By: Xavier van Leeuwe, Matthijs van de Peppel, and Matt Lindsay
Director of Marketing and Data, Manager of the Data Intelligence and Customer Relationship Management Team, and President
NRC Media and Mather Economics
Subscription-based businesses like those in the news industry have a significant advantage over other companies in regard to the number of times they interact with their customers. They can observe a customer’s behaviour over time and adjust the nature of their services to maximise the lifetime value of that relationship.
One of the ways they can adjust their relationship with a customer is through the pricing of their product or service. This is particularly true when it is time for that customer to renew his or her service.
Publishers have a lot of demographic data on their subscribers, but they are not using the data in the audience department. The customer demographic data is used exclusively to market the audience to advertising customers. Publishers can show advertisers the demographic profile of their readers, the geographic distribution by ZIP code, and the overlap between their readership and advertisers’ customers.
Publishers are not using this data to understand which of their customers are most likely to cancel after a price increase, churn for another reason, or upgrade a subscription.
In 2002, Matt Lindsay had recently left Arthur Andersen Business Consulting following the Enron paper-shredding controversy. He was working as an independent consultant for Knight Ridder, the second largest newspaper publishing company at the time. The company hired him to analyse its subscription pricing strategy following several years of declining audience volumes and revenue, despite aggressive discounting of subscription prices.
“We had a Harvard MBA work on this, but he did not find anything,” they told him. “We do not think you will either, but we are willing to give you a try.”
Knight Ridder sent data from the subscription system at one of its newspapers, the St. Paul Pioneer Press, to begin the analysis. When this data was plotted as a retention chart, it was clear that retention behaviour of subscribers was remarkably similar to data
Lindsay had seen on patient lifespans in healthcare economics in graduate school. The similarity brought to mind survival analysis as a possible approach for modelling subscriber retention and price elasticity.
Survival analysis is the field of statistics used to analyse patient data. This same type of econometrics can be used to predict machinery failures and comparable events in other fields.
The goal of survival models in healthcare is to understand what factors affect the lifespan of a patient. In applying this analytical approach to the publishing industry, we were interested in predicting how long a particular subscriber would remain active as a customer. The “failure event” to be predicted in healthcare economics is a patient’s death. In models for publishers, it was a cancelled subscription.
Using survival analysis, it was clear price increases played a role in a subscriber’s likelihood of stopping. However, the effect of price was very different across different segments of the customer base. Certain groups of customers were 20 times more likely than others to stop after a price increase.
It quickly became clear that a one-price model was not the best strategy for publishers. We needed to protect those vulnerable customers from a price increase. Much like other industries that had been through significant change and disruption, such as airlines and hotels, publishing could use sophisticated pricing analytics to increase revenue and operating margins.
Our first few tests of this insight involved targeted pricing adjustments for groups of subscribers who appeared to be paying much less than they would be willing to pay for the product. It was not uncommon for publishers to have below-optimal subscription prices due to the value incremental subscribers brought to the advertising revenue stream. Indeed, the first tests demonstrated some customers showed very little reaction to price increases, while others had significant reactions.
Eventually, we developed a recurring process at Mather Economics where subscriber data was sent by our clients every week, and we returned suggested renewal prices for each account. Statistically representative control groups were used to measure the effect of prices on retention and revenue. The industry gradually accepted that this type of pricing was beneficial for both the customers and the publishers.
We now perform this kind of analysis for about 500 publications in 12 countries on four continents. These clients provide us with subscriber data on about 30 million households every week. We call it market-based pricing.
You may wonder if charging different prices for the same product works.
It really depends on local customs, and you need to understand what feels right for the organisation and for your customers. Some compare this differentiated pricing to progressive taxation. We do not expect everyone to pay the same amount of tax every year, because the tax is determined by income levels and certain life circumstances.
