Has the time come for AI in support?

After decades of dashed hopes, AI techniques are, finally, yielding results for support teams. Not the utopia of automated issue resolution, but many useful insights and assists are now possible, thanks to better technology. For instance, AI can tell us:

  • Whether a particular customer is technically knowledgeable, or historically demanding, or upset today–all information that is immediately useful to the support engineer trying to help that customer (and not easily captured in the CRM system).
  • Where to route cases based on their descriptions and the requesting customer, with a level of refinement that’s much superior to what we can obtain with metadata alone. For instance, do not route upset customers asking about feature X to a new hire.
  • Whether a case is likely to escalate. We know that support engineers should sense that, but some do not, and others are afraid to let it be known that they are struggling.
  • How skilled support engineers are at troubleshooting and communicating. Yes, with AI you can monitor all┬ácases to get a full picture of quality.
  • Why customers contact support and how best to help them proactively. Support organizations have evolved complex systems for capturing and analyzing root causes, but if we could read each and every case, and AI systems can, we would get to much finer levels of detail.

A few things to keep in mind as you venture into AI experiments.

  • Your data may get a very useful assist from outside data. For instance, gauging the sentiment or the technical level of a customer may be aided greatly by their LinkedIn profile (or even their Facebook profile, for consumer support).
  • Classification matters. For instance, how you define the cutoff between “competent” and “expert” on the technical scale will make a big difference in the types of customers classified as experts — hence how you can route their issues.
  • You need to understand the data and business deeply. For instance, cases coming from particular geographies may not follow the same rules for determining sentiments, as they are culturally more abrasive (or less).

Interested? Here are the steps to follow to get started with AI.

  • Define the problem. What exactly would you like to do with AI? Help individual support engineers get a more nuanced view of customers? Anticipate escalations? Analyze root causes? Each use case will require different data and approaches.
  • Develop a model, either using the functionality offered by a vendor or by creating your own. If you are focusing on problems that are unique to you (e.g. analyzing log files or clickstreams), a custom project is best. If you are looking at a more mainstream analysis (e.g. extracting signals from support cases), you will save time and money working with an established vendor.
  • Train the model, using relevant data (which will need to be cleaned ahead of time). Machine-learning models are very sensitive to the dataset so make sure that it is trained with yours.
  • Enjoy

Have you started AI experiments? What use cases are you focusing on and what outcomes are you seeing?


Tagged under:

Leave a Reply

three × three =

Your email address will not be published.