In my first newsletter in 2023, I will discuss subscription analytics with a Berkeley marketing professor and illustrate prompting generative AI tools with examples from New York Times offers.

If you have questions or suggestions, e-mail me today at greg.piechota@inma.org or meet me in person in Stockholm in March at the INMA Media Subscriptions Summit.

In subscription analytics, start with small data, Excel, and don’t leave it all to algorithms

Mapping the customer journey, tracking lifetime value, personalising offers, and balancing data and human judgment are the table stakes in subscription marketing analytics, said Professor Przemyslaw Jeziorski of the University of California/Berkeley in an interview with INMA.

As data became increasingly important to understand news media subscribers and improve profitability, Professor Jeziorski urged media companies to start small.

At a recent quarterly meet-up, he told executives of 160 news brands benchmarking with INMA: “Data collection is easy and cheap, so many companies build lakes and fill them with trash. The problem with this is that data retrieval and analysis is costly.”

He recommended marketers start with the data they already have and plan analyses based on the customer lifecycle, which includes an acquisition stage, development stage, and retention stage.

“While everybody is excited about data lakes, machine learning algorithms, and real-time analytics, you can actually do the key analyses in Excel.”

Map the customer journey: To effectively use data, advised Professor Jeziorski, it’s important to map the customer journey, understand how people make decisions at each stage, and start using data to tailor products and offers to individual customers.

“I would start with retention analytics and identifying customers at risk of leaving,” Professor Jeziorski said, explaining his reasoning: 

• Most likely you already have all the data you need, such as each subscriber’s transactional history and insight about her Web and app reading. 

• The mathematics of predicting who is at risk of churning is simple because there are only two possible outcomes: a reader retains or does not.

• The project is cheap because you can run a basic logistic regression algorithm even in Excel.

• The insight can be easily applied to business: Send a discounted retention offer to a subset of at-risk subscribers, and measure a difference in retention rate vs. a group that has not received a discount.

Track customer lifetime value: “As media companies change their business model from mass-distributed goods to services, they need to measure the value of individual customer relationships,” Professor Jeziorski said.

Lifetime value is a simple mathematical model that gives companies one number to summarise the profitability of a customer and compare it to acquisition costs. It takes into account things like churn rate.

“Let’s say your average customer lifetime value is lower than your acquisition cost. Why is it?” asked Professor Jeziorski as he considered possibilities: “Is it because you are spending too much on promotion, charging too little for the subscription, or people are churning too quickly?”

By calculating lifetime value, companies can do back-of-the-envelope scenarios to understand the potential impact of changes, such as a decrease in churn rate: “How much would lifetime value increase if we cut our churn rate by half?”

Lifetime value can be calculated for any customer, including old and new customers, and can be used to make decisions about investments in marketing, product, or data analytics. 

Personalise offers: In digital subscriptions, there are various paywall models that target potential subscribers based on criteria such as Web site usage or content type. However, targeting can also be based on individual motivations and preferences. 

“Let’s take a complex product such as The Wall Street Journal that delivers all sorts of benefits to different people. Some sign up for the latest news. Some others want to support quality journalism. Think of these two benefits as two products,” Professor Jeziorski said. 

Applying predictive modelling, we could not only decide whom to target with our acquisition or retention offer but also which benefit to highlight to whom. These offers can convert at higher rates than traditional one-size-fits-all messages.

“Another benefit of personalisation is reducing waste of customers’ attention. Although the monetary cost of sending e-mails is low, the mental cost for the recipients remains high, and people who receive irrelevant offers might ignore them altogether.”

Personalised prices can also be effective in increasing customer lifetime value. Many publishers differentiate introductory discounts or retention prices depending on customer behaviour (e.g., the lower engagement, the lower price). 

Professor Jeziorski warned publishers to consider the potential negative effects on customer trust and perceptions of fairness.

Balance data-driven vs. data-informed decisions: For Professor Jeziorski, data analytics is a support system rather than a decision maker. 

For example, when talking about reengagement campaigns for lapsed readers, Professor Jeziorski suggested using a combination of algorithmic recommendations and curated picks. 

Churn modelling often leads to engagement as the biggest driver for retention. The problem is we know little about disengaged readers, our data is scarce, so it is difficult for algorithms to recommend content most likely to get them back to reading.

“Use the limited data to choose the topic or the section the reader engaged with in the past, but ask editors to pick the best articles from those sections for the final recommendation,” proposed Professor Jeziorski.

“Data is helpful at making small decisions, but human judgement is necessary for strategic decisions and decisions where we face the unknown,” he added. 

