Triage is a complicated process, and it is likely that you've already implemented some level of automation. If you are wondering how to leverage artificial intelligence with your existing workflows, this article is for you.
We'll cover a few pitfalls and best practices on how to combine the best of both worlds. Forethought will advise you during the design of your unique solution.
Combining the best of both worlds
If you're wondering whether you should use workflows/triggers or AI, the answer is both. Think of AI as the building block of your solution. On one hand, Forethought Triage models excel in language understanding. On the other hand, if you can frame the issue with an IF THEN ELSE business rule, why even bother using AI?
Example:
"If a ticket is about a refund AND it hasn't already been refunded, THEN send to the refunds department."
- Understanding that a ticket is about refunds is likely to require artificial intelligence, especially if the question is long and complex.
- Understanding that the ticket has not already been refunded requires looking at previous transactions for this user, which can be defined by an "if" statement.
Therefore, the solution will leverage a combination of workflows and AI, but which one comes first?
AI is the Forethought, not the aftermath
If someone had to read the ticket to fill out some fields/metadata before the Forethought actions, we're probably not doing it right. In other words, AI is the very first process step, before humans and rules-based workflows.
If your process is complicated and involves multiple business units passing information from one to another, consider putting the AI as early as possible in the chain to maximize its value.
Example:
- A ticket is created in your help center. It contains a subject and a description: "Hi, I would like my money back for the service that I never received."
- Forethought reads the ticket and routes it in the right "Refunds" queue before it is read by a human.
- A workflow rule then triggers in your system that a refund has already been processed. It marks the ticket as "already refunded" and sends an email to the customer.
Bottom line: AI is used as the first step of the process that will then allow for a set of automations. When building the Triage model, it is therefore very important to ask ourselves:
"What information do we have when the ticket is created?"
This will drive the inputs that we use to train an AI model.
Focus AI on language understanding
Process example
Let's suppose your organization triages tickets based on the issue type field. The possible values you tag them with are the following:
- Technical
- Refunds
- Approved
- Product
- Already Refunded
The process is the following:
- Everyone in the team spends 2 hours a day triaging tickets.
- When something is sent to Refunds, the department will process the ticket and mark it as either:
- Approved
- Already Refunded
Problem
If the model is trained to only read the text of the ticket, it will have a hard time making the difference between Refunds and Already Refunded. This workflow status is unrelated to the content of the ticket and the AI would struggle to make the difference just as a human agent who is unaware of the process would.
So can we train Forethought's AI to use this information as an input?
Technically, yes. Forethought allows for multiple input models. However, if this information requires human verification later in the process, this defeats the purpose of AI to eliminate the need for human intervention.
Solution
The solution is to focus the AI on language understanding. To do this, we'll narrow the list of issue type values to:
- Technical
- Refunds
- Product
A complimentary workflow status field will have:
- New
- Approved
- Already Refunded
Forethought will focus on the issue type field to send tickets to the right team. Automated workflows can pass the workflow status field.
Doing so will tremendously increase the accuracy of your Triage models while being non-disruptive for your business.