AI product due diligence

AI is catalysing the inception of new businesses, the rollout of innovative features, and the influx of venture capital funding. However, as we now know, it’s evolving so rapidly that companies risk obsolescence overnight. In this ever so dynamic environment, what are the crucial questions I consider during a product workshop?
 
I would split my agenda over 3 aspects (a) technology, (b) moat and (c) commercial
 
Technology
 
- are the models proprietary or piggy-backing on other models? if the latter, are they based on SOTA models and can you adapt as new models are released?
- is it true AI? or older school algorithms (i.e. ML, NLP, Rule-based decision...)
- how is it trained? can it scale? (i.e. is there a human-in-the-loop component to take into account - for annotation or enrichment purposes)
 
Moat
 
- is the data for training accessible (think copyright), of good quality (i.e. annotated), and in sufficient volume?
- is there a unique/proprietary dataset generated?
- what are other potential moats? Think UX, embedded workflows, industry expertise,…
 
Commercial
 
- is AI critical to the product/platform or is it a feature?
- does it help create clear and sustainable value to the customer?
- is it being used by high % of the customers?
 
These points are only supplemental to the standard questions typically explored in product dd sessions. To the very least I try to touch on: org. structure, talent and development velocity, feature set and how it solves the pain point, roadmap (historical and forward-looking), deployment and roll-out, ROI and architecture / security / compliance posture.