WTF is a chief AI officer, with The Washington Post’s Sam Han

For The Washington Post, that reason why came when AI went from being a tool for certain parts of the company to becoming a critical component of the news publisher’s business and operations. “That transition point is where the [chief] AI officer’s needed,” said Sam Han, who was appointed the Post’s chief AI officer in June.

“We want to reach 200 million paying customers. How can you use AI to improve user engagement by providing AI-powered products? How can you use [AI to] help newsroom and businesses to improve their operations? And then at the same time, how can you use AI to optimize subscription revenue and advertisement revenue? That’s the core mission of [chief] AI officer,” Han said.

In a live recording of the Digiday Podcast during the Digiday Publishing Summit in Miami last week, Han explained how his new role is supporting parts of the Post’s business, such as in driving subscriptions through its dynamic paywall. Case in point: the Post tested an AI-powered paywall against its traditional paywall and saw the AI-powered paywall deliver a 20% increase in customer lifetime value, according to Han.

“The best benefit of having that is no more meetings to decide what rules [to set as the basis for the paywall],” Han said. “Thanksgiving sale’s coming up, how should we change the paywall rules to accommodate more subscription during the sales price? We will have meetings after meetings after meetings, and then the executive with the largest voice will win, right? That’s all gone now. It’s all driven by machine learning.”

Here are a few highlights from the conversation, which have been edited for length and clarity.

AI-powered dynamic paywall

We built a machine learning model called reinforcement learning and analyze a customer’s history in terms of engagement: how many times this customer saw free articles, how many times that customer saw the paywall [or registration] wall, [a person’s subscription churn history, an article value’s, etc.]. We put all these factors into consideration. And then the machine learning algorithm makes a decision for this customer.

AI-powered subscription offers

Our subscription marketing team created flexible access products, [such as] weekly pass, daily pass. We are experimenting with pay-per-article now. We want to bring all those into the paywall decision. It’s not just showing a paywall. It’s a paywall containing which products.

AI-related rules

The newsroom had its own AI policy, like human in the loop, but we didn’t have guidelines for the overall organization. We have our own language model, chatbot version [for] internal use. We also have enterprise ChatGPT version available. And we have Google Docs, which also has Gemini in there. And then individual reporters or engineers have their private or personal accounts. 

We have very strict guidelines saying, if it’s business- critical, sensitive information, use language model hosted by us internally. Otherwise you can use enterprise version of ChatGPT and so on. If documents are in Google Docs already, it’s already safe, we have correct classification of document and so on, so then you’re allowed to use Gemini to do certain tasks. 

AI in the org chart

I work as head of AI under [the] CTO organization. So it became a natural transition to sit under CTO, and I work very closely with the CTO and great support from [the] CTO office. So I think that was the good fit for our organization to have it under there.