Attribution vs tracking
Tracking is the recording layer. It logs that a click happened, that a UTM parameter arrived intact, that a specific person opened an email on a Tuesday afternoon. Attribution is the interpretive layer sitting on top of that record. It takes a pile of tracked events belonging to the same person and decides which one, or which combination, gets credit for the sale, signup, or booking that eventually happened. You can have precise tracking and zero attribution, if nothing in your stack ever links those events to a conversion. You can also run a named attribution model on top of sloppy tracking, in which case the model produces a confident, wrong number.
That distinction is the reason attribution vs tracking gets searched as its own question. People start out wanting to know where a sale came from, discover they need campaign tracking just to capture the raw events, and only then run into attribution as the separate step of turning those events into a credit assignment.
How the credit assignment actually works
Every attribution model needs the same two ingredients: a way to identify that touchpoint A and touchpoint B belonged to the same visitor, and a rule for splitting credit once that chain exists. The identity piece usually comes from a cookie, a click ID, or a logged-in account. The rule piece is where the models diverge, and it's the part that actually defines an attribution model.
| Layer | What it captures | Question it answers | Example |
|---|---|---|---|
| Tracking | Individual events as they happen | Did this happen, and with what parameters? | A click landed with utm_source=instagram at 3:14pm |
| Attribution | A rule applied across a person's tracked events | Which touchpoint gets the credit? | That Instagram click, not the email opened three days later, gets the sale |
The three attribution model families
Every named model on the market is a variant of one of three approaches. Knowing which family a model belongs to tells you what it actually needs to work before you spend time setting it up.
| Family | How credit is assigned | Data it needs | Realistic fit |
|---|---|---|---|
| Single touch (first touch or last touch) | 100% of the credit to one touchpoint | One cookie or click ID per conversion | Solo creators, small stores, most service businesses |
| Multi touch (linear, U shaped, time decay) | Credit split across every recorded touchpoint in the journey | A stitched identity across multiple sessions and devices | Teams running several paid channels at real, sustained volume |
| Algorithmic or data driven | A model infers each touchpoint's contribution from historical conversion patterns | Hundreds of conversions a month, ideally more | Large advertisers, and the bidding systems built into ad platforms themselves |
Single touch splits further into first touch and last touch attribution, which disagree with each other constantly and are worth understanding as their own comparison.
Why attribution is never the full picture
Attribution is not a record of what actually caused a sale. It is a convention, a rule you chose, applied to whatever fraction of the customer journey you happened to observe. Most journeys are not fully observed. A person sees a Reel on their phone, thinks about the product for two days, switches to a laptop, searches your brand name directly, and buys. Your tracking sees a direct visit with no referrer and a purchase. It never sees the Reel. No attribution model, however sophisticated, can assign credit to a touchpoint it never recorded.
This is why cross-device journeys, offline conversations, and app-to-web handoffs quietly break every attribution report, and why two tools measuring the same business can report different winning channels without either one being wrong. They are both accurately modeling an incomplete record.
Which model should you actually use
For almost any operator running one or two channels, the honest answer is the simplest model that still changes a decision: last touch attribution, or first touch when what matters more is what introduces new people than what closes them. Multi touch and algorithmic models only earn their complexity once conversion volume is high enough that the models would actually disagree with each other in a way that matters. Building one on a trickle of monthly sales just adds noise dressed up as precision.
A link tracking tool is usually enough to run single touch attribution properly. Raydar, for example, captures both a first-touch and a last-touch cookie on every click through a tracked link, along with the UTM parameters and referrer, so a small operator can answer which post produced a given sale without standing up a dedicated attribution platform. The window you choose for counting a click toward a later conversion matters just as much as the model, which is covered in what is an attribution window.
Once volume genuinely supports it, multi touch attribution gives a fuller picture, but treat it as a later problem, not a starting one. Get first touch and last touch to roughly agree first, and check why your different tools never quite match before trying to model anything more advanced.
Attribution meaning, in one line
If you only remember one sentence: attribution is a chosen rule for splitting credit, applied after the fact, to a partial record of what a customer did. It is not a measurement in the way a scale measures weight. Two businesses can look at the exact same tracked events and produce different attribution numbers, simply because they chose different rules. That's simply how the exercise works, and knowing it up front saves a lot of arguing over which dashboard is lying. Usually neither one is. They just chose differently.