Lead Attribution
Lead attribution is the practice of assigning credit to the marketing channels and touchpoints that contributed to creating a lead.
Key takeaways
- Identity stitching decides whether any attribution model has real data to divide.
- Last-touch overpays the channel nearest the form; first-touch overpays discovery.
- Store first-touch and last-touch separately and never overwrite either field.
- Multi-touch models need complete timelines; on partial data they are worse.
- Offline and private-channel influence cannot be credited and is silently missing.
In depth
Attribution works in two stages that are often confused. First identity resolution: anonymous sessions, form submissions and CRM records have to be joined into one timeline for one person, usually through a cookie or click identifier carried into the lead record. Second the credit rule, which distributes a fixed amount of credit across the touches in that timeline. The rule is arithmetic and easy to change; the stitching is engineering and determines whether the arithmetic has anything real to work on.
Model choice is a trade between simplicity and fairness. Last-touch is cheap to run and systematically overpays the channel closest to the form, which is usually branded search or direct. First-touch overpays discovery and ignores everything that closed the gap. Multi-touch spreads credit and requires far more complete data to be worth anything, so a linear model on a half-stitched timeline is less accurate than a well-understood last-touch view. Longer cycles and more channels push every model's error upward.
A workable setup captures source at the earliest possible moment and never overwrites it: a first-touch field written once, a last-touch field updated each visit, and the raw campaign parameters stored alongside. Reports then run both models side by side and the disagreement between them is itself the finding. In a quiz funnel this is straightforward, because the entry URL parameters and referrer can be stored with the result and travel into the lead record when contact details are captured.
Some influence is structurally invisible. A podcast mention, a conversation in a private community, a recommendation from a colleague and anything that happens on a device you never see leave no trace to credit. Consent choices and browser restrictions remove more of the trail every year. Attribution is therefore a model of the visible part of demand, and the honest use is directional: it ranks channels well enough to shift budget, and it is not an accounting of who caused what.
Example in practice
How to measure it
Start with data quality, not with credit. Report the share of leads carrying a known source, the share with a complete touch timeline, and the share labelled direct or unknown. Those three numbers cap how much any model can be trusted. A channel report built on data where a third of records have no source is a ranking of your tracking, not of your marketing.
Then validate the model against something it did not see. Pause or geo-hold one channel and watch whether total qualified leads fall by roughly the amount the model attributed to it. Repeated a few times across channels, this tells you which parts of the report are load-bearing. Where a pause changes nothing, the credited channel was riding on demand created elsewhere.
Common mistakes
The most common failure is losing source data at a redirect. A campaign link passes through a shortener, a consent wall or a language redirect, the parameters are stripped, and the lead arrives labelled direct. Test every real entry path end to end, not just the tagged URL in a spreadsheet. Then check the share of leads with no source: if it is large and growing, fix the plumbing before arguing about which model to use.
The second is switching models until one flatters the channel someone wants to keep. Each switch produces a new set of winners and no new information. Pick a primary model, write down why, and treat the others as secondary views reported at the same time. If a decision would reverse depending on the model, that decision needs evidence from a holdout test rather than a different weighting.
Frequently asked questions
What are the main lead attribution models?
The most common are first-touch, last-touch, linear, and time-decay multi-touch. First- and last-touch credit a single interaction, while multi-touch models spread credit across the whole journey.
Why do so many leads show up as 'direct' or 'unknown'?
Usually the source data, such as UTM parameters or referrer information, was lost before the lead was recorded. Persisting that context at the moment of capture, as a quiz funnel does, dramatically reduces unattributed leads.
Is single-touch or multi-touch attribution better?
Single-touch is simpler but ignores assisting channels, while multi-touch is more accurate but needs cleaner data. Many teams start with last-touch and graduate to multi-touch as their tracking matures.
Which attribution model should I start with?
Start with first-touch and last-touch reported together, because both are cheap, easy to explain and require only one stored field each. The gap between them already shows which channels open journeys and which close them. Move to a multi-touch model only once your touch timelines are reliably complete, otherwise the extra sophistication multiplies existing gaps.
How do I attribute leads that come through several devices?
Only after login or an email match ties the sessions together, which usually happens at the moment of form capture. Before that point the same person looks like two visitors. Practical setups accept the loss on the early touches and attribute from the first identified session onward, stating the limitation rather than pretending the timeline is complete.
What should I do with leads attributed to direct or unknown?
Treat them as a data-quality metric rather than a channel. Report the share separately, investigate the biggest contributors, and never redistribute them across known channels to make a chart look tidy. Common causes are stripped parameters, consent blocking, untagged internal links and people typing your name into a browser after seeing something offline.
Is lead attribution possible without cookies?
Partly. Self-reported source questions on the form, unique landing pages per channel, campaign-specific offer codes and server-side click identifiers all work without third-party cookies. Each is coarser than a full timeline, but together they usually rank channels well enough to allocate budget, which is what most attribution is actually used for.
How does lead attribution differ from revenue attribution?
Lead attribution credits the creation of a contact; revenue attribution credits closed money, which arrives much later and in uneven amounts. A channel can lead on volume and trail badly on revenue if the leads it produces rarely close. Run both if your cycle allows, and let revenue attribution settle any conflict between them.
Should I ask leads how they heard about us?
Yes, as a supplement rather than a replacement. Self-reported answers capture influence no tracker sees, such as a podcast or a colleague's recommendation, but people misremember and pick whatever option is first. Keep the list short, add a free-text option, and compare the answers with your tracked source to find where the two systematically disagree.