Attribution Done Right to Maximize Incrementality

Tuomas Mäki Dec 05 2018 3 PM | 4 min read

Attribution is a somewhat tricky concept to wrap your head around but one that we simply could not pass up, given its impact on driving ROI for your brand if you get it right. If we get back to the basics, we first need to understand what matters most when measuring the effectiveness of your digital advertising efforts – that is incremental conversions, i.e. users who converted from seeing your ad. Now that we’ve nailed down the ‘What’, on to the ‘How.’

  1. How do I know which attribution model is the right one to measure my marketing efforts?

  2. How can I use the results from my attribution model in the best possible way?

One thing we know for sure – attribution helps advertisers better understand where to spend for maximum ROI and maximize incremental results. For the best results, an attribution model should reflect incremental conversions.

Let’s look at the four common types of attribution models, and the pros and cons of each. Making the best choice involves a trade-off between accuracy, ease of implementation, ease of use and investment cost.

  1. Econometric models: Time-series based regression models that model the impact of marketing spend on sales, excluding any external factors like seasonality. They are a great way to complement user-based bottom-up models and are the only viable option when a substantial amount of budget is used in channels where users cannot be identified, i.e. TV and print. A challenge with these models is that there is often not enough variance in historical channel-specific marketing spend amounts to regress the correlation reliably.

  2. Ad Network native models: Bottom-up models, for example, a 7-day click attribution model. This means anyone who clicks an ad and ends up converting on the advertiser’s web page or app within seven days from the click will be attributed to the campaign. On the flip side, conversions are not de-duplicated across channels, and this model assumes all conversions are incremental.

  3. Rule-based models: The simplest group of the user-based multi-channel attribution models, where we allocate conversion attribution to the different touch points along the customer journey, based on pre-set rules. However, one major limitation is that they generally assume all conversions are incremental.

  4. Algorithmic models: Machine learning models on a user journey level that aim to capture the incremental effect between ads and user actions, whose logic is typically based on comparing users that have been exposed to ads against users that have not been exposed to ads. Due to the complexity of the model, it can prove difficult to get hold of all the required data and verify it’s correct.

We’ve looked at the different types of attribution models, but how do you measure the results? Unfortunately, attribution models will never be 100% accurate, but one way to make sure the results are as accurate as possible is to always cross-check the results with lift tests as these are the gold standard for measuring incrementality. The lift test calibration method is one method that measures the ratio between attributed and incremental conversions and converts this ratio into a multiplier. The multiplier is then applied to convert the attributed conversions into estimated incremental conversions. Note that the results are accurate only at one point in time, one spend level and one market. The further away we move from those targets, the higher chance of the results being flawed, and the lift testing cycle repeats. However, lift testing provides benefits over the attribution models as long as advertisers are aware of how and when the method can be used.

There’s no such thing as a one-size-fits-all approach when choosing the right attribution model, but following these key steps can take you one step closer to finding one that works:

  • Understand the assumptions and limitations behind each model.

  • Cross-check results with lift tests.

  • 100% accuracy doesn’t exist with attribution models, but applying the right methods to a simple model will bring you closer.

  • Everyone has a part to play with understanding attribution, however it helps to assign attribution efforts to at least one person in your organization who can also educate other colleagues.

Watch the video below to learn more about incrementality from our Data Scientist Dr. Lauri Kovanen.

We’ve also created a short guide to help you understand the models available, and the limitations behind them. We hope that this will help you evaluate the right model to drive ROI and make sure your advertising dollars are being channeled effectively. Read our latest e-book to learn more.

Tuomas Mäki

Read Next