TikTok U.S., Disney-Kimmel + WTF is a chief AI officer with The Washington Post’s Sam Han

Summary

---

The Digiday Podcast

Episode: "TikTok U.S., Disney-Kimmel + WTF is a chief AI officer?"

Featuring The Washington Post’s Sam Han

Date: September 23, 2025
Hosts: Kamiko McCoy & Tim Peterson

---

Overview

This episode dives into three of the week's most pressing media stories: the future of TikTok in the U.S. amid regulatory changes, Disney’s suspension—and rapid reinstatement—of the Jimmy Kimmel Show, and mounting questions over the reality of "state run" or "state-directed" media. The headline segment features an in-depth conversation with Sam Han, The Washington Post’s new Chief AI Officer, about how publishers are harnessing AI not just to survive, but to optimize revenue, streamline newsrooms, and drive long-term growth.

---

Key Discussion Points & Insights

1. The State of Media: A Challenging Landscape

• DPS Takeaways (00:26–01:26): Tim reflects on the Digital Publishing Summit, describing the mood as one of camaraderie amid challenge:

  "It was not incredibly encouraging... there's a lot of conversation about, like, all the challenges media companies are facing... I think everyone's managing expectations." (Tim Peterson, 00:29)

• No silver bullet solutions, but a sense of being "in the same boat" provides some comfort for struggling publishers.

---

2. TikTok U.S.: The Ownership Shakeup & Algorithm Mystery

Segment Start: 04:15

• Deal Structure & Ownership:
  • TikTok U.S. is forming, with Oracle and Silver Lake as primary American stakeholders. Fox may enter as well.
  • Critically, the U.S. government will not own equity, but will supervise the algorithm.
  • Users will not need to download a new app. The algorithm, TikTok's "secret sauce," will be licensed from ByteDance and re-supervised stateside. (04:44–06:54)
• Concerns of Government Oversight:
  • Tim voices skepticism about state supervision, drawing parallels to state-run media:

    "This is, I feel like the most important detail... the US government is going to have supervision over the algorithm... I don't know that I want to be using TikTok US if this government that we have... is going to have that kind of hand..." (Tim Peterson, 06:54)

  • Editorial fear: Will TikTok become bland or politically skewed, like X (formerly Twitter) or Truth Social?

    "Does TikTok just become like the QVC network and everything plays it safe...?" (Kamiko McCoy, 08:11) "If the US government is going to... supervise the TikTok US algorithm, it's going to be able to put a finger on the scale" (Tim Peterson, 09:46)

• Memorable Moment:

  "It's going to have a finger on the scale when it comes to the content on TikTok US." (Tim Peterson, 10:18)

---

3. Disney’s Kimmel Dilemma & State-Directed Media

Segment Start: 10:18

• Background:
  • Disney suspends Jimmy Kimmel’s show after backlash over remarks made following the killing of Charlie Kirk and resulting pressure from the FCC and White House.
  • The suspension was reversed hours after recording, but hosts analyze the deeper implications regardless. (11:11)
• Government Pressure & Free Speech Concerns:

  "The fact that you have an FCC commissioner dictating programming, dictating content decisions, seemingly violating free speech... like we're coming after free speech." (Tim Peterson, 12:30)

• Business Motivations:
  • Disney, seeking government approval for key M&A deals (esp. ESPN/NFL), may be quick to appease regulators.
  • Comparisons to earlier mergers and concessions (Paramount/Skydance, Omnicom/IPG).
• State-Directed vs. State-Run:

  "We're not yet seeing state run media like it seems like we may be seeing with TikTok US, but we're definitely seeing state directed media because Disney’s effectively taking a directive from the government..." (Tim Peterson, 14:50)

• Advertiser Reaction:
  • Brands' stances generally track with business interests; past moments of principle (post-George Floyd) largely faded.

    "Advertisers are ultimately in the business of selling products and so it's the business of capitalism, not so much politics." (Tim Peterson, 16:45)

---

4. Content Licensing: Fresh Deals & AI Tension

Segment Start: 18:21

• Meta is showing signs of rekindling relationships with publishers, pitching content licensing deals (notably with Axel Springer, Fox, News Corp).
• Deal Shifts:
  • The movement is from "training deals" (lump sum for data access) toward usage-based compensation:

    "What publishers want to move to is usage based pricing... if an article is getting cited by Perplexity or by Gemini or by chatgpt... publishers want to get paid for that." (Tim Peterson, 19:43)

  • Reddit is a notable leader, seeking usage-based deals with Google and OpenAI.
• Legal Front:
  • Publishers are contesting content scraping and AI-generated summaries in court (NYT v. OpenAI, Britannica v. Perplexity, German publishers v. Google).
  • AI Overviews (Google’s search summaries) are blamed for decreasing publisher traffic.

