Ben Young
Ben Young
March 8, 2024

Edition #439

Learning to ride again, predictions and Ikeas wonderful partnership with Annie Leibovitz.

I’ve been teaching my eldest to ride a bike, which is pure joy. And we’re at the part where you feel they have the balance, so let go at the back of the bike, and off they go for a bit. And each time just trying to extend and getting a bit further. You’re getting that intuitive feel, and go! This got me thinking about predictions..

And then I listened to Boston Dynamics founder Marc Raibert, the creator of those robots we’ve all seen that remind you of Terminator. Well Marc was talking about the algorithms in the robots – and how they have a continuous prediction loop, using visuals to predict in the next second, second and a half, what’s going to happen. And then this of course, helps avoid obstacles, climb over things, adjust etc.

“I think that the number is only a couple of seconds for Spot. So there’s a limited horizon type approach where you’re recalculating assuming what’s going to happen in the next second or second and a half. And then you keep iterating at the next, even though a 10th of a second later you’ll say, okay, let’s do that again and see what’s happening.”

Pulling the thread further, when we look at predictive analytics, the mega hot topic right now. Using these to make predictions in the future. Making a prediction for a limited time horizon. And doing that on a continuous loop. It’s not unlike what the robots are doing.

You see, when you query OpenAI, you’re getting a prediction for now. It doesn’t say, oh if you were in 2025, I’d write something different. The time horizon matters. And to help understand that horizon, you want to know what it was built on or up to. Again OpenAi as an example, it was built on the culmination of knowledge to a date.

But most other use cases in our industry for predictions will be built off smaller windows. i.e. with ad targeting, it needs to be dynamic to now, so yes having historic data is helpful, but it needs to make predictions based on what’s happening in market now. And bid/place accordingly.

So I think this idea of time horizons will become more prevalent, how far back is the time horizon, how long is this time horizon for, is this a real-time prediction, or a slow time, or an enduring prediction.

Notable stories this week

  • Enjoyed Bens (first hand) take on Vice.
  • John Lewis plans to launch its own in-house content agency.
  • LinkedIn doubles down on news as social rivals retreat.
  • Time to hustle’: Content marketing is under pressure to demonstrate ROI.
  • How Notion unlocked influencer marketing gin the early days to drive millions of sign ups.
  • Spotify expands content marketing efforts to include fortune-telling.
  • Telegram has an interesting monetization strategy.
  • How one Google featured snippet is killing commercial list-based content.
  • From AI content to fake local news, brands must play programmatic dodgeball this election season.
  • Why media companies are pushing podcasts at SXSW ’trojan-horse’ style.
  • Let’s get clear on attention metrics.
  • Kielce Brothers are shopping their hit podcasts.
  • Taylor Lorenz launches video podcast with Vox Media.
  • New ways Google is tackling low quality content on Search.
  • Rogue editors started a competing Wikipedia that’s only about Roads.
  • How a better understanding of media quality can help unstick culture.
  • Cookie deception sparks tension between buy and sell side in IAB Tech Lab.
  • IAS expands “made for advertising” driven measurement.
  • Edge is moving towards deprecating third party cookies too.
  • The Atlantic calls TikTok broadcast tv.
  • Facebook is deprecating Facebook News.
  • FTC is cracking down on mass data collectors.
  • LiveRamp reported to UK, French regulators.
  • Why podcast companies are investing in AI-generated podcast translations despite questionable quality.
  • President Biden would ban TikTok.
  • Playground XYZ launches optimal attention.
  • NTWRK folds into Complex.
  • Nilay Patel explains why websites have a future.
  • How the media industry keeps losing the future.
  • Threads enables post-scheduling in live API pilot, plans to add analytics capabilities.
  • Google open sources a MMM. Super smart.
  • Why is your news site going out of business?
  • Reality checks.
  • [Long form] Shane Smith and the final collapse of Vice News.


  • Investment is (slowly) trickling back into ad tech.

Campaign of the week

View all 2024 best campaigns.

Smartest commentary

  • “This thirst for measurement and a shift toward more data-driven decision-making is driving marketers to think more carefully about the metrics they use to gauge the success of their efforts – and to pay more attention to analytics tools and technologies that can offer insights into the performance of their content.”Shareen Pathak.

Datapoints of note

  • Taboola saw traffic to the open web go up 30% when Facebook/Instagram were offline.
  • 5% of videos generate 89% of the overall views on TikTok. Via Shane.
  • Brands outperform competitors in most KPIs with creator marketing.
  • Google expect that recent changes will collectively reduce low-quality, unoriginal content in search results by 40%.
  • IAS analysis from over 40 global agencies and brands found traffic served on sites classified as non-MFA have a +278% better conversion rate than traffic served on sites classified as MFA. Further analysis by IAS found that quality media was more cost efficient than sites classified as MFA, delivering lower cost-per-conversion by 63%.
  • There has been a 12% jump in time spent on X since last year.
  • The Beehiiv ad network is now at a $3m run rate, and on pace to 10x this year. Wow.

View all 2024 datapoints of note.

That’s it for this week.