It's hard to overestimate the impact of this technique, though it is 5-10 years too soon. :)
What this does is lets people reason about change in massively multivariate data sets. Like GPT level multi-variate. I look forward to seeing this applied to the state of ML models, as this is what we're going to need to express the complexity of a lot of techs we will all depend on.
There is a casting pearls before swine problem they will need to overcome, as the level of management understanding to interpret and respond to the changes this framework expresses is going to take some years to get people with that level of technical sophistication into decision making positions.
Near term, my impression is it will be amazing for getting funding for scientists of all kinds, and being used in hyper-competitive data fields like americas cup sailing, F1 racing, and spooky intelligence/social media company social engineering analysis, and I could see it being used in a new hybrid of quant trading funds, as any sufficiently advanced data vis is essentially arbitrage. All very rarefied niches with a high degree of autonomy. It may be too cool for making policy or lower level decisions in the near term, but this is super epic.
When you look at the recent history of how data viz has impacted fields, it has always been way out ahead of the industries it served. I hacked around with 3D "coral" graph viz back around 99-01 (early CAIDA stuff) for internet security analysis in govt, and in spite of its incredible analytic power, the anecdata was it didn't get traction because it was unmanageably powerful. However, I think we're just entering a data viz renaissance, and this is the bar. Amazing.