Interested in more insights about subscription analytics? Join 160 brands from 33 countries that benchmark their subscription performance with INMA.

Example prompts to analyse, iterate the best-in-class subscription marketing assets with generative AI

New generative AI tools, such as ChatGPT, make it easier for other publishers to study, copy, iterate, and learn from the expertise of the top selling news subscription brands.

With 10.75 million paid subscriptions across print and digital, The New York Times is the world-class subscription marketer. No wonder many — as executives often confide — study and copy what The Times is doing.

New generative AI tools promise to make the marketers’ work more effective and efficient. ChatGPT, Dall-E, and others can produce text, images, audio, and video, and they are as easy to use as mobile messengers. 

Literally, if you know how to write a text to a friend, you know how to use ChatGPT, a prototype AI chatbot developed by OpenAI. 

Prompting AI tools to produce quality marketing assets is a new skill combining art and science. Here are some examples:

Prompt: Act as … a marketer of a digital subscription at The New York Times. Here is a copy of its subscription landing page (for Europe): “Special offer. Unlimited access to all the journalism we offer. €0.50/week (Billed as €2 every 4 weeks for your first year). €20.00/year (Billed as €20 for your first year). Subscribe now. Cancel or pause anytime.” Can you analyse this offer?

Hint: Focus your conversation ChatGPT on a specific subject and use the relevant language. The opening question helps set the playing field for the conversation and boundaries for responses. You can also make the bot pretend to be a person of your choice, and this results in answers written from the perspective of that person. 

Example response from ChatGPT: “As a marketer of a digital subscription to The New York Times, here are some points to consider regarding the offer text:

• The offer includes “unlimited access” to all of the journalism offered by The New York Times, which can be appealing to readers who want to be able to access all of the content on the site without any restrictions.

• The offer is presented in two different formats: a weekly rate of €0.50 and an annual rate of €20.00. This can allow readers to choose the option that best fits their budget and reading habits.

• The offer includes the option to “cancel or pause anytime,” which can be reassuring to readers who may be hesitant to commit to a long-term subscription.

• The offer includes a discounted rate for the first year of the subscription, which can be a strong incentive for readers to sign up. However, it’s important to note that the rate will increase after the first year, so it may be helpful to include this information in the offer text.

Overall, the offer text clearly communicates the benefits of the subscription and provides readers with multiple options for signing up. However, it may be helpful to provide more information about the rate increase after the first year to give readers a more complete understanding of the terms of the offer.”

Prompt: Create … a new version of this offer. Focus this time on The Times’ quality journalism. Write it in the style and length that I showed you above.

Hint: After setting the playing field, follow up and discover that ChatGPT remembers your previous prompts. It can also create new text from scratch or revise the text you entered. Set the boundaries for the style and length. And if the answer is not satisfactory, click “regenerate response.”

Example response: “Special offer. Unlimited access to award-winning journalism. (...) Subscribe now. Cancel or pause anytime.”

Prompt: Brainstorm … 20 more examples. Each should emphasise different benefits, such as the mission to search for truth, verifying information, witnessing events, investigating facts, and sense-making. Please write in the same style as in the first example.

Hint: ChatGPT is a master of remixes. When you find a satisfactory response, provide feedback to the bot (a simple “I like it” will suffice). And ask for more iterations. The chatbot can create endless lists within the boundaries you set, based on your tips or beyond them.

Example responses: “Join the quest for truth with The New York Times … Be part of the conversation … Get the whole story … Experience history … Make a difference … Be an advocate for justice …”

Prompt: Suggest a prompt … for an illustration for this offer’s landing page. Feature a reporter doing an interview. Make it playful and draw it in a cartoon style.

Hint: Use ChatGPT to suggest prompts for other generative AI tools, such as image-creating DALL-E by OpenAI. In general, the more specific and detailed prompt, the richer the image generated. 

What prompts did you find useful? Let’s create an INMA AI phrasebook. E-mail me at greg.piechota@inma.org.

Interested in more examples of generative AI use in subscription marketing? Join the Readers First Initiative’s Webinar on January 18. Register now.

About this newsletter

Today’s newsletter is written by Grzegorz “Greg” Piechota, INMA’s researcher-in-residence and lead for the Readers First Initiative. In his letters, Greg shares original research, analysis, and best practices in growing reader revenue.

E-mail Greg at greg.piechota@inma.org, message him on Slack, meet him at the next online meet-up or in person at the INMA Media Subscription Summit in March.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .