zapp2photo - stock.adobe.com
Amazon Connect integrations with the native Lex chatbot service last year enabled enterprises to offload many call requests that would otherwise be handled by humans.
Instead, those tasks are now handled by virtual agents -- called bots -- built on Lex. This combination extends Amazon's Alexa technology for voice interactions into enterprise call centers, and some early adopters have seen significant improvements in how many call requests their teams could handle, without the need for extra staff.
Despite some early limitations, it's possible to use these integrations to track customer interactions, including abandoned calls, average wait time, length of calls and agent performance. These metrics demonstrate how Lex applications, called intents, reduce the burden on humans and help business managers justify the development and improvement of Lex applications used in their call center workflows.
Analyze Connect data
Intents are a fundamental component of the Lex programming model. Developers associate the different ways in which a user makes a request with a particular intent, which is connected to a Lambda function to process the request on the back end. For example, the user input could be as simple as an automatic transfer to a particular call service team. More complex intents could retrieve information from a database or schedule an appointment without agent intervention.
Amazon Connect generates both historical and real-time metrics, called contact trace records (CTRs). These records capture events associated with a contact and can be used to analyze the number of calls handled or abandoned, as well as average wait time and other metrics.
Developers can set up more sophisticated analytics pipelines with other Amazon Connect integrations; for example, they can configure an Amazon Kinesis CTR export stream to transfer raw data to S3. AWS Glue can catalog and transform the data into a format that is easier to analyze with tools like Amazon QuickSight, Athena or Redshift Spectrum. This kind of insight can help quantify customer experience, measure agent performance and correlate these items against the introduction or improvements of Lex capabilities.
Know what to program in Lex
Native Amazon Connect metrics don't currently provide any insight into what users call about, which might be useful in Lex intents. However, there are Amazon Connect integrations with several third-party voice analytics tools that could provide this kind of information. Services such as CallMiner Eureka, DialogTech and VoiceBase can analyze the speech in call center interactions to identify caller intent. These services are often used to help coach human agents, but this data could also be used to automatically quantify the types of queries these agents handle.
However, this level of analysis might be overkill as an introduction to Lex intents. It's probably more practical to just ask the call center workers. For example, when the National Health Service Business Services Authority in the U.K. embarked on its first Lex-Connect integration, its team quickly identified requests about the European Health Insurance Card as a top priority. This focus enabled them to build their first Lex app in a couple of weeks and reduce call center traffic by 26%.
How to improve Lex
Developers and analytics teams don't currently have access to the voice samples of Lex interactions. This kind of data would make it easier to identify different ways that customers make a particular kind of request. It could also provide managers with guidance when Lex is unable to successfully execute a request. For example, callers to an airline might ask the following variations of a question when they request information about their miles:
- How many miles have I accumulated?
- What's my mileage status?
- What's my current mileage?
Business managers might want to include service agents on Lex development efforts to help make recommendations about the different ways users ask for things.
Amazon Connect integrations with Lex are still in their early phases. Early reports suggest they are suitable to capture numeric data from people or to give consumers options they can quickly navigate. However, some developers have run into problems when they try to capture more variable inputs, such as names and addresses. At this point, it's probably best to start with Lex intents that play to the service's strengths.
It could also be a good idea to track any gaps agents see with Lex. To do this, a developer could add a field in a customer relationship management application about Lex limitations that agents could fill in when they take a call. This would help enterprises identify what kinds of new intents to program into Lex and could also help identify cases when Lex couldn't understand customer verbiage used to make a request. Developers can then use this information to add new phrasing into triggers for specific Lex intents.