Similarly, a US$500-per-year newspaper subscription is a much greater percentage of disposable income to a school teacher, on average, than it is to an investment banker. Arguably, investigative journalism is a public good that benefits everyone once it is produced. Differentiated pricing could be seen as benefiting society as a whole due to greater support for journalism and greater access to independent news.
Many of our big-picture findings about subscriber retention are intuitive. As expected, the more a subscriber’s renewal price is increased, the more likely it is that customer will cancel the subscription.
But that fact hides important distinctions between individual subscribers. Price sensitivity varies considerably across a publisher’s subscriber base, and our detailed knowledge of the differences can be used to minimise customer cancellations due to a price increase. This type of pricing strategy can reduce cancellations due to a price increase by up to 75%.
As we suspected, customers with higher incomes are less sensitive to price increases than those with lower incomes. At the end of the four years, for the average subscription cohort, there are about twice as many high-income subscribers remaining as low-income subscribers. This result would suggest that income is an important variable to consider in pricing and retention strategies.
A further disaggregation of the retention data to isolate subscribers acquired by one channel (in this case, inserts) shows the effect of income on retention is much lower within a single acquisition channel. While high-income customers still have higher retention, the variation across income levels is smaller, particularly in the first few years of the customer lifecycle.
If we plot retention across acquisition channels for the high-income group only, we see that there is considerably more variation in retention across channels within the income tiers than we saw across income levels. This insight suggests the nature of the acquisition offer and method of acquisition play an important role in retention — perhaps more so than income levels.
These are but two of the factors affecting retention that we include in our survival models. Other variables include subscription term, payment method, seasonal patterns, demographic variables, and macroeconomic indicators.
One advantage of regression modeling (such as survival analysis) is that it is able to measure the effects of each important factor in isolation.
Visualisation tools are very helpful in uncovering relationships within the data and presenting results to a broad audience. Complementing regression modelling with A/B testing, performance measurement, and reporting is the best way to establish a closed-loop pricing and retention optimisation process where insights from prior price changes are incorporated into future changes. This improves the efficiency and performance of the pricing strategy.
With digital subscriptions, the number of variables that can be included in survival models increases substantially. A number of predictive metrics are consistently important in our models of retention of digital subscribers. We can classify these metrics into consumption, interaction, attitudinal, time, and socioeconomic categories, although there are other metrics that do not fall into these groups.
Consumption metrics describe the quantity, frequency, and time spent with the content during a particular period of time. Interaction metrics describe actions taken by customers while on the site or while they are engaged with the content. Attitudinal metrics are those measuring the level of an individual’s enthusiasm for or loyalty to a topic or community.
Time metrics reflect when events occurred during a subscriber’s lifecycle, and the overall time of activity on the account, often called the account tenure. Socio-economic metrics include factors characterising an individual’s demand for the product, such as disposable income, price sensitivity, age, gender, macroeconomic indicators, education, and household type.
Using survival analysis and other types of customer analytics, we have made adjustments to retention strategies, operational processes, and prices. We found data-supported actions can reduce churn by 15% in the first few weeks, and the performance usually gets better as publishers learn from their initial efforts. Pricing strategies can reduce price-related customer losses by as much as 75%.
Given the changes to the advertising industry, it is likely the majority of revenue for publishers will come directly from their audiences. Using data and analytics to manage customer acquisition, retention, and pricing strategies as well as possible will make the difference between the publishers that survive and those that do not.
It is important to note that analytics on customer price elasticity are only the first step in an effective subscription yield management process. There are necessary technical implementation steps within the billing and customer relationship management systems.
It is imperative testing and reporting processes be in place so the effectiveness of the pricing analytics can be monitored, and insights from the pricing changes can be incorporated into future yield management decisions.
An often overlooked but critical element to a successful yield management programme is a thoughtful strategy for customer communication and messaging of the price changes. This includes customer service scripts for responding to questions from subscribers.
We have found messaging and customer communication are very important to the ultimate performance of the programme, because they have such a significant effect on the bottom-line yield from the price changes.
For full article, click here.