The question first, the technology second
Most enquiries about artificial intelligence start from a technique and then look for a problem to attach it to. I turn that around: what is meant to get better, for whom, and how would anyone notice that it had? Once that is settled, it can be checked whether an AI method is the right means, and often enough it is not.
That check is cheaper than a project that runs aground on the data situation a year in.
What I bring
I explain the basics as far as they are needed for deciding: what machine learning does, where its results come from, where its errors come from. No further, because you are meant to be able to decide, not to build it yourself.
For designing an application I use cards rather than code: the AI Toolkit lets a team play through the idea, the data sources, the interaction and the learning objective without anyone needing to program. What comes out is a concept prototype solid enough for a funding application or a tender.
Ethics is a design question
Fairness, accountability and data protection cannot be checked at the end like spelling. They are decided where it is settled which data gets used, who sees the results and who is allowed to object. That is why those questions come early with me, not at acceptance.
What I am not the right person for
I do not train models and I do not build infrastructure for running them. Once it is settled what should be built and only the implementation is left, you need a development team, not me. Until then, in the phase where the decisions are made, I am useful.