    "The decline from Google is not complemented by OpenAI or Perplexity kind of traffic." (Unnamed publisher via Tim Peterson, 24:21)

---

5. Featured Interview: Sam Han, Chief AI Officer, The Washington Post

Segment Start: 26:28

Setting the Stage

• Origin of the Role:
  • The technology’s centrality spurred the need for a dedicated executive; previously, AI was siloed in engineering/product.
  • "About 10, 11 years ago, I started machine learning team there... focus was more like creating a personalization experience, recommendation models... [now] with the language model ChatGPT kind of explosion, we start seeing differences in the adoption of AI." (Sam Han, 26:47–28:25)
• AI Hub → CAIO:
  • Early experiments included AI-written sports/election stories (Heliograph), comment moderation, dynamic paywalls.
  • Recent years saw company-wide projects, requiring coordination, standards, and now, executive leadership.
  • "This year we realized that we need a role to really overall and then in a way that’s more like recognition of the need, but also announcement... saying AI is critical to our business." (Sam Han, 30:59)

Mission and Metrics

• CAIO Focus Areas:
  • Driving business growth through AI: subscriptions, engagement, advertising.
  • Empowering newsroom and other departments via operational AI.
  • Productizing for third parties via the Arc XP CMS platform.
• Direct Impact:
  • Paywall optimization with reinforcement learning:
    • Dynamically determines if and what kind of paywall to show.
    • Weighs user history, content value, ad potential.
    • "We did AB test... and we saw like 20% improvement in terms of overall customer lifetime value using that model." (Sam Han, 35:45)
    • Key benefit: ends "meetings after meetings" about paywall rules; now data-driven and flexible for special events (elections, sales).

AI in the Newsroom & Beyond

• AI as 'Team Sport':
  • Nearly every team can use AI, requiring standards and collaboration.
  • The challenge: avoiding duplicated efforts, ensuring oversight, and maintaining quality.
• Staffing & Talent Challenges:
  • Competing with Big Tech for AI talent is hard; mission and impact are key recruitment drivers.
  • "We offer... not by salary or compensation, but by mission. We make sure that we provide exciting work... you see your work implemented, go to production and showing value..." (Sam Han, 43:03)
  • Current AI team: ~16 machine learning scientists, ~6-7 data engineers, ~10 software engineers.
• AI Governance:
  • Han now sets policy for AI use across the Post: safe handling of data, selection of models/tools, and compliance with guidelines.

Notable Quotes & Insights

• On evolving the Chief AI Officer role:

  "It was kind of organic transition. But however, the mission for a chief AI officer is to use harness the power of AI to drive business growth, innovation and revenue optimization." (Sam Han, 31:59)

• On dynamic paywalls and reinforcement learning:

  "The best benefit of having that is no more meetings to decide what rule to do." (Sam Han, 35:45)
  "Now it’s all driven by machine learning."

• On talent strategy:

  "The advantage we offer is... you work on a project, you see that going to production, you work with engineers and you see the value... But it's very challenging." (Sam Han, 43:03)

---

Memorable Moments & Quotes (with Timestamps)

• "If the US government is going to be able to supervise the TikTok US algorithm, it's going to be able to put a finger on the scale to say... this is the content that needs to get shown to everyone." (Tim Peterson, 09:46)
• "We're not yet seeing state run media... but we're definitely seeing state directed media because Disney’s effectively taking a directive from the government, from the FCC with this decision." (Tim Peterson, 14:50)
• "What publishers want to move to is usage-based pricing... so it's the publishers being able to get paid for the value they're actually delivering and make sure that it's an equitable exchange." (Tim Peterson, 19:43)
• "We did AB test... and we saw like 20% improvement in terms of overall customer lifetime value using that model." (Sam Han, 35:45)
• "The best benefit of having that is no more meetings to decide what rule to do... Now it’s all driven by machine learning." (Sam Han, 35:45)

---

Key Timestamps

• State of Digital Publishing Summit Recap – 00:26–03:23
• TikTok U.S. Deal Deep Dive – 04:15–10:18
• Disney-Kimmel/State-Directed Media – 10:18–18:02
• Meta Content Licensing/Publisher AI Battles – 18:21–26:09
• Featured Interview: Sam Han, Chief AI Officer – 26:28–47:06

---

Conclusion

This episode provides a vivid, insider view of how media organizations are responding to existential issues: government intrusion, platform volatility, and the relentless advance of AI. The conversation with Sam Han is both technical and practical, demonstrating that for top publishers, AI is not about future promise—it's a day-to-day operational necessity, with measurable impact on both subscriptions and advertising. The episode closes on an optimistic note: AI, in the right hands, can restore some control to publishers, if only they evolve fast enough.

---

Transcript

[Interviewer]

Foreign.

[Kamiko McCoy]

Hello. Hello and welcome to the Digiday Podcast, a show about the business of media and marketing. I'm Kamiko McCoy, senior marketing reporter here at Digiday.

[Tim Peterson]

And I'm Tim Peterson, executive editor of video and audio Digiday Media. What's up, Kimiko?

[Kamiko McCoy]

Tim, welcome back.

[Tim Peterson]

Thank you.

[Kamiko McCoy]

Glad to have you back. How was dps?

[Tim Peterson]

It was good. It was not incredibly encouraging of being in the media business just because there's a lot of conversation about, like, all the challenges media companies are facing and like media companies are always facing challenges. I don't know that we got to solutions as much as I would have hoped. At the same time, like, I think everyone's managing expectations in terms of like, it's unclear to what extent like, solutions are out there. I think it was more the sense that I got was the most encouraging part for everyone. There was just hearing that kind of all publishers are in the same boat. It's not a, it's not a great boat. But, you know, there's something to be said for camaraderie. Like, what's the, the thing about friends and foxholes?

[Kamiko McCoy]

Something about communal suffering.

[Tim Peterson]

Something like that. Has life been outside of Miami?

[Kamiko McCoy]

It's been good. It has been a busy, busy news week, as we'll get into later in this episode. But you actually had a live conversation from the Digital Publishing Summit where you got to talk to Sam Hahn, the chief A officer at the Washington Post, right?

