The pyramids, somewhat similar to Maslow's Hierarchy of Needs (in this case, the needs for talent), also reflect the demand-supply realities in the AI space. For example, for every 1 person to qualify for the L4 level, one can get about 3-5 people at L3, about 30-50 people at L2, and about 300-500 people at L1- (all ballpark estimates, not cast in stone, no formulas here, will vary by complexities in tech used and domain) in both the pyramids- be it technical skill-mix or domain-skill-mix.
So, in target resource planning, if one may start with X at L4 in both pyramids:
- L3 will most likely be in range of 3X-5X,
- L2 will most likely be 10X of L3, and
- L1 will be about 10X of L2.
These ratios more or less reflect the current realities in the talent supply chains as well, given the over-enthusiastic coders who want to ride the AI hype thinking Python coding and AI are synonymous, and often shying away from the hard maths of data sciences that's required, to go up to L2 and so on.
Also, the other critical aspect regarding talent is to acknowledge the fact that in most use-cases and applied AI scenarios, domain expertise is more important than just generic AI/ML technical skills. This is because of the fact that empty shells of algorithms cannot deliver any value without relevant data models, knowledge graphs, good knowledge items (nodes in the graph), training data- labelled/ annotated or not, domain lexicons and ontologies. SMEs of domains are the most critical talent levers that enterprises can ill-afford to ignore.