a reasonable compromise that avoids the extremes of pure first- or last-touch.
What it hides: Nuance in the middle of the funnel. A webinar or sales call in the middle of a long journey might be the thing that actually drove the decision, but position-based attribution will give it minimal credit.
Data-driven attribution
How it assigns credit: A statistical model analyzes your actual conversion data to determine which touchpoints most frequently appear in converting paths versus non-converting paths, and weights credit accordingly.
Example: The model finds that customers who clicked a YouTube ad at some point in their journey converted 40% more often than those who didn't. YouTube gets proportionally more credit across those paths.
When it's useful: Businesses with high enough conversion volume to train a model reliably (typically several hundred conversions per month, minimum) and who want attribution that reflects their actual customer patterns rather than a manually chosen rule.
What it hides: The model is only as good as the data you feed it. If your tracking is incomplete (missing touchpoints, attribution gaps between click and conversion), data-driven attribution will confidently produce misleading results based on a distorted picture of the journey.
The most important point about all of these models
Switching attribution models changes how credit is distributed, but it doesn't change what actually happened. The customer journey is the same. The touchpoints are the same. The sale happened once.
An attribution model is a lens, not a source of truth. You can only switch lenses meaningfully if you've captured the full journey in the first place.
This is the critical constraint: if your tracking doesn't capture the click-to-lead-to-sale chain. If touchpoints in the middle of the journey are invisible, or if revenue events don't trace back to their originating campaign, then no attribution model can give you accurate results. You'll be distributing credit across an incomplete picture of the journey.
Before debating which model to use, the more important question is: do you have complete journey data to run any model against? That means first-party tracking at the click level, UTM data carried through to lead capture, and payment events tied back to their originating source. With that foundation in place, attribution model selection is a meaningful decision. Without it, it's a choice between imperfect guesses.
A practical starting point
For most growing businesses, linear or position-based attribution is a good default, as it distributes credit in a way that avoids the worst distortions of first- and last-touch, without requiring the conversion volume that data-driven attribution needs to be reliable. As your data quality improves and conversion volume grows, you can experiment with more sophisticated models.
The more actionable discipline is often simpler: build complete journey tracking first, measure your blended ROAS as your top-line health metric, and use whatever attribution model you choose as a lens for channel-mix decisions, not as the single source of truth for any budget call.
Where Cavor fits in
Cavor captures the full click-to-lead-to-payment journey through your CRM and payment processor, giving you the underlying data to run attribution models against actual conversions, not platform-reported ones.