[Tim Peterson]

Yeah, this is a fun conversation. You know, there was a lot of enjoyment to be had at the Digital Day Publishing Summit. I think it's more the remove from the event where it's just like, oh, how is the business doing? But, but yeah, this was. I really enjoyed this conversation. So chief A officer is like one of the newest titles in the industry. And so Sam was appointed chief A officer at the Washington Post back in June. And. And so this whole session is me basically being like, sam, what is your job? What does a chief AI officer do? And so that was really interesting to learn about. But also we got into really the nitty gritty in terms of the work that they're doing and working with different departments at the Washington Post around things like, for example, how they're working on driving subscriptions and experimenting with different AI powered paywalls or applying machine learning to the paywalls that the Washington Post has to be able to figure out what kind of subscription offers to provide to people or when to put a paywall up. So I thought it was a really interesting conversation. It was one of the conversations during DPS where it's just like, as much as AI is a threat to publishers, it's also a tool and that can be used to benefit publishers. And so this was one of the conversations we had during, during the week about how publishers are actually working to take advantage of AI to their benefit.

[Kamiko McCoy]

Well, that's hopeful and a glimmer of light that we'll need for this episode.

[Tim Peterson]

Absolutely. Yeah. Because it's on. The flight back from DBS was when the Kimmel news broke with Disney taking down the Jimmy Kimmel show after political pressure from the FCC and the White House. Also looming all over last week into this week has been the TikTok deal. What's happening with TikTok? And then it's weird that the light part of the week is, oh, hey, meta's gonna be, you know, is talking with publishers about doing content licensing deals, which is nice timing because a lot of the conversation during DPS was around content licensing. So we have a number of, you know, big stories to get into this week. Maybe should we start with TikTok?

[Kamiko McCoy]

Yeah. With no further ado, let's get into this. This week's juicy scoops with Tick Tock. I feel like I've maybe spoke too soon here. Why? I feel like I'm, we're starting to finally get some closure or at least I'm speaking way too soon, but at least a look of what the hell is happening with the, the overall framework of how this could all shake out. I know you're. I'm gonna ask you to explain how this shakes out. You've got the algorithm being divvied up. What the hell is happening?

[Tim Peterson]

There's a lot that's happening and all credit to the Wall Street Journal and Axios and Bloomberg and the New York Times that I've been referring to to stay up on like, okay, how is this actually transpiring? So from what I understand from all of the reporting that those outlets have done, like you said, the deal isn't done yet, but a framework is in place. I guess China still has to review the details. We're still kind of getting of in terms of like what this deal actually is. But what we know so far is Oracle and Silver Lake become part of this US ownership group of like TikTok US is I guess what we'll call it for now.

[Kamiko McCoy]

Okay.

[Tim Peterson]

Trump has said, oh, actually Fox, you know, maybe is going to be part of this too. So maybe Fox will become one of the owners of TikTok US. We'll see about that this morning Bloomberg reported the US government will not hold equity in TikTok. So TikTok US is going to have a lot of owners. The US government will not be one of them. I think there's some comfort in that in the government. We're going to talk a good amount about state run media in this episode. It seems like TikTok at least won't be state owned Media media, a state owned platform. Another thing we know people in the US won't need to download a new TikTok app. So there will be a transfer of data and effectively a new algorithm powering TikTok US. But that's not going to mean people are going to have to delete TikTok and install a new one. Which is a good lesson from. I think it was HBO Max when they rebranded to Max required people to install new app. I don't, I remember that just being.

[Kamiko McCoy]

Annoying for the algorithm itself because that's been a big contention point here. Right. Is that the China ByteDance did not want to hand over the algorithm, which is essentially TikTok secret sauce. So now do you have like a duplicate of the algorithm?

[Tim Peterson]

Yeah. So according to the Wall Street Journal, ByteDance is going to license a copy of of the TikTok algorithm. So they're going to duplicate the algorithm, license it to this TikTok US ownership. And so then Oracle and the US government are going to recreate the algorithm, are going to supervise. This is quoting from the Wall Street Journal. Under the supervision of Oracle and the US government. This is, I feel like the most important detail of this entire deal that the US government is going to have supervision over the algorithm that is going to be governing TikTok US. I don't know that I want to be using TikTok US if this government that we have, given the crackdown we're seeing across media is going to have that kind of hand in terms of how content is going to be recommended.

[Kamiko McCoy]

Well, not only that, I know that we said that it's not a state run media platform, but like if arose by any other name. Right. If not in name but like kind of in. Kind of in practice.

[Tim Peterson]

Yeah, that state owned platform. But it does seem like it's going to be at least partially state run platform.

[Kamiko McCoy]

Yeah, to your point, which is still spooky. Ookie. And just in time for Halloween season. Love that. A good jump. Scare for you. But also, I mean you got to think about creators here where I feel like this has been a big back and forth for them as well. Of how much? Because especially when you had like a lot of them that are like politically activists and using that as their platform and whatnot. Like, you know, I don't know, does TikTok just become like the QVC network and everything plays it safe and those type of creators flee? I don't know what happens to a state run to clarify a social media channel.

[Tim Peterson]

Does it become like how what we've seen happen with X where it's become even more tech bro, and seemingly right leaning. I mean, I've stopped using X, so that's just me going off of the reports I've seen and just kind of what the conversation around X has been. Is it going to be like Truth Social where it just gets very right leaning? Because we've seen this administration and others from I guess the conservative side of the aisle go after platforms like Meta alleging that there's political bias in these platforms that they're penalizing certain types of content. Well, if the US government is going to be able to supervise the TikTok US algorithm, it's going to be able to put a finger on the scale to say the this is the content that needs to get shown to everyone.

[Kamiko McCoy]

Right.

