Based on these questions, Wehkamp decided to create and run an experiment for 4 weeks. Their test approach: geo experiments. By dividing the Netherlands into two parts they created a test and control group. Before running the experiment a trend analysis based on historical data was done, to be able to measure causal impact: what would have been the performance if the campaigns were not switched off? Good to know:
1) The attribution model Wehkamp used: last non-direct click with a range of 7 days
2) It’s all non-branded campaigns (branded campaigns are already phased out at Wehkamp because of an earlier incrementality experiment)
3) The experiment was done for Google and Microsoft Ads
4) First-party data was used to generate insights into active, new and reactivated customers
The short term impact
On the first day there was already a noticeable impact on traffic. At the end of the experiment, regardless of attribution, these effects can be considered the incremental impact of paid search for Wehkamp: -15% traffic. +2,7% conversion rate, -13% orders, -9,1% revenue, -13% profit, -40% acquisition (new customers) and -31% reactivated customers. The result: 30% overvaluation on ROAS (return on ad spend) or 8% overvaluation on POAS (profit on ad spend).
The long term impact
The impact it had on new users and reactivated users was the biggest insight and most concerning. Why? When diving into the numbers more closely, there showed to be less loss on revenue with loyal customers (-12%) in comparison to other customers (-24%). In addition, when looking at the long term impact and calculating the ‘missed future value’ of customers not acquired in the test period by analysing the performance of the active customer base, there turned out to be linear growth in number of orders and revenue generated year after year by those loyal customers. So, not requiring a customer short term is long term loss because new customers cán become loyal customers when attended to.
(long term impact = loss of new customers + loss of reactivated customers * customer value the next year)
The results are of course dependent on the attribution model used. Wehkamp used the non-direct last click model with a range of 7 days. Here the incremental ROAS after the experiment was -30%. When using the MTA (multiple touch attribution) model the incremental ROAS was +2%. This shows the impact the use of an attribution model has on the results and also how important it is to decide on: what is the ‘truth’ for your company?
– Incrementality experiments help you to measure the value of paid search and increase the effectiveness of marketing budget. It also helps having the conversation with higher management.
– Geo experiments are a great way to accomplish this.
– You want to get as close to the truth as possible, so make decisions on what is the truth for your company? The more detailed the metrics, the better: prefer POAS over ROAS for example and use a multi touch attribution model (MTA) instead of a last-click model. Also, try to use as much first party data as possible.
– The value of customers increases overtime. Loyal customers are brand ambassadors. Although less impacted by switched off advertising, do not stop advertising on loyal customers. You don’t want them moving on to a competitor.
– No company is the same and no experiment results will be the same. Start testing yourself and find out: what is the value of paid search for you?