Artificial intelligence is powered by well annotated training data, but the data you need to train a customer support assistant includes more information than you might expect. The "tribal knowledge" of your customer support teams is critical to developing a customer service AI agent that can most effectively help your business.For example, it is very likely that your customer support knowledge base has a lot of outdated or redundant content. (Most knowledge bases do.) Your customer service team probably knows which articles and entries are the "right" ones to reference, and they often provide customers with ad hoc advice that isn't explicitly in the content ("Mac users should skip to step 2," or similar advice).
Capturing this knowledge about your support documentation is often vital to training a customer support AI solution. If the AI knows which content items are correct -- and what content is missing -- it can better tune its own recommendations and performance. With enough tribal direction, the AI can even learn to distinguish good support content from bad and can direct your documentation team on how best to clean up your knowledge base.
Tribal support knowledge is also invaluable to training an AI system on your support process. In many organizations, support team members have unwritten policies or unspoken rules for accelerating or optimizing the support process. Training an AI on these unwritten rules is critical if it is going to adequately supplement the work of human support reps.
For example, if your solution is heavily used in educational settings, your support team may expect a rush of novice users in autumn when school starts and might informally assume that many user-reported "bugs" in that time period are in fact just new users misunderstanding how the solution works. Instead of taking bug reports at face value, your team may place more emphasis on confirming the user understands how your solution should properly function. More simply, they'll be ready for a seasonal rush of RTFM issues.
It's unusual for this kind of experience to be formally documented, but if your AI agent is allowed enough time to train alongside human support agents, it will likely pick up on this pattern and add it to its own internal issue-diagnosis algorithms.
Adopting AI is a different proposition than adopting other SaaS solutions, as AI takes time to ramp up and needs a different kind of data to work best. Part of this adoption cycle means capturing as much unwritten "tribal knowledge" from your team as possible, and that need is especially high amongst customer support teams (which have a lot of unspoken shortcuts and job-hacks from which AI could benefit).
An AI support assistant that can sit and learn alongside your team may not seem impressive at first, but this exposure to your secret support shortcuts will improve its performance in the long run.
Talla is building an AI assistant designed specifically to support customer service teams. If you'd like to start your journey towards next-generation customer service automation, contact Talla today.