How to use AI to augment Agent Intelligence
Every Cx Manager we have met struggles to analyse large volumes of customer queries, discover what trends, what needs to be fixed or escalated. This is because their analysis and reports rely on tags manually applied by agents to each customer conversation, and these tags are simply not reliable. You either ask agents to navigate a complex maze of tags and they end up applying randomly the same top 10 tags all the time, or you limit yourself to a choice of 10 and you lose any granularity in your vision.
So this is where us AI vendors come in and say: Fear no more! We shall classify while thou shall have coffee!
But it’s not that simple. Indeed AI is great at understanding language patterns in customer conversations to spot the ones that carry the same intent, describe similar problems, about similar things. AI is great at tracking every details of each conversations, spotting which brands, products, features or issues are mentioned and how frequently.
But when it comes to figuring out who’s fault it is… well, AI is little bit dumb.
AI for the masses
A good example would be a business customer complaining about a complicated software they use. They could be saying “I went to the report screen and could not print it”. AI will be super quick at identifying and correlating “report screen” and “could not” and “print”.
But who’s fault is it? Why is this customer not able to print: is it insufficient user privilege? A bug with the button not displaying correctly? General outage in the printing cloud service?
This would be good to know at the end of the month. Once you discover there were thousands of similar queries, what will you decide to do: train the agents on how to print, escalate a bug to Product team, tell IT to improve their printing server uptime?
This is where Agent intelligence comes in, and will beat AI all the time: support agents know or can figure out over the course of each support conversation whether the user has the wrong set-up, whether it is a bug or a general outage. What will be difficult for them to do is manually tag all the details that are so important too, such as “report screen” and “can not print”. It sounds easy to do on one conversation, try it on 100,000 with 50 agents tagging differently from a choice of 250 reasons to call.
How it will help the agent
However, it is totally feasible to train every agent on how to apply 5 tags correctly: “general question”, “bug”, “user config”, “feature request”, or “outage”.
And it’s really for AI to discover all the other details: button, printing screen, shipping company, integration, IT assets, products type (shirts, trousers, etc.), and so on.
Then, marrying the two gives you an incredibly rich perspective, incredibly easily, and incredibly fast:
- From all the customer queries about “printing”, how many were related to bugs, product knowledge or outage?
- From all the bugs we got reported by customers over the last 6 weeks, what were the top 10 features mentioned in these?
This is where AI becomes the great augmentation layer to Agent Intelligence. Working together makes the job easier for both. And ultimately makes your Cx Manager job easier.