Everything you need to know about Intents

What are Intents?

‌An intent is the purpose of a message or ticket sent from a customer. Y Meadows uses NLP (Natural Language Processing) to read the message, and derive the intent or point of the message.

Users train the AI models with specific Intents. For example, if a service team has a high volume of change password requests, they would create an Intent for that use case. Once the model has been trained it should be able to accurately categorize messages with that Intent, even when keywords or sentence structures are completely different.

​

‌

Some Intents can be easily recognized, like "password reset" requests, but most are specific to each organization. Most organizations have their own method of categorizing support messages. These categorizations are usually a good jump off point for creating Intents in Y Meadows. For example, an airline might have messages coming in that agents tag as "missing luggage" or "cancellations". Those categories might be usable as Intents, or could be broken down into even more granular sub categories that could be individual Intents (like lost luggage and rerouted luggage). That would be primarily based on if different scenarios in those categories might have different resolution processes.

‌Choosing the right Intents:

‌Now that we know what an intent is, we have to carefully select the best intents to maximize the impact of Y Meadows customer service software.

There are 3 main criteria for choosing what intents to focus on in the implementation stage.

• Clearly defined Intents

• Common or repetitive messages

• Simple resolution actions

‌Clearly defined intents are essential to getting accurate results from the NLP models. A good way of finding some initial intents to start with is to ask yourself some basic questions:

• What are the most common tasks your team deals with

• What tasks are very time consuming? Do they require educated guesses or creative solutions (if they do, they won’t be good prospects for automation)?

• Do we have tasks that involve “swivel chair” (multiple steps or logins) steps to acquire the information needed to resolve the message? Can we automate that to improve team efficiency?

• Complexity - judgement or questions that require deep thought, advice, or educated assumptions are not good candidates for automation.

In Summary:

‌Choosing intents sounds far trickier than it actually is. Typically, internal support teams already know what messages they would like to have automated, even before they become familiar with the Y Meadows process. As teams grow more accustomed to the software, they become much more proficient at filtering out what messages to select and which should be handled by agents.