# Attribution models, worked through on one B2B deal

> What first-touch, last-touch, linear, time-decay, position-based, W-shaped and data-driven attribution each do, worked through on one deal, and how self-reported attribution complements them.

An attribution model is a rule for splitting the credit for a conversion between the touches that came before it. No model is true; each answers a different question, and the skill is knowing which question you are asking. This guide runs one €30,000 opportunity through six rule-based models, then covers data-driven models and self-reported attribution.

## What a model does

Take an account that saw a LinkedIn ad, read a post, signed up for a webinar, clicked a nurture e-mail and finally searched for you and asked for a demo. Sales created a €30,000 opportunity. Every touch contributed something, and no log can tell you how much each one changed the outcome.

A rule-based model assigns credit by position or by time. A data-driven model estimates it from the difference between journeys that converted and journeys that did not. Neither measures cause and effect; for that you need a holdout test, where a randomly chosen group does not see the campaign.

Two choices matter as much as the model. The lookback window decides how old a touch may be and still earn credit; 90 days is a common default in B2B. The unit decides whose touches count. In B2B, count every touch from everyone at the account, because the person who books the demo is rarely the one who first saw the ad.

## One deal, six models

The touches, oldest first, with the days before the opportunity was created, and the euros of the €30,000 opportunity each model credits to each touch.

| Touch | Days before | First touch | Last touch | Linear | Time decay | Position-based | W-shaped |
| --- | --- | --- | --- | --- | --- | --- | --- |
| LinkedIn ad click | 30 | €30,000 | €0 | €6,000 | €800 | €12,000 | €9,000 |
| Organic post click | 21 | €0 | €0 | €6,000 | €1,950 | €2,000 | €1,500 |
| Webinar sign-up (lead created) | 14 | €0 | €0 | €6,000 | €3,890 | €2,000 | €9,000 |
| Nurture e-mail click | 7 | €0 | €0 | €6,000 | €7,790 | €2,000 | €1,500 |
| Search ad and demo request (opportunity) | 0 | €0 | €30,000 | €6,000 | €15,570 | €12,000 | €9,000 |

Time decay uses a seven-day half-life. Computed with the attribution code Demander runs, rounded to the nearest €10.

## What each model is good for

- **First touch.** All credit to the touch that introduced the account. Good for judging which channels create new demand. Blind to everything that moved the deal forward afterwards.
- **Last touch.** All credit to the final touch. Good for judging what converts people who are ready. In B2B it over-credits branded search and direct visits, which often only record a decision made elsewhere.
- **Linear.** Equal credit for every touch. A fair default when you have no reason to favour a position, and easy to explain. It rewards channels that produce many touches, useful or not.
- **Time decay.** Credit halves for every half-life before the conversion. Suits short, sales-led cycles where recent contact does most of the work, and undervalues the awareness that started the journey.
- **Position-based (U-shaped).** 40 percent to the first touch, 40 to the last and 20 shared by the middle. Matches the common view that introducing and closing matter most. The 40/20/40 split is a convention, chosen and never measured.
- **W-shaped.** 30 percent each to the first touch, the touch that created the lead and the touch that created the opportunity, and 10 percent to the rest. The closest rule-based fit to a B2B funnel, provided your milestones are recorded reliably.

## Data-driven attribution

Data-driven models learn the weights from your own data. Each ad platform offers one, but each sees only its own touches, so each tends to credit itself. A model across all channels needs your own journey data.

A transparent method is the Markov chain with removal effects. Treat each journey as a chain of steps from a start to either a conversion or nothing, and estimate the probability of moving from each step to the next. Then remove one channel, sending every journey that passes through it to nothing, and measure how far the overall chance of converting falls. That fall, the removal effect, is the channel’s importance. Scale the effects to add up to one and you have credit.

