In this issue...
Morning, folks. Taking this week a bit easy after a friend's wedding last weekend. I'm visiting old friends in Zurich, and doing a bit of data wrangling on the side. You gave me a lot of feedback about MapSplit, and I'll add to it in due course. Maybe turn it into a game, who knows 😃
Quantum #671 had an open rate of 28% and a click rate of 8%, and the most clicked link was the brilliant 200ms interactive.
Also, someone made me realise that the past few newsletters were over the 102KB Gmail threshold, and this likely explains while open rates suddenly dropped – I had attributed this to peak August leave. Interesting observation, and something to look into. I'd never looked at how big Quantum was, and I've suddenly realised because I could no longer read it all on my mobile in signal-free areas. You never stop learning!
·The one article that caught my eye this week is in the Dataviz section and it's about tennis. Ben Shelton, the American tennis player, has a notoriously powerful serve. I'd say, as a tennis amateur, that his serve is probably the best in the circuit. What I had never noticed, and the chart excels at showing, is how bonkers is second serve is. You know that in tennis you have 2 attempts at serving: usually the first is your best, strongest one, aimed at preventing the opponent to return, while the second is a "just make sure it's in". It turns out that Shelton is a statistical curiosity in that his second serve has an unreturned percentage that is way above that of any other pro. Obviously, a serve isn't everything in tennis. But you're here for the data, not for the tennis, right? 😃
Last week at work I had a great chat with some tax advisors about the use of AI in tax. One of the clear questions was "how can we trust that AI does the right thing?", or "how do we correct errors made by AI in tax processes?". I found myself saying: "these are very good questions, but can I ask: how can you trust that humans do the right thing? How do you correct errors made by humans in tax processes?"
Fundamentally, I think this is the crux of any AI deployment, and they're all about learning how a human process works and what's its baseline error before we can deploy AI effectively. This is one of my favourite side-effects (in my case, it's becoming more and more an intentional, by-design choice) of working in AI-driven innovation: the ability to point the finger at what doesn't work now, in a non-AI process, and – finally! – develop the right performance metrics. I always say that I'm a data guy before being an AI guy, and this is increasingly true. I told my colleagues that I'm not a true believer in AI and I'll never be; but I'm also not a true believer in humans 😃 And what I mean by that is that capturing the right metrics (and I mean the right metrics – which is a tough question in itself) should be the basis for any process analysis and improvement. No hype, no doom. Just evidence.
·This week's LOL:
·'till next week,
Giuseppe @puntofisso.bsky.social