[Tim Peterson]

And on the other side of that thing, like here's the content we're actually going to suppress. So I feel like the TikTok as we know it is going away with this deal. I would love to think that I'm wrong, but just by virtue of what we've seen with the US government and its war against media and its war against free speech, it feels reasonable to look at all of this and think it's going to have a finger on the scale when it comes to the content on TikTok US.

[Kamiko McCoy]

Yeah, closure absolutely was not the right word there, but I can't think of a better one at this time. Staying down the path of state run media. For those of you that have been living under a rock, or maybe if you're off the grid, which in that case, congratulations. Jimmy Kimmel's late night show has been suspended indefinitely by abc. This comes after the backlash over Kimmel's comments following the killing of Charlie Kirk. You've also got President Trump's got a lawsuit which has now been dismissed against the New York Times citing defamation, defamation, excuse me, and claiming $15 million in damages. And then finally you've got the Paramount Skydance, which is preparing a bid for the Warner Brothers discovery. So all of these things happening simultaneously kind of flicks at what we were saying about state run media. But let's Talk about Kimmel first.

[Tim Peterson]

Hey, it's Tim. A few hours after KO and I recorded this, Disney announced that it's bringing back the Jimmy Kimmel Show. I think our discussion still holds up, though, so let's get back to it.

[Kamiko McCoy]

How disappointed are you that that yet another talk show, late night talk show host has gone off air?

[Tim Peterson]

I mean, I don't watch late night tv, so, like, that's something Tim doesn't watch. That doesn't get to me. But the fact that you have an FCC commissioner dictating programming, dictating content decisions, seemingly violating free speech, like we're coming after free speech. And also, like, one thing that I feel like hasn't been talked about enough is, like, what Jimmy Kimmel actually said that sparked this whole thing. And so I'm just going to quote it. This is from the New York Times coverage of this. And so their transcript quote, we hit some new lows over the weekend with the MAGA gang desperately trying to characterize this kid who murdered Charlie Kirk as anything other than one of them and doing everything they can to score political points from it. So. So it seems like it's that part of the line where he says trying to characterize this kid who murdered Charlie Kirk as anything other than one of them. Where FCC Commissioner Brendan Carr took that as and others on the right have taken that as like Kimmel saying that the person who murdered Charlie Kirk is part of this MAGA gang. Where it's my reading of this is he's just saying that they're using this person to then attach this person who murdered Charlie Kirk to, like every other group that they want to drag down. Could it have been written more clearly? Yes.

[Interviewer]

Language is important.

[Tim Peterson]

Words are important. We're in the business of words, you and I. But from this now we have the FCC commissioner threatening Disney and demanding Disney take Kimmel off the air. Disney then following suit and actually taking Kimmel off the air, which seems kind of like a panic move. And in some ways the worst thing Disney could have done here because then it looks like Disney bowing to the FCC and to the Trump administration. It also isn't super surprising because when I saw Disney did this, my immediate thought was like, well, yeah, Disney's trying to close this deal between the ESPN and the NFL where Disney or ESPN is going to acquire the NFL's media properties and the NFL is going to get a 10% stake in ESPN. But for that deal to close, Disney needs government approval. And so even when that deal was announced back in, I think it was August Immediately, Mind mind went to what we were already seeing with Paramount, Skydance, where Paramount had to bend over backwards to appease the Trump administration in order to get that deal closed. Also what we've seen with Omnicom ipg, where these two agencies want to merge or Omnicom to acquire ipg. And so they agreed to terms with the US Federal Trade Commission that they will not block advertiser spending because of political reasons. And so this again is Disney needs to get a deal done and so is willing to make a move like this, take Kimmel off the air to appease the government. So it's not, we're not yet seeing state run media like it seems like we may be seeing with TikTok US, but we're definitely seeing state directed media because Disney's effectively taking a directive from the government, from the FCC with this decision.

[Kamiko McCoy]

There are direct parallels that could be made between that and what happened with Stephen Colbert. And to your point, you know, these other deals that are being teed up to come down the line, which to me kind of leaves the question of like, who gets the final say on like what free speech is in terms of like state. How did you call it? State directed media.

[Tim Peterson]

Yeah, well, yeah, because that's the ridiculous thing is like these are people who are in Trump and Brendan Carr who are supposedly championing free speech, but then they're coming after Kimmel because they didn't like what he said, then that's not. Then you don't believe in free speech. Like, what are we doing here?

[Kamiko McCoy]

No. Which is a really, really big question. And then I also question kind of how advertisers fit into all of this as well in determining like how to spend their dollars and whatnot. And because to the point you're seeing a boycott from people right, on behalf of to Disney being like, oh, well, if you're going to take him off the air and you know, take away free speech or our interpretation of free speech, now we're not going to be buying Hulu Disney subscriptions, yada, yada yada. And there's questions to like, well, our advertisers going to do the same. Senior marketing reporter Sam Bradley and I have been like tinking around with a story around that and I've kind of come to the conclusion after our conversations is that like advertising advertisers have always been on the side of making money. If you consider things like the, you know, Omnicom IPG deal, Garm, that was disbanded a couple years ago after lawsuits from X and advertisers Right. How much space there is for advertisers to be like, well, in favor of free speech. We're going to support with our ad dollars in this way.

[Tim Peterson]

Yeah. I mean, advertisers are ultimately in the business of selling products and so it's the business of capitalism, not so much politics. So I don't, I don't disagree with what you said.

[Kamiko McCoy]

Direct about face from, you know, post, excuse me, George Floyd's murder.

