Let’s set the stage. You are a risk executive at a large commercial bank, and you receive a news alert that Silicon Valley Bank and Signature Bank are going into receivership.
You immediately turn to the device at your desk or in your hand or on your wrist and ask it to create data visualizations in Power BI showing your bank’s commitment and outstanding balance exposure to those two banks in relation to your overall portfolio.
You then ask it to embed those visualizations in a PowerPoint presentation with your notes from the meeting that just ended with other senior executives at the bank. You also ask it to write an executive summary, attach your presentation, and create and send an email with the two outputs to the executive team.
Knowing that you and the team will need more details, you then ask Microsoft Excel to create spreadsheets with information like which borrowers are impacted, what are the commitment details for facilities in the top 10 impacted industries, and what are the outstanding loan balances by loan officer and risk rating.
With the additional details in hand and the executive team on the same page, you ask Microsoft Teams to schedule a high-priority meeting with heads of each business unit that appeared in the Excel output to bring the wider team into the loop.
Today, this kind of response would take many hours and more likely several days to organize and complete, but if the promise of Microsoft 365 Copilot holds true, all of this could be done in minutes. Talk about a game-changer!
Of course, for all of this to work, the information you use to train the AI models must be well-defined in the context of your business. Each data point must have a business definition that is specific and unique, and these definitions need to be accessible to the models.
This is where I see one of the biggest challenges for making data from online transaction processing systems available to AI modeling tools. In my experience, very few systems have consistent naming conventions and certainly do not include detailed business definitions within the product’s metadata. Without those business definitions, it is difficult enough for a human to understand the data they query from a system. How can we expect anything more from a AI model? This change in system design should begin immediately to take advantage of what looks to be an incredible shift in the way we are able to work.
This former Microsoft Office programmer could not be more excited about what we are going to be able to do with Copilot. Let’s go!