Schibsted has seen a large increase in the number of digital subscribers over the few last years. As the focus on campaigns increases, Schibsted media publishers get new subscribers who enter on trial periods over a short period. In fact, many of subscribers never end up paying full price for a subscription.
We had to define some rules for orders from trial periods and campaigns to exclude unprofitable subscribers. Based on the defined rules, we started a process with the intention of developing a model to reduce repeat purchase behaviour. The model was in production for all our media publishers in 2019.
These are the three main reasons why we wanted to develop a model for repeat purchase behaviour:
- We needed to reduce the number of subscribers shifting from one trial to another, where profitability ends up being unacceptably low.
- We wanted to make sure former subscribers understood that trial and campaign periods only apply to new subscribers.
- We wanted to use price as a factor to build a media publisher brand perceived as high-quality journalism.
The overall goal with the model was not to increase income from a short-term perspective. The model was developed with a long-term perspective and ambition to build a strong value proposition and profitability per subscriber over a longer period of time.
Based on this, we made the following hypothesis: Can we use communication over a longer period to convince buyers to buy and stay on a subscription at a higher price than a reduced trial and campaign price?
The model is based on two variables: one value parameter and one score parameter, which is combined with a discount system. The value parameter is the value set on a trial or campaign offer and is based on the discount percentage on a trial or campaign measured against the full subscription price.
The score parameter is given to subscribers when they buy into a specific trial or campaign and is active for 180 days. This means every subscriber who bought on a trial or campaign in the last 180 days has an active score.
The model has been active since 2019, and 50% of full price for a subscription has been used as the alternative to a trial and campaign price. Subscribers with a score higher than allowed for a specific campaign or trial received tailored communication and a sales promotion with 50% off, explaining they are not allowed to buy the offer because they recently received a discounted price.
To illustrate, here is an example:
The trial offer for a digital subscription cost 1 NOK for 30 days. This is equal to a 99.6% discount since the full price is 249 NOK for 30 days. When a subscriber buys into this offer, they receive the model’s highest score because the discount is 99.6%.
The value parameter is always constant and based on the discount measured against the full price for a subscription. The discount always determines what value is set for a trial or campaign. However, the score parameter can easily be adjusted up and down.
For example, we could set a campaign score to 5. This means subscribers with a score equal to or higher than 5 are redirected to the alternative offer with 50% off full price. The possibilities for adjustment allow us to exclude based on what strategy and goals each media publisher has for a specific campaign.
Analysis and results
From analysis we know subscribers shown the model offer have, on average, five subscriptions over the last three years. Therefore, when comparing churn rates, we need to compare the model offer against the 1 NOK for 30 days trial with the same number of subscriptions and years.
The model has been used on both sales promotions and article sales (i.e the subscriber buys a subscription to read one article behind a paywall). The sales promotions is an offer visible on the front page, usually at the top of the site throughout the year.
Before activating the model, the sales poster had a conversion rate of 60%. After the model was implemented, the corresponding figure dropped to 13%. By exposing former trial subscribers to the alternative offer, we received significantly lower sales at a higher price. When comparing the profitability of sales from 1 NOK against 50% off full price, we expect to earn more money with the 50% off full price offer over the course of one to two years.
However, since results show sales are lower on 50% of full price, expected income is reduced. As a result, profitability from the 50% of full price offer is only 43% of the 1 NOK offer. The reduction of sales on the 50% of full price offer gives us a much lower profitability, although profitability per sale is higher at 50% of full price.
For large campaigns, we see the same trend as for the 1 NOK trial offer. Introducing the model explains a big drop in sales on the Web when comparing 2018 and 2019. This figure shows how many subscribers dropped off after being exposed to the alternative offer from the model.
The number of active subscribers after first renewal is lower if the subscribers have been on multiple trials or campaigns with a discount. Even though the renewal rate is lower for this segment, we still have subscribers in this segment who end up as loyal, full-paying subscribers. Since the model is not able to identify which subscribers buy several times on a discount and then become loyal, and subscribers who never pay full price, both groups are shown the alternative offer.
Thus, the economy in implementing the model is not as desired right now, but it should not surprise anyone. Moreover, it was never an expected the alternative offer would give us more subscribers than the 1 NOK trial or campaign offers. That is not why the model was implemented. The goal was to change how the media publishers’ subscription offer is perceived by the subscriber base. One step at the time, we are going from a discount mindset to creating loyal, full price-paying subscribers.
Conclusions, limitations, and future research
We are not currently able to fulfill our hypothesis. There are several reasons why, but we need to highlight some of the most important gaps.
- It is difficult to optimise volume and balance the model logic in large campaigns.
- Model regulation varies too much in campaigns for media publishers, which has led to difficulties in measuring results.
- It has been challenging to detect errors due to model complexity.
In 2020, we need to dig deeper into the identified gaps. We also realised it is critical to evaluate the price point and subscription history when we want to convince former trial and campaign subscribers to buy at higher prices. It is critical to find out how we can maximise the positive effect per subscription type and among what type of segments. Furthermore, we need to simplify the model and improve the analysis.
Based on the gaps and findings, we are now looking into an improved model with a more flexible structure.