Attribution issues disrupt the optimisation problem

In an ideal world the optimisation problem would be resolved by marketeers continuously experimenting with strategies they believe will be most effective and measuring the results against marketing KPIs. Measuring whether a strategy is effective requires the accurate and proportionate attribution of conversion events* to the relevant marketing activities. However, in reality attribution is a hard problem to resolve (primarily due to restrictions in the data available) meaning most marketeers are unable to accurately measure the effectiveness of a strategy. As marketing efforts scale up, not only do the stakes of the optimisation problem get raised but the number of concurrent interdependent marketing activities increases, making the attribution problem even more challenging.

*A conversion event is any action a prospect/customer takes which generates revenue or increases the probability of generating revenue examples include a website visit, subscription, signup and checkout.

Digital marketing platforms don’t resolve the attribution issue

Digital marketing and affiliate platforms report marketing KPIs such as ROAS and CAC against activities. These KPIs are based off the spend, number of conversions and assigned conversion values. A conversion is recorded when a user interacts with the marketing activity and subsequently completes an action that is registered as a conversion event/goal in the platform within the given attribution window. These actions are monitored by tracking pixels and conversion APIs. However, these KPIs cannot be taken at face value because:

• The same underlying conversion can be counted multiple times - Each platform operates in a silo meaning it is unaware of any marketing activity away from the platform that contributed to the conversion. This gives rise to instances where multiple platforms all report conversions for the same underlying conversion event. As a result the KPI figures each platform presents typically exaggerate the effectiveness of the activity.
• Can’t distinguish between new and existing customers - Platforms do not have access to the source of truth for conversion events (the database, CRM, ERP) so cannot reliably distinguish between new and existing customers. This leads to many conversions being incorrectly categorised as new customer acquisitions.
• All interactions are not equal - Digital marketing activity is optimised to drive the best in-platform results, this means displaying adverts to people who are most likely to convert. In many cases these people would have converted anyway, meaning the interactions did not increase the probability of the conversions at all (this is especially the case for view-through conversions). As a result the KPI figures each platform present typically exaggerate the effectiveness of the activity.
• Attribution windows are restrictive - The maximum attribution window available within most platforms is 30 days, if a conversion event occurs later than this timeframe it is not recorded. For companies with long sales cycles this means either a portion of conversions are not recorded or an upstream conversion event such as a website visit or signup must be used in place of a revenue generating conversion event. The results obtained when measuring the performance of activity using an upstream conversion event are not always representative.
• Halo and cannibalisation effects - Platforms report on each activity in isolation meaning they fail to capture the interaction effects between activities. It may be that brand awareness adverts have a significant positive impact on the conversion rate of search adverts or that branded search advertising is capturing users who would have visited the site through an organic result. In the former case brand awareness advertising is having a halo effect and in the latter case branded search advertising is cannibalising organic traffic. Platform reporting will not capture either of these effects, and the values they do report will be distorted by such effects.

Using the reporting from digital marketing platforms to resolve the attribution problem is not an issue with a small marketing budget and only a couple of digital marketing channels but as marketing teams grow more sophisticated they typically adopt a method for resolving attribution.

Methods of attribution - direct and indirect

There are two broad methods of attribution; direct and indirect. Direct attribution sees individual conversion events attributed to marketing activities e.g. the registration of user x can be attributed to the Facebook advert ab13 and email yz51. Indirect attribution sees the total number of conversion events within a given timeframe attributed to marketing activity e.g. 36 out of the 210 registrations which took place on the week beginning 20/05/2023 can be attributed to Facebook in-stream adverts.

Direct Attribution

Direct attribution examples:

• Click-based attribution
• Post-purchase surveys
• Discount codes

Direct attribution methods are only possible when:

• Engagement with marketing activity by a user is recorded.
• Conversion events completed by a user are recorded.
• The same user who completed a conversion event can be identified as the user who engaged with the marketing activity.

Direct attribution benefits:

• Verifiable: Any KPI figures derived from direct attribution methods can be verified because the it is possible to see exactly which conversion events were attributed to the marketing activity.
• Granular Insights: It is possible to calculate KPI figures from direct attribution methods at a very low granularity e.g. an individual advert. The only constraint is the granularity at which engagement is recorded.
• Feedback time: It is possible to frequently make adjustments to the mix of marketing activity and measure the effects of these when using direct attribution methods. The only limiting factor is the typical time from interaction to conversion event for a user.

Direct attribution drawbacks:

• Excludes some marketing activity: Any engagement which doesn’t meet the criteria listed above is excluded from the model, such as Television advertising or watching a TikTok in-stream advert outside of the view-through attribution window. This means KPIs figures for activities which are frequently included (e.g. Google Branded Search) are inflated, KPIs for activities which are frequently excluded (e.g. TikTok in-stream adverts) are deflated and KPI figures cannot be calculated for activity which is always excluded (e.g. Television advertising).
• Excludes some users: Users who block/opt out/ignore the data capture are excluded. This means results are only based on a sample of users which may not be fully representative.
• Does not directly measure the importance of the engagement: Just because a user saw/clicked/reported seeing an advert does not necessarily mean that the advert increased their probability of converting. Direct attribution is unable to accurately infer the impact of that engagement.

To summarise, direct attribution methods are suitable when you need to frequently measure and compare the impact of performance marketing adverts and campaigns. It is not suitable for a holistic attribution across all marketing activity.

Indirect Attribution

Indirect attribution examples:

• Marketing mix modelling
• Geo-based lift tests

Direct attribution methods are only possible when:

• The total engagement or reach of the marketing activity can be measured or accurately estimated
• The marketing activity generates a significant proportion of total engagement within the timeframe it is active (marketing activity can be grouped together to satisfy this constraint)
• There is significant variation in engagement across geographies, product ranges or time
• There is a significant history of conversions data (typically 2 years is sufficient)

Indirect attribution benefits:

• Able to include all marketing activity: Indirect attribution does not require individual interactions and conversions by individuals to be recorded meaning all marketing activity can be included, which in turn allows KPIs to be calculated for all marketing activity.
• No users excluded: Since indirect attribution deals with aggregate numbers, there is no subset of users excluded from the model. This means results are fully representative and the true scale of the impact of the marketing activity can be accurately measured.
• Importance is readily inferred: Indirect attribution centres around comparing engagement and conversions across dimensions (for example time or geographic region). If changes in engagement for a particular marketing activity consistently and significantly impact overall conversions, it can be reasonably inferred that the marketing activity plays a crucial role in driving conversions.

Indirect attribution drawbacks:**

• Non-verifiable: It is not possible to see which specific conversion events were attributed to the marketing activity meaning the KPI figures cannot be verified.
• High-level Insights: Due to the constraints in the level of engagement required with the marketing activity in order to get results, KPI figures can only be calculated at a high level of granularity e.g. Facebook in-stream adverts.
• Feedback time: Unless the model is extremely well tuned, it typically takes multiple sales cycles for the model to accurately reflect changes in the marketing mix.

To summarise, indirect attribution methods are suitable for building a holistic understanding of attribution. They are not suitable for comparing the performance of individual adverts or frequently measuring the impact of incremental changes to the marketing mix.