10808_AI_Bussiness_School_ExtraVid_FINAL_v3

[AI Business School] >>Norm Judah: One of the interesting elements that comes out of the maturity model work that we did was what is the right organization model as you set up a center of excellence. And, then obviously a different model, one is very centralized, one is heavily decentralized with some hybrid in the middle and so the centralized model would be reporting in possibly to the chief digital officer or the chief information officer. The center for excellence for AI that does strategy governance but also executions so because that is a scientist so hard to define, we cluster them into one place and do it centralized but the projects actually still happen in the business silos or in the business divisions and these people sort of pause down. There is this other model which is completely decentralized which says you have no centralization and governance, but each one of the business units actually has their own mini groups so you have a set of local standards for each one and local data scientists that happen there as well. In that case, there is the propensity of course to develop different standards and different silos and so when I actually prefer at this point is the hybrid model which says from a central perspective you have strategy and governance, model management the ownership of the maturity model, the activities of that, but then execution actually sits in the individual silo, so the application architects, the data scientists all sit in a silo in a business silo wrapped around a particular problem, but the only way that that is successful in reality is if we still form the community. The community of the central governance but also the architects on the individual team bringing those people together to make sure that you are sharing data models to make sure that you are sharing data itself, that you have understanding of the ML models, you have understanding of the problems that that things solve because one of the things that AI will do is you will get insight in a big data science from one cluster of data that becomes useful in another cluster. This thing that you learn about credit card processing in a bank that might be useful in retail customer, so this has to be point of understanding of the breadth of data and the breadth of the model, so today, the model that seems to be most effective is this partially centralized decentralized, we have governance and strategy and operational technique coming from a central org, but then the applications being developed by AI experts sitting in the business as close as they can be to actually get into that [INDISCERNIBLE] agile mechanism of developing great solutions today. [Microsoft]