Go to Top

Natural Language Processing based Workflow Engine for Customer Inquiries

The client was into providing ecommerce services, where it was facing major issues in dealing with post-sales customer service. Most of the customer queries were coming in via direct email or indirect email generated by IVR (Interactive Voice Response) system. The client was dealing with 5000-6000 such emails on daily basis. Given that it was a small organization with only 15 people in post-customer sales response team, it was a major challenge to deal with such a large volume on daily basis. This was resulting into lot of customer satisfaction and retention issues.

The solution was to devise a NLP (Natural Language Processing) based system, which could analyze emails in Incoming Queue and tag each email with a product, inquiry type and corresponding post-sales response handler team. Once an email is tagged, it was moved from Incoming Queue to Unprocessed Queue with the information on product, inquiry type and post-sales response handler team.

Once email is available in Unprocessed Queue, people assigned to that handler team can accept the customer inquiry one by one. Once an authorized person (in the corresponding post-sales response handler team) accepts customer inquiry, it is removed from Unprocessed Queue and moved to Assigned Queue. Authorized person would take action such as sending email response with requested information to customer or processing order cancellation.

Once authorized person completes processing of customer request, it is removed from Assigned Queue and email is sent to customer. At the same time, a backend cron job monitors Unprocessed Queue and Assigned Queue on business rules as per SLA (Service Level Agreement) for each category of customer inquiries. For example, each customer inquiry must move from Unprocessed Queue to Assigned Queue within 2 hours. In case of exceptions, when such business rules are not followed, email notification is sent to the supervisor.

This video interview solution was developed using LAMP, Amazon Comprehend, Amazon SQS (Simple Queue Service), Amazon S3 and Java. From start to finish, it took four months to provide this solution.

Once the solution was implemented, client was able to process 5000+ customer inquiries within same business day.