Marketing BS Briefing: AI Inflation Edition
Marketing in Inflation, Naming Stadiums, Sentient AI, Unarmed Human Combat, and More...
From 2016-2018 I was CMO of General Assembly. This weekend Bryan Caplan, author of “The Case Against Education” published a piece I wrote about how Tech Bootcamps work on his blog “Bet On It”. You can read it here. (And welcome to the new subscribers who joined from Bryan’s blog!). Onto the briefing:
Marketing
Inflation: Last week the month-over-month inflation rate was +1% (+8.6% annual. The highest since 1982). How should marketers be thinking about this new world of high inflation that most of us are experiencing for the first time in our careers? This is worth a full essay, but in the meantime, I highly recommend listening to the entire Aswath Damodaran episode of Invest Like The Best (Apple Podcasts link). Some of my take-aways from the episode:
High inflation is “fine” and can be dealt with. The challenge is variable inflation. And high inflation is almost always variable. 2% inflation might vary from 1.6% to 2..4%, but 10% average inflation may vary from 4% to 16%.
High inflation hits hardest on non-discretionary goods. And we will soon find out which goods are actually discretionary (Netflix subscription?)
It is often hard for companies to raise prices - they will lose sales to competition. Extra hard if you are in a regulated industry (don’t expect rents in New York to rise with general inflation. So being a landlord there may be tough!)
If you aren’t thinking about how you will raise prices, now might be a good time.
Conversion Rate Optimization: Most CRO optimization is over-rated. It works, just not as good as the bulls think it does. What is most impactful is often getting simple stuff right. Here is a good example: Dylan Price shares that adding a demo store to his Shopify app listing increased his conversion from 2% to 13%. Showing your customer how your product will work in their life can be an easy win.
False Advertising: Lying a little bit in an ad is illegal as consumers may be deceived. But blatant “over-the-top” lying is okay as it is assumed the consumers know you are lying and therefore they are not deceived. Specifically courts have found claiming your bread is “fresh local quality” is fine even if the bread is made out of state (and not local). Everyone knows it’s a lie, right?
YouTube: Mark Ritson sat down with Google to talk about how Australian brands are using YouTube. You have to be careful with these house-organ-type videos, but Mark generally tells it like it is.
Market Research: Tyler Cowen uses survey data to show that Americans are not as polarized as generally believed. If you ask Republicans and Democrats how much “income redistribution” should be done on a 7-point scale the modal answers at 1 and 7. But if you ask the same people “Do you consider the amount of property taxes you pay to be too low, about right, or too high?”, only 43% of Republicans say too high, compared to 37% of Democrats. He repeats the exercise with a couple of other examples. Another case of “trust what people do, not what they say they care about.”
More Market Research: Drilling home the above point, Ryan Wallman shares “every customer survey report from a consulting firm”:
Marketing to Employees
Naming Stadiums: Part of my book goes into the ego that drives marketing executives to sponsor sports stadiums. “Axiomalpha” assumes that making a poor decision on a stadium sponsorship is a signal that LOTS of decisions in the company are being made poorly. They test the hypothesis with 98 companies and find out if a purchase is followed by a decreasing share prices over the following years. The hypothesis plays out (almost) all of the time. The exceptions are companies like Target and Qualcomm who manage to negotiate very sweet deals on the price they pay (due to extenuating circumstances). The fact that the real value from these sponsorships comes from impact on employees is mentioned only briefly in the Target example. I think it is true in every example, it is just that the cost vastly outweighs the benefits.
Summer Worker Shortage: WSJ has a piece on summer camps and how hard staffing has been this summer. “Andy Pritikin, owner of Liberty Lake Day Camp in Mansfield Township, N.J., said before the pandemic, 85% of his advertising budget went toward attracting campers and 15% went toward attracting staff” - now that ratio is reversed.
Business / Strategy
Rush to Quality: I had a theory that post-COVID there would be a rush to quality for events and experiences. If the “fixed cost” of leaving the house increased, and the in-house alternate improved (zoom conference!), then, on the margin the demand for “lower end” experiences would drop more than for “higher end experiences”. I am not sure it will actually play out that way (although it might for conferences!). But Arpit Gupta points out that it is already happening for office real estate. Traditionally there was a lot of low-end office space that was needed for companies that just had to get some sort of location for their employees. Now that non-creative work can mostly be done from home, will those spaces be needed? High end office space for the “creative class” to develop new ideas and products and work collaboratively will still be needed, but that may be a smaller percentage of the commercial real estate stock than many think.
Diversity: Is diversity (real diversity, not just surface level) good or bad? Pro argument is that it leads to more variety of thought and more different ways to attack a problem. Anti-argument is that a homogeneous group can work together with less friction. While it is the “nice thing” to say that diversity is always good, it is not obvious which effect dominates in any particular setting. This new paper published in Science Direct suggests that “the cost associated with misaligned incentives is minimized if experts of similar abilities are placed in the same team. Consequently, surplus maximization may lead to non-diversified teams”. This paper makes the case that a company talking about diversity makes things worse for under-represented groups (at least the sense of belonging for members of those groups). Conversely Vincent Geloso looked at early Canadian history and found that regions settled by more homogeneous settlers under-performed vs more diverse communities: “Homogenous areas were the least likely to innovate. The fact that this holds independently of the ethnic background of the homogenous area (i.e., wholly French or wholly English) is just damning”. Related: The world’s biggest plant has been found off the coast of Australia. Scientists were sampling DNA in multiple places in the Ocean to see how much genetic diversity was in the area, but found what they thought were multiple plants were actually the same plant. Something about how less genetic diversity is bad, unless it hits zero.
