Building capability
Training
The goal of training here is that the organization carries this work itself. The people responsible for an AI decision should be able to make it, judge what the model produced, and hold the risk without an outside party in the loop. The four levels below are how we describe capability, and the fourth is where that ownership changes hands.
- personal skill
- the organization holds the risk
The four levels
- 01
Aware
Knows which AI tools are in use and what the organization's policy says about them.
Read in fullCollapse
Someone at this level can name the tools in use around them and say what the policy allows for each. Asked whether a particular kind of data may go into a particular tool, they either know the answer or know where it is written down. Regular use is not part of this level, and plenty of people here have never opened one of the tools.
- 02
Capable
Uses AI for real work and comes back with a result worth keeping.
Read in fullCollapse
The work at this level is real work rather than a demonstration. A person here takes something from their actual queue, a first pass at a supplier policy or a summary of a long report, and gets output they can edit into shape. What they cannot reliably do yet is catch an answer that is confident and wrong.
- 03
Fluent
Matches the approach to the task, and recognizes the tasks that should not go to a model at all.
Read in fullCollapse
Judgment is what separates this level from the one below it. A fluent person picks the approach to fit the task and can say why another approach would have been wrong for it, and turns work away from AI when the input cannot leave the organization or when an ordinary script would be exact where a model would only be close. The visible sign inside a department is refusal, since fluent people are the ones saying that a particular task should not go to a model.
- 04
Accountable
Owns the decision, supervises and defends AI output including someone else's, and carries the risk without outside review.
Read in fullCollapse
Everything below this level is personal skill. At this one it becomes the organization's: a person here approves output a model produced, and output a colleague produced with a model, states the basis on which they approved it, and holds that position when an auditor or a customer asks. A department with people at this level does not need an outside party in the loop for this work, which is what the whole progression is for.
Role tracks
Executive
An executive finishes able to question a model-backed recommendation before approving it and to state which risks the organization will carry, holding a named person to the decision rather than the tool that informed it.
Department head
By the end a department head can place their own people on the four levels above, decide which of the department's work is appropriate for AI, and defend that line to the people doing the work.
Analyst
Analysts leave able to check a model's output against the source data it claims to rest on, and to say which part of an answer is evidence and which part is the model filling a gap.
Engineer
For an engineer the outcome is a system that fails visibly, one where a model sits only where a model belongs, a deterministic step sits everywhere else, and enough is logged that a bad output can be traced back to what produced it.
How training appears in the product
Training is one of five roadmap categories, and whether a roadmap carries one is the model's choice. Items are titles and descriptions, with no course or completion record.
Read in fullCollapse
In the platform, training is one of five categories a roadmap item can carry, alongside security, governance, operations, and technology, and whether a given roadmap carries a training item at all is the model's choice at generation time. A roadmap with no training item is a valid result. Learning paths are generated separately from roadmaps, organized by subject and phase across generative AI, predictive analytics, data readiness, decision integration, and responsible AI. An item on either one is a title and a written description, so what the platform holds is the plan for the work, with no course, no completion record, and no note of where anyone stands on the four levels above.
What the answer decides
Answering the assessment teaches as it goes. Each question carries its own description, branching picks the next question from the last answer, and ideas from respondents carry votes.
- the assessment's mechanism
- the people answering
Read in fullCollapse
Answering the readiness assessment is itself a teaching pass. A question carries its own description, and branching rules decide what is asked next from the answer just given, so someone who has not met a control before meets the definition at the point it is being asked about. Questions also accept ideas from the people answering, and those ideas carry votes, so a disagreement between two teams about the same policy shows up inside the assessment rather than after the report.
What training does not fix
Training supplies no policy and no decision owner, and it is not an input to the secure score, which holds no record of training at all.
Read in fullCollapse
Training does not supply a policy, and it does not supply a decision owner. A department of fluent people with nobody named against the decision only produces unowned decisions faster, and no amount of skill closes that gap. Training is also not an input to the secure score, which reads assessment answers, completion rates, department membership counts, and submitted ideas, and holds no record of training at all.