5 minute read

I recently attended the M365 Conference in Las Vegas where AI, specifically Microsoft's Copilot, dominated the keynote presentations. Copilot is a suite of AI-powered products, including Copilot for Office, which helps with composing Word documents, creating visually appealing PowerPoint presentations, and performing complex tasks in Excel, and Copilot for SharePoint, which designs professional-looking SharePoint web pages based on what a user types into the Copilot text box.

This emphasis on AI from Microsoft is reminiscent of their pivot towards the internet in 1995 when Bill Gates famously declared that "The Internet is the most important single development to come along since the IBM PC was introduced." Bill is no longer the head of Microsoft, but CEO Satya Nadella asked him for his guidance, and Gates has recently written that "The development of AI is as fundamental as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone. It will change the way people work, learn, travel, get health care, and communicate with each other. Entire industries will reorient around it. Businesses will distinguish themselves by how well they use it."

While Microsoft has not announced pricing for Copilot, it is likely to be a significant investment. OpenAI's ChatGPT, for example, costs $20 per month for a subscription to GPT4. It is reasonable to expect that each Copilot use-case will fall within a similar range.

So, who will pay for LLM solutions from Microsoft? The answer is simple: those who get more value from it than it costs. The first formal study showing the value of AI has been published by authors from MIT and Stanford, which found that contact center agents using AI had greater productivity, job satisfaction, and improved client outcomes. This is a win-win-win for AI and the business involved. Interestingly, the AI assistant was most helpful to newer contact center agents, helping them to act like much more experienced agents from day one.

The authors of the study did not name the Fortune 500 company or give many details about cost structures, but they referenced a very high industry average turnover of contact center employees each year at a cost of $10,000 - $20,000 per agent. This experiment was done with 5,000 agents.

Doing some rough math here: If the company loses 150 people each month at a cost of $2M per month, and reduced attrition causes only 100 people to leave each month (50 fewer than before), that's a savings of $670K per month. If the AI tool for those 5,000 agents costs $20 per month, or $100k per month total, the ROI is almost 700%. And that's just the savings on attrition. There are also productivity savings and additional value in customers getting answers more quickly and being happier.

Measuring the impact of Copilot for workers who are not in contact centers will be tougher, but we need to come up with methods to do so. If you are in a company with 5,000 employees, you will have to justify an expense on the order of $100k per month (or potentially much higher if you need multiple Copilots in your business). Building a business case will be essential to justifying the investment.

Some organizations may wait until they have iron-clad justification for investing in AI, but they risk getting left behind by competitors who could become much more agile and productive almost overnight. The best companies will work on developing measures to ensure that money is not being wasted and will rely on business models that show the potential, but they won't wait to start working with Copilot in at least some scenarios as soon as it becomes available.

AI Credits: The image at the top was created by Microsoft Designer, prompting it with the title of this post. When I started writing this post, I asked ChatGPT for an outline based on the topics that I wanted to cover. The outline got me over the hump to start writing, but I didn’t end up using any of its suggestions. After I finished writing, I passed the entire piece through ChatGPT to make it more concise. I made several edits of the output and then published this post.