September Success Spotlight: Next Steps After Implementing AI for Issue Classification

By Tracy Oppenheimer

Signing the dotted line and implementing AI to start classifying issues is just the tip of the “AI-ceberg.” Once support teams are comfortable with the accuracy threshold that they have set for triaging issues, it’s time to ask, “what’s next?”

Customer Success Manager Mckenna Morey is here with the answers for this month’s success spotlight. She provides context for actionable next steps to take once your support team has started leveraging AI via Helpshift Predict.

Start Small with Objectives

Enabling Predict doesn’t mean you need to change your workflows immediately. Predict can label tickets and agents can provide feedback before any workflows need to be changed. This means you can optimize your model by incorporating feedback without affecting your workflows. Essentially, label first without any associated actions to make sure nothing breaks.

After optimizing the model through agent feedback, you’ll want to pick a measurable goal and measure the success of that goal specifically. For example, if you previously used keyword-based tagging for routing tickets to a queue, take note of the percentage of uncategorized tickets being sent to the default queue and how it decreases once Predict kicks in. This is a great goal because it is well defined and easy to measure.

Pay attention to reporting and specific metrics in order to make incremental improvements. You can view accuracy information for each label within the Helpshift dashboard. Due to natural variances in ticket data, you’ll want to monitor this on a weekly basis. If a specific label has a low level of accuracy, consider disabling the label or replacing it with more accurate data.

Make Providing Feedback a Priority for Better Modeling

The more feedback that agents provide and the more ticket data your model receives, the smarter it will become. Remember, Predict learns continuously, so it is critical for agents to be trained on providing feedback — and should do so before resolving each ticket (at least in the beginning.)

Once the model is “smart enough,” it can be used for more advanced workflows. This allows for new levels of segmentation as more granular labels can be used to prioritize increasingly specific issue types.

Set Up Predict to Hand Off to Bots to Automate Even More of the Process

Once a ticket is categorized by Predict, specific rule-based bots can be triggered to gather more information from your user before handing off to an agent, or auto-resolve certain ticket types. This gives the support agent more information to work with and increases rates of first contact resolution.

Over time, these actions will become increasingly automated through bots — so implementing AI successfully now will set you up for success!

Subscribe to Weekly Digest

Follow us on



cta-circles

Get started today