[Tim Peterson]

But also like what we've been talking about, even when, you know, we were doing the episodes immediately after the election late last year and then what, you know, as we were talking about advertisers coming back to X, advertisers pulling back on diversity, equity and inclusion, like they kind of go whichever way the wind blows. And since last year's election, even leading up to last year's election, but certainly since the ones have gotten stronger towards appeasing Trump and what Trump wants. And advertisers, many advertisers, I don't want to say all because I would like to believe there are hopefully plenty of brands and agencies that are standing on their principles, whatever they are. But we've seen a number that have backtracked on their supposed principles because I think ultimately they're just looking to take care of their businesses and their bottom lines.

[Kamiko McCoy]

Yeah. And then the last point I'll make about this is that's why you're seeing the point of contention between Ben and Jerry's and Unilever. Unilever, excuse me, with their founder deciding to leave co founder Jerry of the Ben and Jerry's duo. On a lighter note.

[Tim Peterson]

Segue.

[Kamiko McCoy]

On a lighter note, you've got publisher companies that are starting to finally see what could be a deal working in their favor with some content licensing deals coming out of Meta potentially. Is that something that you guys touched.

[Tim Peterson]

On at dps, not Meta specifically. There was like some publishers talking about how they've seen a rise in traffic from Facebook this year. We've even seen that at Digiday and Sarah Waglioni has been reporting on that for I think more than a year. I think last year she was even reporting on publishers seeing an uptick in traffic from Facebook. And one of the suspicions, conspiracy theories that we've had with that is like, well, if Meadow wants to curry favor with publishers in order to get content licensing deals, because it seems like it seems like all these AI companies, companies need content from publishers. A good way to do that would be to start sending some traffic to publishers for meta to be able to say, see everyone, we're friends. You should do deals with us because we can be important to you again. Just forget everything that we did to you in 2018 and since 2018, before 2018 and messing with your traffic, messing with payments, messing with the reach of your content. Let's be friends again. According to the Wall Street, Meta is talking with Axel Springer, Fox and News Corp about some sort of content licensing deal, but we don't know what the model is. And so when we had Seriglioni, our senior media reporter, and Just Davies, our senior media editor on the show a couple weeks ago, we talked about how there's this shift going on with publishers in terms of how they want to get paid for their content from the AI companies. A lot of these content licensing deals have been what have been termed training deals. Basically AI company gives like a lump sum payment for access to a publisher's library of content and that's kind of the extent of it. And maybe that deal gets renewed after a few years or so, maybe it doesn't. What publishers want to move to is usage based pricing. And so if an article is getting cited by Perplexity or by Gemini or by chatgpt in results, publishers want to get paid for that, which makes a lot of sense because then it's the publishers being able to get paid for the value they're actually delivering and make sure that it's an equitable exchange. Unclear if that's the path Meta's going down, but that's definitely the path publishers want to be going down. I did a interview on stage at DPS with Karen Saltzer, who's the CEO of Bloomberg Media, and I opened up by asking her, Bloomberg hasn't done a content licensing deal with any of these AI companies. Why? What she said is that Bloomberg's not interested in one of these training deals in just a pure lump sum content licensing deal. That it's being very measured and thoughtful about making sure if we're going to deal with an AI company, it needs to serve our best interest in the longer term as opposed to just being a short term payment.

[Kamiko McCoy]

Yeah, yeah, I was going to say if it's an upfront payment, then you kind of run to your point that you just mentioned. There's no guarantee that that deal is going to be renewed. You know, once the platform has got what it needs from the for the publisher. But to guarantee consistent revenue or consistent income, then the usage based payment system would make more sense. There is a cabaret maybe this time joke in there somewhere. I'm not going to make it have to sing that song, but I hope that our listeners will infer it given, you know, the publishers have been burned before by the likes of Meta, so maybe this time.

[Tim Peterson]

So interesting because like Reddit. So according to Bloomberg reported this last week, Reddit's looking to sign new licensing deals with Google and OpenAI. It already has those deals in place, but according to this report, Reddit's looking at usage based pricing as the model for these new deals. And so it does seem to substantiate kind of this groundswell towards usage based pricing, especially because Reddit has a lot of valuable content for these AI platforms. And so it seems like it would stand the best chance of getting usage based pricing. And then we'll have to see does that trickle down to other publishers as well?

[Kamiko McCoy]

It's curious to see if there's a precedent that's going to be set here because on the one hand, which we've talked about before, you've got some publishers that are striking deals and then on the other hand you've got a handful of others that have lawsuits. Right. Against these platforms for, for scraping and copy. Not copyright, content infringement. I don't know what the actual word is. What is it?

[Tim Peterson]

In some cases it's copyright, but in other cases, yeah, it's content infringement. It kind of depends. There's like, it gets weird because in the case of the New York Times, I think they're suing regarding copyright because they're able to like have copyright because of print newspaper articles.

[Kamiko McCoy]

Okay, that makes sense. That makes sense. Yeah. So on the one hand copyright in copyright and on the other hand content infringem. So I don't know what the precedent that's going to be set here. And I think maybe, I don't know. In the US usually a legal lawsuit is what sets a precedent. I stand corrected.

[Tim Peterson]

Yeah, no, I mean there are more and more lawsuits. Obviously there's the New York Times, OpenAI one that's still underway. But within the past little over a week we've seen Encyclopedia Britannica sue Perplexity Penske sue Google over AI overviews. German publishers have filed a complaint against Google over AI overviews. And that was one thing that also emerged during DPS is AI overviews. Google's kind of like AI generated summaries that are appearing at the top search results. That was kind of the villain of dps because publishers are attributing, publishers kind of across the board are seeing search traffic from Google decline and they're tying that to AI. Overviews, just the prevalence of AI overviews at this point. Sarah Gaglioni had her media briefing last week where she recapped some of what was said during our behind closed door town hall discussions at dps. And one publisher in there said, quote, the decline from Google is not complemented by OpenAI or perplexity kind of traffic. So overall, definitely we're seeing a drop in search traffic and at the same time they're seeing the rise of AI overviews. So it's the. Of all the threats, it's one of many threats facing publishers at the moment.