AI
The Economist: The paper published four pieces on AI last Thursday: Their Leader is about the “promise and perils” of Machine Learning. Their longer Briefing goes into more details on “Foundational Models” that were used for things like GPT-3 and Dalle-2. The issue also includes two essays “by invitation” on how close deep learning language models are to consciousness: Soon. Far. Interesting because until now these new AI language models have been fairly unknown to the wider public. I have spoken to dozen of CMOs and most had no idea what GPT-3 was, let alone its capabilities. A cover story in the Economist will change that.
Sentience: The Washington Post dropped a story on Saturday about a Google engineer who was believes that Google’s latest AI has already achieved “sentience” (his claims were explored and dismissed internally and he is now on paid administrative leave due to breaking company confidentiality rules). The article contains a chat the engineer had with the bot where the bot claims to be sentient. I am skeptical. The language models are very good, but they spit out what you want them to spit out. If you do the right fine tuning, and set up the right prompt, it will say things about being sentient - especially if you are careful to edit the questions and answers before showing it to the public (which happened in this example). Astra Star Codex believes that language models COULD become sentient, but this is not that.
Imagen: Google’s new text-to-image generator is outperforming Dall-e 2. More here.
State of the Art: Current State of the Art benchmarks in AI have been collected here.
Prompt Engineer:
Dall-e Secret Language: When you ask Dall-e to create text, it creates letters in the form of words, but it all reads like nonsense. Giannis Daras figured out that those “nonsense words” actually have meaning to the algorithm - it is “Dall-e Language”. You can even make prompts in the “Dall-e language” and get results. I am still not sure why this is happening. "Apoploe vesrreaitais" means birds. "Contarra ccetnxniams luryca tanniounons" means bugs or pests. “Vicootess” means plates of vegetables:
Counter Dall-e Language: Benjamin Hilton tried to replicate the language findings and believes that the claims are over-stated. When he plugs in the words for “birds” he gets a bunch of animal images, but not exclusively birds. Vicootess does give plates of vegetables, but when combined with English words like “Vicootees, cartoons” or “Vicootees, paintings” it gives things like characters, flowers and landscapes. Something is going on, but it is not as simple as the words Dall-e spits out being a language that could be used directly.
Stained Glass Windows: Astral Codex Ten got access to Dall-e and tried to create stained glass windows featuring things like Empiricism, Tycho Brahe and Occam’s Razor (instead of the standard religious themes). In the process of creating the windows, Scott learns a lot about how the tool works. Recommended.
Kermit: An anonymous twitter account asked Dall-e to create images of Kermit the Frog in the style of different movies. The results are pretty fantastic. It is hard to choose just one, but my favorite: “A still of Kermit The Frog in The Grand Budapest Hotel (2014)”
Weird Dall-e: Here is a twitter account that just tweets Dall-e generations.
Dall-e Mini: Don’t have access to Dall-e 2 yet? You can use “Dall-e Mini”. Unfortunately it is very very slow. And the results are not great. See my attempt at “Kermit the Frog as Terminator”:
Careers
Early Experiences: It seems that many top performers had early experiences where they pushed to achieve something. Sally Rooney “One of the foremost millennial writers”, writes about her previous career as the number one competitive debater in Europe: "I don’t think I will ever again want something so meaningless so much". I had the same feelings when I was younger competing in competitive improvisation and life guarding…
Performance Anxiety: My first year Psychology course taught me that being watched increases anxiety, and usually hampers performance for amateurs and improves performance for professionals (the “ideal” level of anxiety is higher the more skilled you are at the activity). Rob Henderson shares a quote from “The Self Explained” that performers are more likely to “choke” when friends and family are in the audience. Unclear how that changes (if at all) based on base ability level? I know that when I speak of perform comedy I am definitely more tentative when friends or family are in the audience - especially if they have heard any of the content before.
Fighting: Did you know that unarmed human-on-human fighting has improved more in the last twenty years than in all of human history beforehand? I watched the first few UFC fights in the 1990s when Horace Gracie demolished all of his much bigger opponents. But apparently today it is believed that even low ranked amateur UFC fighters would destroy the Gracie of the 1990s. It is amazing how controlled competition can result in so much innovation. “A good purple belt from a legitimate school today would submit a 90s Royce Gracie 5 times in 3 mins.”
Fun
Grapefruits: Per Capital Grapefruit consumption is down 80% over the last 40 years and no one seems to know why (oranges -27%, peaches -61%). Source.
Life Hacks: David Brookes at the NYTs summarizes some of the best ones. Related: Milan Cvitkovic’s list of “things you are allowed to do” made the rounds on Twitter and Newsletters again this week. Recommended if you haven’t already read it.
General Knowledge: I did not know that men outperform women (by a lot) on general knowledge (exceptions: Fashion, pop music, medicine, cooking). Studies from USA 2001 and 2002, and replicated in Australia.
Howard Johnson Restaurants: The last Howard Johnson closed in early June. Only relevant because when I was about 8-years old I had lasagna there, followed by the worse food poisoning of my life. I swore I would never return. It is hard to know if one will really keep a promise forever. Most of the time you can’t be sure until you die. But now I know. Promise kept. RIP.
Keep it simple,
Edward