The method needs journeys that did not convert as well as those that did: a channel that turns up as often in lost deals as in won ones has not earned much. It also needs volume. With a few dozen opportunities a quarter the estimates are noisy, and a rule-based model you understand may serve you better.

## Self-reported attribution

Self-reported attribution means asking people how they heard about you, usually in an open field on the demo form or as a question on the first sales call. It captures what tracking misses: a podcast, a recommendation from a peer, a post seen in a private community, a talk at an event.

Its weaknesses mirror tracking’s strengths. People name the touch that made the biggest impression, which may be the latest or the fifth, and they compress months into one answer. Answers need cleaning before they can be counted, and the form of the question shapes the answer: free text tends to be more candid than a drop-down that suggests options.

Use both views. Where they agree, you can be confident. Where they disagree, the gap is information: many people citing a colleague while tracking shows a direct visit first tells you word of mouth is doing work your reports cannot see. Demander does not collect the answer itself; keep it on the CRM record next to the opportunity and compare.

## Attribution in Demander

The tracker records one touch per visit with its source: the campaign and ad behind a paid click, the organic post, the e-mail step or the referral. E-mail sends and replies and direct messages are touches too. CRM milestones such as opportunity creation, stage changes and meetings appear on the account timeline but never take credit themselves. Impressions are shown in aggregate beside the paths, because platforms do not report them per person.

Reports show first touch, last touch, linear, time decay, position-based and W-shaped side by side, and a Markov removal-effect view over converting and non-converting account journeys. Credit rolls up by channel, campaign, persona, message or creative, and every row drills down to the conversions it fed. The path explorer shows which sequences of touches convert and which never do. The attribution window is 90 days by default and can be set per conversion.

Starter includes first and last touch; multi-touch models come with Growth, and holdout tests for incrementality with Scale. See [pricing](https://demander.to/pricing/).

## Mistakes that distort the picture

- **Judging awareness on last touch.** Channels that start journeys always look weak under last touch. Cut them on that basis and the loss shows up in pipeline a quarter later, when nobody connects the two.
- **A window shorter than the cycle.** A 30-day lookback in a 120-day sales cycle hands all the credit to late touches and none to what started the conversation.
- **Crediting people instead of accounts.** In a sale with four or five people involved, a person-level view splits one journey into several short ones and loses the early touches entirely.
- **Switching models mid-argument.** Every channel owner can find a model that favours their channel. Agree on the model before the budget review starts.

> **Pick one model for decisions.** Several models side by side help you understand your funnel. Deciding with several at once is how teams end up arguing. Choose one for budget decisions, write down why, and use the others to check it.

## Questions

**Which attribution model should a B2B company use?**

If you record lead and opportunity creation reliably, W-shaped is the closest rule-based fit to a B2B funnel. If not, position-based is a sound default. Check either against first touch, which shows where demand starts, and test the channels that matter most with a holdout.

**Why do the ad platforms report more conversions than my CRM?**

Each platform counts every conversion it can link to its own ads, with its own window and often with view-through and modelled conversions included. Add the platforms together and the same deal is claimed several times. Cross-channel attribution credits each deal once.

**How long should the lookback window be?**

Longer than your typical time from first touch to opportunity. For many B2B products 90 days covers most journeys, which is why Demander uses it as the default. For enterprise deals with long research phases, set it longer.

**Does attribution tell me what happens if I cut a channel?**

Not reliably. Attribution describes who touched the deals you won. Whether those deals would have happened anyway is a causal question, and only an experiment such as an audience or regional holdout answers it.

## Related

- [Attribution](https://demander.to/features/attribution/)
- [Thompson sampling for budgets](https://demander.to/resources/thompson-sampling-budget-allocation/)
- [Offline conversions](https://demander.to/resources/offline-conversions/)
- [Website tracking](https://demander.to/features/website-tracking/)
- [Pricing](https://demander.to/pricing/)

---
Source: https://demander.to/resources/attribution-models-explained/ · Demander · reviewed 2026-10-02