Click-based Attribution

What is click-based attribution and the extent to which it resolves the attribution problem.

How click-based attribution works

Event Tracking

Click-based attribution leverages event-tracking data (e.g. GA4.) on individuals’ visits to a site. Event tracking data captures the source of the traffic (this is passed through utm parameters in the url) and the tagged events that took place during a session. Event tracking can be done via the browser or a server (client-side and server-side tagging). Whilst server-side enables event data to be captured for a greater proportion of traffic and improved resolution to the individual, the implementation is more involved.

Attributing across user touchpoints

Tracking user events and traffic source is not sufficient for click-based attribution, crucially it is necessary to resolve which sessions were the same underlying user. It is possible to do this using the standard user identification of the event tracking platform e.g. Google’s pseudo_user_id or if server-side tracking is used it is possible to build a more persistent individual user identifier based on the identifiable data captured. Click based-attribution then works by apportioning each conversion event across the traffic sources of the user’s preceding sessions, the relative weight assigned to each source depends on the shape of the attribution model chosen.

Why click-based attribution overcomes the issues faced by in platform reporting

  • Uses source of truth - Click-based attribution uses conversion data from your own database/commerce platform meaning there is no issue with double-counting conversions or identifying new vs existing customers.
  • Importance is weighted - The relative importance of an interaction is based upon the shape of the click-based attribution used rather than all being given equal weight.
  • View-through conversions excluded - Only interactions which generated a click are included in turn alleviating the issue of including interactions with marketing activity which did not affect the probability of converting.
  • No attribution window limit - There is no restriction on how long anonymised event-tracking data can be stored for, meaning there is no strictly imposed conversion attribution window.

Issues with click-based attribution

  • It is only possible to capture engagements that result in a click through to your site from users who consent to tracking:
    • Blocking tracking - A significant and growing proportion of traffic is not tracked because users/browsers block event tracking
    • Ignores impressions - Most engagement for digital marketing is in the form of impressions and views rather than clicks, especially brand marketing
    • Excludes offline activity - It is not possible to asses the effectiveness of offline marketing as it is not possible to click through to the a site from them
  • Identifying the same user - Resolving whether or not two sessions are the same underlying user is a difficult problem with no perfect solution and it is becoming increasingly difficult with users now able to share very little identifiable data.
  • Overlap not interaction effects - By using click-based attribution it is possible to identify when the same user engaged with multiple activities, and then measure the KPIs for these ‘overlaps’. However, it is not possible to infer to what extent interaction with one activity caused an improvement/deterioration in performance.

Despite these issues, click-based attribution rightly remains an integral part of attribution for digital performance marketing. If implemented properly marketeers can use the results to compare the performance of a specific activity over time or compare the performance of like-for-like activity. Which in turn enables rapid experimentation.

Tradeoffs in weighting methods

The weighting method determines how much weight to assign each recorded user interaction prior to a conversion. First click attribution assigns 100% of the weight to the first interaction meaning the conversion is wholly attributed to that marketing activity, whilst Last click assigns 100% of the weighting to the last interaction. The simplicity of first and last click attribution methods make them easy to interpret and verify, however neglecting all other interactions with marketing activity typically results in them producing distorted attribution results. In turn, marketeers have responded by creating a number of different multi-touch weighting methods from linear weighting where each touchpoint gets equal weight to u-shaped weighting where the first and last touchpoints get significantly more weight than the rest. Whilst these methods do not suffer from the problem of ignoring touchpoints like first click and last click do, the weighting is set by an arbitrary ruleset rather than reflecting the importance of the the user’s interaction with the marketing activity. This has given rise to data-driven multi-touch attribution.

Data-driven multi-touch attribution

Ideally attribution weighting would be determined by the relative importance of each touchpoint in driving the conversion, unfortunately this data is not available, but by analysing the user journeys across thousands of users and millions of touchpoints it is possible to accurately estimate the extent to which each touchpoint increases the probability of converting, Which is the best available indicator of importance. Kleene calculates this estimation using Markov Chains and Shapley Evaluation. In simple terms this means simulating scenarios where each marketing activity and all associated user journeys are removed and the number of conversions and journeys lost is recorded. The number of conversions lost relative to the total number of journeys lost determines the relative weighting an advert receives. The respective relative weightings are then assigned to every touchpoint for each conversion.