It's a data classification exercise. To your point, not all data is the same. Some data will be low risk, and it's not, for example, health care. We need to be incredibly sensitive and careful around how health care is treated.
Yes, there are other types of transactions and data that are perfectly well suited for software as service applications where geography isn't as important. However, health care has some incredibly importantly data, of course. Many of our energy companies have decades of seismic data that needs to be protected, and there's intellectual property that needs to be protected.
There's a real data classification exercise that needs to run by sector. Some of those we can push to AI services with more ease than others.