Ross Smith interviews Jason Weum on the deployment of artificial intelligence in support at Microsoft. 

 

Ross Smith: Hey Jason,. as a leader in the deployment of AI in Customer Support, I’m interested in learning more about your background and how you started engaging with AI..   

 

Jason Weum: My background is in software development and engineering having spent a large part of my career both developing software (custom & boxed product) as well as managing the process through Technical PM and Product Management roles.  I have spent time in consulting, development, PM, testing, and support roles throughout my career.  Most recently I moved into the Supportability function within Microsoft Support where we focus on crafting an AI-Frist Customer support experience while making our Support Engineers more efficient and effective through developing cutting-edge solutions. 

  

Ross: Wow, that’s great – you have both and engineering and support background. ,  so you haveThis gives you  a good great perspective on this new world of AI that blends cutting edge technology and support practices. We know that Customer Service and Support is a top enterprise scenario for  the use of AI. When did you first get involved with AI? 

 

Jason: My team started discussing how we could disrupt shake up some of our current plans in support by leveraging AI back in the summer of 2022.  We were laying the initial foundation of understanding for using AI in Customer Service.  Then in December of 2022, ChatGPT burst onto the scene and because of our workwe’d already done a bunch of worksignificant research, experimentation, and work  in this space, we were primed for seizingto seize the opportunity.  Quickly in January 2023 we started building models based on our data sets combined with the power of GPT 3.5, and learning as the technology was changing so rapidly.  It was and continues to be quite a wild ride. 

  

Ross: So you were ininvolved with experimenting with AI this early on, before ChatGPT was even introduced to the world. You’ve clearly got a lot more experience than most. What are some of the things you’ve learned? 

 

Jason: One of the most important things I learned is that AI, like many technologies, has a learning curve for users to see the promised efficiencies.  Early on many people thought it was this magical tech to solve all problems and, while it is amazingly powerful, it does take investment to realize the potential.  Another key learning is that there isare certain things AI is really good at and others where it tends to struggle.  Understanding the parts of your processes or tooling that are good candidates for AI is key to unlocking the most productivity gains.  Finally, AI is only as good as the model or knowledge source you feed it.  We learning learned that having a good set of high-quality content is important to ensuring the best possible response quality. 

  

Ross: Sounds like you all have learned a lot about what works and what doesn’t. Can you tell us all a little more about where you see AI totally shine and where it tends to not perform as well in customer scenarios? 

 

Jason: In the customer service space, we see AI perform well at tasks that typically require less deep knowledge about the thing you are supporting.  For example, helping draft high quality customer emails, helping create initial troubleshooting or scoping emails to your customer, categorizing your support case based on a defined taxonomy, answering customer questions on how to use a particular feature –, basically things that typically take up time in a support engineers’ day but don’t require deep specialized knowledge.  Conversely, deeply technical nuanced issues, issues specific to an individual implementation, issues on new areas that aren’t well documented, these are candidates more suited for your human engineers. 

  

Ross: You have a long history with building diagnostics and automation to help customers help themselves with their technology issues. How will AI be integrated or will it be different from traditional diagnostics we read about here Save time and effort with Microsoft 365 self-help diagnostics – Microsoft Community Hub 

 

Jason: Right now, we believe AI will help in a few ways.  First, AI continues to help our developers generate code more efficiently, allowing them to accomplish more in the current methodology of development.  Secondly, we see diagnostic services being able to leverage LLMs to perform tasks that would normally be embedded in business logic.  For example, we have diagnostics that analyze large log files looking for predefined patterns.  Going forward we will be able to leverage the pattern recognition and categorization capabilities of AI to be able to find new patterns in logs files we haven’t encountered yet.  This will shorten the development cycle and increase the effectiveness of these diagnostics.  Lastly, we can use AI to generate natural outputs along with advanced troubleshooting steps that evolve over time.  With diagnostic results being able to prompt an LLM for additional insights we can ensure diagnostics can provide the most up to date guidance in a low code way. 

 

Ross: This is such a fascinating time of technological change and rapid advancements. What are you excited about for the future? 

 

Jason: As many have said, we are just on the verge of this AI revolution and what it can unlock for do for many around the world.  I am excited to be a part of one of the most exciting times in computing history and being able to learn every day as new things continue to be discovered.  Many problems we have tried to solve in the past with marginal results now seem much more feasible.  I am excited to see where we can go, what we can do, and what problems we can solve in the months to come. 

 

Ross: You have clearly come a long way in your AI journey, Jason. What advice would you share with customer support professionals who are trying to figure out how to deploy AI in their organizations? This is an intimidating space. What should people do to get started? 

 

Jason: AI is another tool that can be used to help you solve business problems.  While it is powerful and exciting, it won’t address everything all at once.  I would recommend being clear on the challenges in your organization and then seeing how AI might help.  Make sure you have a foundational understanding of where AI excels today and take that into careful consideration before planning out where to start.  Also, be aware that AI will change your existing processes and you may need to rethink how things work to get the most out of these new capabilities.  I like to start small and learn, building out based on positive results.  Finally, remember that any AI solution is only as good as the data in the model.  Taking time to ensure you have quality knowledge source content first will go a long way in ensuring the results AI delivers meet expectations. 

 

Related: ChatGPT & AI: 6 Tips for Managing Support Content (microsoft.com) 

 

Ross Smith is the Worldwide Support Leader for the CSS Modern Work Supportability team.  

 

Jason Weum is a director of supportability for OneDrive and SharePoint support, and he led strategy development for Modern Work GPT.