Alternative Data Weekly #236
May 2, 2025
Thanks for being here!
The theme that emerged in this week’s email is … AI is changing how we work, but the pace of change has me recalling a quote attributed to Bill Gates “most people overestimate what can be done in a year and underestimate can be done in ten years.”
QUOTES
“To implement AI at scale without making a mess of things, don’t build anything until you’re sure there’s a clear business case.”- Cassie Kozyrkov
News Articles
Podcasts
Cool Charts
Final Thoughts (Gell-Mann for AI?)
#1– Jeremy Korst, Stefano Puntoni, and Olivier Toubia published How Gen AI Is Transforming Market Research. April 2025.
My Take: Marketing will change significantly with the use of AI. Perhaps more than any other area of business. Of most interest to me in this article is Digital Twins (“AI people” that replicate your customer). Digital Twins will allow marketing to endlessly ask questions of “customers”, with early feedback on responses being 85% accurate, and only getting better. This will change the balance between broad shallow surveys & narrow deep surveys … now marketers can have both.
#2 – Nomad Data published Why AI Hasn’t Transformed Work Yet — And What Needs to Change. April 2025.
My Take: Context is key. AI solutions are not one-size-fits-all. AI tools are generalized. “The burden of adapting these tools to real-world business processes within an organization has been unfairly placed on the users themselves, who are rarely technologists.”
#3 – Sven Balnojan published AI states the obvious. April 2025.
My Take: AI does some things very well. What is interesting is the stuff that it does OK (like 90-95% well) but can finish the task in 20 seconds vs 2 days.
“Like the junior analyst who stays up all night preparing every possible data view before the executive meeting, AI can methodically work through all variations, combinations, and edge cases without fatigue or complaint.”
I find myself just trusting the output because it used to be such a pain. Hearing this at law firms too where huge briefs are compiled in seconds and attorneys are missing errors. It is very tempting. And the pace of improvement, the 90-05% cases should be 98-99% accurate in relatively short order, which is great if you are a junior lawyer tasked with mindlessly pulling together docs, but bad if you are a client of that attorney trusting their judgement. Plus this introduces Gell-Mann Amnesia for AI. Second week in a row of citing this phenomena. See Final Thoughts below.
BONUS: Ben Lorica published Real-World Lessons from Agentic AI Deployments. April 2025. “What will truly distinguish industry leaders from those left behind goes beyond mere technological adoption. The companies that thrive will need to reimagine their organizational structures around human-AI collaboration rather than treating agents as mere cost-cutting automation tools.“ (JF: this stuff is fascinating!)
What else am I reading?
Abraham Thomas published Data and Defensibility. April 2025 (repost from last week … check it out)
Alex Boden published How will S&P Global and CME Group spend their $3.1bn OSTTRA windfall? April 2025.
Pierluigi Vinciguerra published Are LLMs capable to replace traditional scrapers? April 2025.
Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, Romeo Dean published AI 2027. April 2025.
Cassie Kozyrkov published Is AI a passing fad? April 2025.
Dan Entrup published 5th Round Data Delivery. April 2025.
Peter Baumann published Leading the AI Transformation. April 2025.
Vin Vashishta published AI Is Replacing CEOs, Just Not The Way We Thought It Would. April 2025.
Source: AI in Financial Services podcast host Matthew DeMello interviewed Arpit Saran, VP of Digital at Regions Bank. March 2025.
My Take: A few takeaways for me on this conversation. First is the need to stay ahead of potential fraud, particularly within financial services (all driven by data). Second is the customer expectation of hyper-personalization (And this is just getting started).
Lastly, early AI companies that are succeeding do one function and do it really well (I am seeing this in the market). This will change as AI systems gets better and more is expected of AI.
Use Cases (every aspect of banking can be impacted by AI):
Financial Advisory
Customer Support
Risk & Compliance
Highlights (25-minute run time)
Minute 01:00 – interview starts
Minute 04:00 – we need to stay one step ahead of fraudsters (Chief Risk Officer involvement)
Minute 07:00 – Regions at the top of list when it comes to technology stack
Minute 09:30 – Klarna removing some major application in house utilizing GenAI
Minute 11:00 – use cases; every aspect of banking can be impacted by AI
Minute 14:30 – hyper-personalization becomes an expectation
Minute 09:30 – it is all about the data (harnessing, mining)
Minute 23:00 – interview ends
Source: Dylan Anderson published Approaching Data Quality in Today's Complex Data World. April 2025.
“The elephant in the data room, and how to address it with everything going on in our industry”
…Dylan has some of the best visualizations (above) and the best memes (below):
Gell-Mann Amnesia for AI.
For the second week in a row, I’ve cited this phenomena.
“Briefly stated, the Gell-Mann Amnesia effect is as follows. You open the newspaper to an article on some subject you know well. In Murray's case, physics. In mine, show business. You read the article and see the journalist has absolutely no understanding of either the facts or the issues. Often, the article is so wrong it actually presents the story backward—reversing cause and effect. I call these the "wet streets cause rain" stories. Paper's full of them.
In any case, you read with exasperation or amusement the multiple errors in a story, and then turn the page to national or international affairs, and read as if the rest of the newspaper was somehow more accurate about Palestine than the baloney you just read. You turn the page, and forget what you know.”
This happens with AI all the time. I prompt for some information about a topic I know well, and immediately see flaws. I prompt for information about something I know little, and I trust implicitly … well 99% trust … and am happy that it only took a second to compile all the (partially correct) information.
Has anyone coined a term for this phenomena yet? Or are we just going with Gell-Mann for AI?
John, helpful - TY for your work!