[Kamiko McCoy]

Yeah, it's bad news bears, but you know, the resolve is what, a lawsuit or you strike a deal.

[Tim Peterson]

Yeah. Or you figure out how to be using AI to strengthen direct relationships with your audience, with readers. Because there was conversation around, okay, what are the ways in which we can be used in AI to streamline internal processes or improve internal processes? There was a lot of conversation about publishers like Bloomberg, like Business Insider, using AI to create these AI generated summaries at the top of articles in order to get people reading and spending more time on page. And then, you know, like we in the conversation we have with Sam Hahn as the featured segment in this episode, using AI to improve paywalls to help drive more subscribers, as well as ways in which the Washington Post is using AI to also increase its advertising revenue.

[Kamiko McCoy]

All right, well, I'm sure it's a juicy conversation, so with no further ado, I'll let you and Sam get into it.

[Tim Peterson]

Thanks.

[Interviewer]

So let's get right into it. Chief AI Officer. So this was a position that the Washington Post announced in June. You are in this position. It's a new position for the Washington Post. It's really a new position for the industry overall. When did you all start having discussions about creating a Chief AI Officer role?

[Sam Hahn]

Sure, sure. By the way, I like Chief Ninja better than Chief AI Officer, but yeah. So let me tell you a little about about that kind of adoption at the Post. I think the critical point of having a officer is where the air is just a tools for an individual department or small subset of departments too. It drives the core business or core growth. I think that transition point is where the AI office is needed. It's similar at the Post. So let me just tell you the AI's journey at the post. About 10, 11 years ago, I started Machine Learning team there and the focus was more like creating a personalization experience, recommendation models, and then categorize articles into different topics or moderating comments and so on and so forth. So it is really focused with engineering product team and focusing on customer facing application. And then we explored additional products called Heliograph where we wanted to create small stories automatically by data. So during election, if the election results have become available during the election night, we want to bring that in and write a small story. And in addition to reporter covered material, we want to put them together and show. And then during Olympics similarly certain medals were awarded. Want to create a short story story and go like that. So that was like first expansion of AI and required a lot of collaboration.

[Interviewer]

With the newsroom because you didn't have large language models.

[Sam Hahn]

Exactly. It was hard. It was really template based. And we went through rigorous testing with the newsroom make sure that it aligns. So it expanded a little bit with including newsroom. And then a few years down the road we started experimenting with the dynamic paywall and how do you optimize the ad and subscription revenue and start working with subscription marketing team. As you can see that the role, the engagement expanded. But still even up to that point it was just one team machine learning team doing majority of the work in collaboration with engineering products and so on and so forth. Now in 2023 with the language model ChatGPT kind of explosion, we start seeing differences in the adoption of AI. Newsrooms start asking how can I use AI to improve their processes? Can you use it for SEO metadata generation, can I use it to analyze video using AI? And all that started happening across the board in customer facing. Can you use AI to provide more conversational experience? And so on. So, so those kind of exploded and require coordination across the team to make sure that we don't have duplication of the work and we have some kind of standards, some kind of way to monitor the quality and performances. More importantly, in the previous years we had just like main machine learning team doing core work. But now with the language model it's everybody like software engineers can do the similar job. Even newsroom data reporting team can do meaningful project using AI and analytics team and so on and so forth. So now there was a need to really coordinate, so needed organizational support. So like two, three years ago, two years ago we started what we called AI hub. So which consists of me and newsroom leader and then product lead and so on. So we gather together, here's a roadmap, here's some kind of standard we want to push for, here's how we educate people and so and so and we had that to really put some of poses into production and so and so and then this year we Realized that we need a role to really overall and then in a way that's more like recognition of the need, but also announcement to the company and the external saying AI is critical to our business. We are going to focus more on AI.

[Interviewer]

So it wasn't until this year that you all decided we need a chief AI officer.

[Sam Hahn]

Right. I mean, we kind of developed that, saw that need. And then this year we say this might be the year we want to do it.

[Interviewer]

Okay. And then with that. So you're within the CTO's office, you report to the CTO.

[Tim Peterson]

Correct.

[Interviewer]

How was that always going to be the case? Or was there any discussion of the chief AI officer role? Where should that sit within the broader organization?

[Sam Hahn]

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

[Interviewer]

And then when you all were crafting this new role, to what extent did.

[Tim Peterson]

You go through the process of actually.

[Interviewer]

Writing out what's the job description? What are your direct reports?

[Tim Peterson]

What did that look like?

[Interviewer]

How did you figure out what the job description for a chief A officer would be?

[Sam Hahn]

Right. It was organic. If I were new to this company, then the process you're describing makes sense. I need to really define the mission and all, but because I've been leading that AI team for many years, it was kind of organic transition. But however, the mission for a chief AI officer is to use harness the power of AI to drive business growth, innovation and revenue optimization. So specific to our company goal, we want to reach 200 million paying customers. And how can you use AI? How can you use AI to improve user engagement by providing AI powered products? How can you 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 kind of the core mission of AI officer. And in addition, we have Arc XP, which is a SaaS CMS platform. We sell products, we develop at the Post and make them available to other news media companies. And another goal associated with that is how can you create AI powered products for ARC so that other news media companies benefit from the work that we did for the Post.

