#12 - [Weekend Listen] 43 Minutes on Recommender Systems @ Newsroom Robots

First, please follow Nikita’s podcast, Newsroom Robots . She hosts the most influential podcast on AI for the news industry.

I'm not a media-savvy person. Given the choice, I could spend months at home reading, working, exercising, and spending time with my family. However, my belief that recommender systems can economically support big media is so strong that I’m taking the uncomfortable step of writing this newsletter and recording my first podcast. Thankfully, Nikita was kind enough to help me. ??

Of course, listen to the podcast. If you can't below are key takeaways:

1. No rich country without low-cost energy (electricity). Similarly, no big media company can thrive without asymmetric algorithmic advantage.
2. Recommender systems deliver monetization through sustained, repeat engagement. Without them, Internet Products eventually end up in stable equilibrium (stagnation) or negative equilibrium (growth in monetization reduces retention).
3. Recommender systems will become the ‘operating system’ for media businesses, managing both content and ad distribution. This allows teams to focus on higher-order problems, like core value proposition.
4. Pulling this off is extremely low probability. But that’s what makes it worthwhile. If everyone could do it, there would be no competitive advantage.
5. Implementing recommender systems requires deep changes—organizational restructuring, new design language, building low-latency and high-consistency data pipelines, serving front-ends directly from servers (not cache), and retraining your product team.
6. Sustained AI performance depends on self-learning models with no hard-coded business rules (shortcuts). Resisting this urge is hard ??. Very hard.
7. As a low-ARPU industry, news companies can struggle to make multi-year bets of this scale.
8. Is personalization evil? ?? Most social media platforms suffer from echo chambers, conspiracy theories, NSFW content, and misinformation because they distribute UGC. In comparison, all content on news products is product by the editorial team. To counter this negative associations, I now refer to what we build as ‘algorithmic distribution.’ ??
9. But my business is too complicated: I know. Most news businesses have diversified into multiple revenue models — direct ads, indirect ads, videos, affiliate programs, sponsored content, and subscriptions. Beyond personalization, your recommender system must act like a mutual fund for each user, adjusting the ratio of monetization methods and hiding irrelevant options.

Enough with the takeaways. Go listen to the podcast! ??