Attribution Model

Every model needs three inputs and produces one output.

• A recorded journeyThe touchpoints one person had with your marketing: ad clicks, organic visits, emails, calls, sometimes impressions. Only recorded touches exist to the model; a visit outside your storage window or on another device is not in the journey.
• A conversionThe event being credited: a form fill, a booked call, a qualified lead, a closed deal with a value. In lead generation the model gets more useful the further down that list the conversion sits.
• A windowHow far back from the conversion the model looks. The attribution window decides which touches are eligible before any rule is applied.

The output is a share of the conversion per touchpoint, summed into the channel, campaign and ad rows you read in a report. Change the model and the rows change; change the window and the journey itself changes.

• First-click

  100% to the first recorded touch. Answers 'what started this journey' and credits awareness and content. Ignores everything that closed the deal. Full entry.

• Last-click

  100% to the last touch before the conversion. Answers 'what closed this' and matches what the CRM stamps as lead source. Overpays brand search and retargeting. Full entry.

Single-touch models are simple and cheap to run, which is why last-click became the default in every tool. Their common fault is that they answer one question and are read as if they answered all of them.

• Linear

  Equal credit to every recorded touch. Nothing disappears and nothing is weighed. Full entry.

• Position-based (U-shaped)

  40% to the first touch, 40% to the last, 20% shared by the middle. The B2B compromise, with W-shaped and full-path variants that anchor on pipeline stages. Full entry.

• Time-decay

  More credit the closer a touch is to the conversion, usually halving every seven days. Fair in short cycles, last-click in disguise in long ones. Full entry.

• Data-driven

  Weights estimated from converting and non-converting journeys rather than set by rule. The default in Google Ads since late 2021, and the only model besides last-click that Google Ads and GA4 still offer after the rule-based ones were removed in 2023. Needs volume and cannot be audited by hand.

Multi-touch attribution has its own entry, and the guide to running it in a B2B pipeline is multi-touch attribution for B2B.

0:39

The same €12,000 deal appears under every model entry in this glossary: a Meta ad click on day 1, a generic Google Search ad on day 8, a newsletter click on day 15 and a brand search ad on day 22 that led to the form, with the deal closing five weeks later. Here is what each model reports for it.

• First-click: Meta €12,000. Everything else €0.
• Last-click: brand search €12,000. Everything else €0.
• Linear: €3,000 to each of the four touches.
• Position-based: Meta €4,800, brand search €4,800, generic search €1,200, newsletter €1,200.
• Time-decay (7-day half-life): Meta €800, generic search €1,600, newsletter €3,200, brand search €6,400.

Brand search is worth anything between €0 and €12,000 on the same deal. Nothing about the customer changed between those rows. A budget decision that does not know which row it is looking at is a decision made by the model, not by the marketer.

• Start from the question'Which channels find customers' wants first-click. 'What closes them' wants last-click. 'Where is the budget going and is it all pulling' wants a multi-touch view. A model chosen to flatter a channel answers none of these.
• Match the sales cycleThe longer the journey, the more a recency model drifts to last-click and the more the first touch matters. Weeks of consideration argue for position-based or a first-click column; same-day decisions can live with time-decay.
• Do not add the platforms upGoogle Ads and Meta each run their own model on their own data and each credit themselves for the same lead. Their totals overlap; the CRM's count of closed deals is the only number that has to add up.
• Attribute the deal, not the formA model that credits form fills scales the channels that produce cheap forms. Run the same models on qualified leads and closed revenue from the CRM and the winning channels often change.

An attribution model is a lens, and every lens distorts something. Single-touch models are clear and wrong at the edges; multi-touch models are fairer and depend on how much of the journey you recorded. Pick the model for the question, read two of them side by side, run them on closed deals rather than form fills, and keep the journeys long enough that 'first touch' means the first one. Marketing attribution software exists to do the last part, and the choice of model is the easy bit once it has.