[Interviewer]

Okay, so two of the things you're being kind of evaluated against are how can your group help to drive some subscriptions and how can it help to drive ad revenue?

[Sam Hahn]

Exactly one of them, yes.

[Interviewer]

How do you even measure that though?

[Sam Hahn]

Yes. So one of the major tool I think some other media presentations were about that is having a paywall. So when reader comes to our site, how do we decide to give free access to article versus putting up a payroll, paywall or registration wall saying you have to pay to read more or you have to register to read more. So we built a machine learning model called reinforcement learning and analyze that a customer's history in terms of engagement. How many times this customer sold free articles. How many times that customer sold paywall gradual. What are the history this customer subscribed and churned. Look at all that and then looking at the content. Does article have ad value? If you have drag ad on the article, it's important. Maybe we prefer like free article view over paywall. We put all these factors into consideration and then the machine learning algorithm make a decision for this customer. It's a free article. This customer paywall. And we did AB test like using rule based. In the past it was more like rule based. Customers have and most 2, 3 free articles before they see paywall and so on. Or if customer coming from pay media campaign they have additional free articles before they see what all these huge rules that maintained by product and marketing team versus machine learning powered paywall. And we did AB test and we saw like 20% improvement in terms of overall customer lifetime value using that model. So that AB test clearly showed now that's used everywhere. And the best benefit of having that is no more meetings to decide what rule to do. So the Thanksgiving sales coming up. How should we change the payroll rules to accommodate more subscription during the sales price? We'll have meetings after meetings and after meetings. And then executive with the largest voice will won. Right? That's gone now it's all driven by machine learning.

[Interviewer]

And are you able to query the model for how you would want to change the recipe based on the data it's able to be across.

[Sam Hahn]

Exactly. So we have a live dashboard of the performance. And it shows what is a subscription take rate how many people just deserted will not come back. And we show all of them right there. And then sales marketing team look at that and say we need to adjust. So that was very evident during the election night. And we had election sales and subscription team wanted to make a change. It was like talking to our engineers change the parameter. Go in the past where we had to committee meeting to decide. By the time we decide, election is over.

[Interviewer]

When did you all start testing that and when did you officially roll it out?

[Sam Hahn]

We started more than a Year ago, I think at least like a year and a half ago. And we did ab and then we had the gradual adoption and fine tuning the models and all that's ongoing. So even now our subscription like related to 200 million paying customers. With a traditional subscription model like monthly and yearly, I don't think we can reach that goal. So our subscription marketing team created flexible access products. Weekly pass, daily pass. We are experimenting paper article. Now we want to bring all those into payroll decision. It's not just showing a paywall, it's a paywall containing rich products.

[Tim Peterson]

Oh wow.

[Sam Hahn]

So that's additional work continuing.

[Interviewer]

And the model is making that decision of do I present a monthly subscription offering versus a pay per article.

[Sam Hahn]

Right, Right. We are thinking of like multiple options. Right. Like out of six, seven options available, maybe you want to package two of them together. Another package with the different pairs. That's the algorithm machine learning model that optimizing the revenue.

[Tim Peterson]

Okay.

[Interviewer]

And you're evaluating this based on customer lifetime value. You've seen a 20% improvement on that. I imagine a challenge with that is if you're evaluating based on customer lifetime value, it can take some time to see that value.

[Sam Hahn]

Exactly.

[Interviewer]

Whereas you may have the marketing team or other parts of the organizations being like, it doesn't seem to be working just yet. We need to change something. How do you navigate those conversations and how do you get them to trust that this will work out over time? Then at the same time, if you don't see it work out over time.

[Tim Peterson]

How do you pivot but in a.

[Interviewer]

Way where they're not then saying yeah, we can't trust the model, I told you.

[Sam Hahn]

Sure, sure. The first step is make sure that customer lifetime value model is agreed upon. So what he did was we looked at past one year worth of data. So someone who started subscription one year back and we looked at actual realized revenue by subscription and advertisement, by that customer coming to our site and engaging with that article and so on. So we get a model. Say if someone get a yearly subscription in one year that customer will generate so much revenue in average.

[Interviewer]

And then we look at inclusive of advertising, everything.

[Sam Hahn]

Everything. Now we look at another customer who is an anonymous user who did not convert but occasionally visit our site, get a free article but never subscribe over one year period. What are the ad revenue that customer generates? So we show them by actual data and then saying hey, for anonymous users, one additional page view has so much dollar or cents value over one year. Now when we make a decision dynamically, we use that data to Make a decision. So for the subscription marketing team in short term they might not see great kind of subscription take rate there, but long term that's better for the overall revenue. But again we provide flexibility to subscription marketing team during sales period. We make sure that paywalls are shown more because we have great deal like model realized because the subscription price is lower, there is better chance of customer taking that knowing that we previse and more paywalls.

[Tim Peterson]

Right.

[Sam Hahn]

So we work together closely like that to make sure that they understand long term revenue or optimization component like subscription and advertisement has to come together.

[Interviewer]

And is it always looking over the.

[Tim Peterson]

Because I guess like one of the.

[Interviewer]

Funny things talking about customer lifetime value. We're not talking about from when someone's a baby to when they're on their deathbed, we're talking about a certain period of. Is it always a year for you?

[Sam Hahn]

All year to three years.

[Interviewer]

Year to three years.

[Tim Peterson]

Okay.

