I promised an essay last week and it did not happen. I will try to make it happen this week. Stay tuned.

Keep it simple,


Edward, Startup Curated

This Week's Sponsor

Live webinar with Vanta! De-risky business: How to assess and mitigate risk

Managing risk is critical to protecting your business and maintaining customer trust — but how do you de-risk without the headache? In this webinar, Matt Cooper, Senior Manager of Privacy, Risk, & Compliance, will share how to leverage tooling and automation to simplify the risk assessment process.

Edward, Startup Curated


Attractiveness: In July 2021 I wrote about intra-elite competition and prejudice against ugly people. New study from Sweden has found that attractive female students get lower grades when non-quantitative classes move remote (attractive men do not see any drop). Also: TikTok has more “lookism” than YouTube.

Everything becomes the same thing: (Essay, April 2021) Twitter is adding TikTok short-video “discover"y” feature.

Unseemly Arbitrage: (Essay, October 2019) CEO of Wordpress (who bought Tumblr, and forced porn off the platform), explains why pornography is not possible on modern social media (due to restrictions by infrastructure companies)

Predicting Divorce: (Essay, Aug 2022) John Gottman claimed he could predict who would and would not divorce with 94% accuracy. Now Todd Kashdan looks at the data and shows that Gottman’s “predictions” were all backwards looking. Predicting the future is much harder than “predicting” the past.


MrBeast: The elite YouTuber has a 6-person team that creates thumbnails for his videos and optimizes them for CTR. They actually figure out what thumbnails they want to create BEFORE they shoot the videos, so they can make sure they have the right shots for the team to test. Data wins. But he adds that lately he has been, “obsessing less about numbers and more about quality”. Data optimization is very important, but can only optimize for the short term. See my essay on Presidential emails.

Unilever: The CPG firm had previously announced new deodorants for “people with disabilities” and kicked off the announcement with a series of award winning ads. But now it seems that the product will never even get to shelves, quote: “Consumer testing after the announcement found that users wanted customizable solutions for their particular needs, not a single product meant for a wider range of disabled people.” and “The biggest headline that came from [the trial] is in trying to be inclusive, we were actually being exclusive”. Someone was drinking their own Kool-aid.

Customer feedback: In theory customer feedback is a great way to collect forward looking metrics — your sales may be great, but if customers are not happy, then future sales may be in trouble. In practice trying to get quantifiable feedback from users is only possible by degrading the customer experience you are trying to measure. Three is little upside for a customer to fill out yet another feedback survey, so the users that do are rarely representative. The WSJ writes about one potential solution: Use machine learning on customer calls to try and predict what their feedback scores would be without having to request them directly. I’m skeptical. But this is a real problem, so it is unsurprising that vendors are finding at least some clients that are ready to give it a try.

Just Do It: Dan Wieden was the creator of the Nike slogan and co-founder of Wieden+Kennedy. He passed away on Friday last week.

Apple’s new ad framework: Apple both kneecapped the leading ad industry, and then provided their own tools that had significant gaps. The good news is the tools are improving. Eric Seufert at MobileDevMemo has a solid writeup on the latest upgrade and what it means.

Meta creative tips: The Facebook team believes you ads will be more effective if they are more creative. They provide a list of tips to help you achieve that goal:

  1. Empower your users (i.e., getting Pampers customers to provide childcare tips)

  2. Elevate unheard stories (i.e., Flare Audio told a story of someone with sound senstitivities)

  3. “Shop the moment” (i.e., An ecommerce company ran a livestream in Asia)

  4. Solve IRL (i.e., an Indonesian ecommerce company helped local businesses drive sales with branded templates)

  5. Metaverse (i.e., ParaOlympics created a VR experience)

They definitely don’t read like silver bullets for most businesses to increase CTR…

Changing opinions: Another example of “never trust customer surveys”. People are very bad at predicting how they will feel in the future.

Edward, Startup Curated


Zig when Zag: Rory Sutherland argues that vacation destinations have become “winner take all”, which means that “tier 2” destinations are very undervalued — much cheaper, marginally less distinct, and far less crowded. I found this to be true last summer when Portugal was cheap, pleasant and empty; while Paris was nuts. Lessons here for choices beyond holidays.

Airbnb amenities: Hosts are adding pickleball courts. This apparently improves conversion, but also, as Airbnb increases search discovery capabilities, improves visibility. If you want to get ahead of this I would look through the new search options and try to find ways to have your property appear under more filters. May I suggest a treehouse/yurt combination?

Amazon Streaming: A good Bloomberg write-up on Amazon’s streaming strategy. The company is planning to spend $15B on programing this year. The writer makes clear that no one knows if Amazon’s streaming strategy is working, just that "Video is the second biggest contributor of new subscribers to Prime after shipping.” (so I guess that means it is better at customer acquisition than Amazon Music? Whole Foods discounts? What else?). Given Amazon’s penetration in the US, I expect that the real value of Amazon Studios is in international markets. The nice thing about paying for content is that, unlike fulfillment centers, the marginal of providing Prime benefits in a new geography round to zero.


Text-to-video: Meta has announced their text-to-video generation tool. I knew this was coming but would have guessed we were a few more years away. This has been the year of surprisingly fast advancements.

Bruce Willis: There were rumors last week that Bruce Willis has sold his likeness to be used with Deep Fake technology in future films. His team has denied it and the claims have been retracted, but this technology is out there, and if not Willis, someone will sell their face sooner or later.

Low Code: Rowy allows you to play with StableDiffusion in a “spreadsheet-like environment”. More and more of these tools will be built on top of the underlying models.

Dreambooth using Stable Diffusion: Google’s Dreambooth AI is a test-to-image tool that can do some pretty impressive things. It starts with an image, but then allows for extensive modification (new angles, adding/changing clothing, changing styles, recontextualizing, new locations, changing expressions, and even modifying properties — i.e., this dog, but if it was a lion). Techbriefly describes how to make this type of tool work using the open sourced Stable Diffusion.

Adept: A tool build on top of GPT-3 that allows “high level requests” written in English to be actioned answer a wide variety of queries like “find me a house in Houston that works for a family of four within my budget of $600K”. Claims it can replace 10+ clicks with a single sentence in Salesforce and is able to solve problems across multiple programs. Nothing too surprising at this point, but still pretty cool.

Google Shopping: Google will now let you shop directly from screenshots.


Toilet Timer: Use case is left as an exercise for the reader.

Edward, Startup Curated

Keep it simple,


Edward, Startup Curated