[Interviewer]

And so when you're having these conversations, how's that organized? Like do you have a person on your team that is the liaison with the other teams? Are you that liaison across the different teams?

[Sam Hahn]

We have a product manager so who interact with the business stakeholders to our team and they start that initial discussion requirement gathering and so on. So and then bring one of the data scientists or machine learning scientists to that discussion and say oh here's the problem. Let's say the new Flex products are introduced. Daily pass, weekly pass paper article. Now how can you bring this into the model? And we'll have discussions like how long will it take? What a B test look like, what are the metrics? We have to look at it. And all that discussion happened through product managers kind of organizing meetings like that.

[Interviewer]

What's the model specifically that you're using for these dynamic paywalls?

[Sam Hahn]

Reinforcement learning, deep learning based reinforcement learning models.

[Interviewer]

And are you using is it GPT? Is it Claude?

[Sam Hahn]

No, it's not. It's a different kind of. It's not like language model. It's its own model that use a lot in the last stage of language model. So other language model developers build a core language model. At the end they do the tune up. So this is for the question answer kind of usage. At the end they do the human in the loop kind of reinforcement learning. So that same kind of reinforcement learning methodology is used for that dynamic reinforcement learning. So what if I give this offer how does that impact revenue and learn from that dynamically and improve?

[Interviewer]

Okay, and in terms of the tech underpinning that is that all proprietary in.

[Tim Peterson]

House are you using off the shelf?

[Sam Hahn]

Our own. I mean, based on the research work out there and all, we have our own development.

[Interviewer]

Okay, and so then you have a team to work on that. How expensive are these people getting with Meta being out there shilling or just like throwing NBA salaries at people as you're hiring or as you're even trying to retain employees? How challenging has that become?

[Sam Hahn]

It has been very challenging. We have lost several key members to tech companies like that. But however, we found that there are a lot of good junior level machine learning scientists who can work with us. I mean, we are very competitive in terms of salary. Not we cannot compete on stocks and all other areas. But in terms of basic compensation, we are competitive.

[Interviewer]

But what's typically the range that you're offering?

[Sam Hahn]

I don't want to.

[Interviewer]

Seven figures.

[Sam Hahn]

Not that. Right. So what we offer is not by salary or compensation, but by mission. We make sure that we provide exciting work for the machine learning scientists to work on. They see their work implemented, go to production and showing value compared to other big tech companies where you work on a huge organization, focus on very narrow aspect of the huge system. The advantage we offer is you see overall, you work on a project, you see that going to production, you work with engineers and you see the value. Make sure that all the machine learning engineer scientists in our organization see that value and then develop the talent. That's kind of how we retain our talent, but it's very challenging.

[Interviewer]

Okay, so they feel more ownership over the projects as opposed to feeling like hogs and machines.

[Sam Hahn]

Exactly.

[Interviewer]

How many people are in your organization?

[Sam Hahn]

So I have about 16 machine learning scientists and then I have six or seven data engineers and then about 10 software engineers focusing on only the AI work.

[Interviewer]

Okay, and how has that team grown this year? I imagine it's grown this year.

[Sam Hahn]

Grown quite a bit. Right. So I think at least 4 or 5 in machine learning side and then a few more from software engineer, one additional data engineer, and so on.

[Interviewer]

And then I imagine some of that maybe is like people inbound applying for jobs. But how do you source the talent? Are there specific industries, verticals, levels of experience, backgrounds that you're looking for as you're hiring for the different positions?

[Sam Hahn]

Not that particular. So we go all the regular sources, LinkedIn and other resourcing available out there. And then we also work with some interns. So we work closely with the Virginia Tech and we have research collaboration with that institution and we have a couple of new employees from that organization. So that's another way we develop talent and bring them into our team.

[Tim Peterson]

Got it.

[Interviewer]

Okay. And so you've been in this role for three months now at this point. What's the biggest difference in your day to day right now versus where when it was in, let's say, May of this year before officially you became chief af?

[Sam Hahn]

Right. I think one is had a chance to work on AI policy across the team. So Noodstrom had its own AI policy. Right. Human in the loop and all that guideline, but we didn't have guideline for the overall organization. If you look at our resource, we have our own language model, chatbot version, internal use, and we also have Enterprise ChatGPT version available for team and we have Google Docs which also has Gemini in there. And then individual reporters or engineers have their private or personal account. So with all that, how should we help our employees to use these tools for their work? So we have very strict guidelines saying if it's a business critical sensitive information use language model Hosted by In US internal and otherwise, you can use enterprise version of ChatGPT and so on. So if documents are in Google Doc already, it's already safe. We have correct classification of document and so on. So then you are allowed to use Gemini to do certain tasks. Right. So providing the guideline has been kind of another new task and then another one was Arc xp. We want to really expand the tools that we developed, make it available to all news media companies. And I've been pushing that pretty hard past two months.

[Interviewer]

Okay, so it used to be you were having to write the rules for the dynamic paywall system. You're able to now have machine learning take care of that. Now you're writing rules for AI use by the human employees.

[Sam Hahn]

Yes, to the degree, yes.

[Interviewer]

This is a really interesting conversation. I really appreciate you taking the time for this. Thank you so much.

[Sam Hahn]

Thanks a lot.

[Tim Peterson]

Thank you.

[Sam Hahn]

Thank you.

[Tim Peterson]

Thanks for listening to this episode of the Digiday Podcast. If you enjoyed it, please leave us a rating and a review on Apple Podcasts, Spotify or wherever you're listening. Get more from Digiday with our daily newsletter sent out each weekday morning. Visit digiday.comnewsletters to sign up. SA.