The problem:
How to make lemonade when IT gives you lemons.

The situation:
[24]7.ai was asked to add a chatbot to the phone and chat agents and services the company already provided for SiriusXM. The company’s Artificial Intelligence Virtual Agent (AIVA) integrates with popular customer relationship management (CRM) solutions, enabling transactional workflows. Users could manage their subscriptions, reset passwords, make payments, and diagnose some technical issues all through the chat window.

The technical challenges were easily addressed, but as we finished our presentation of the first transactional dialog, the representative from IT said, “There is no way I’m letting any chatbot connect to our systems.”

Lesson #1: Get IT’s buy-in before the ink dries on the contract.

The solution:
Eventually, we were given access to the API that resets a user’s radio.

When a satellite radio “forgets” its ID number, it can only receive two channels: the promotions channel and the demo channel. So, users will describe this problem as, “I lost all my channels.” These types of utterances would get predicted to the “radio_reset” intent.

The chatbot acknowledges the user’s intent with passive confirmation, authenticates the user with a secure card, then presents the user with a list of radio ID numbers (family accounts may have multiple associated radios). The user can select the radio IDs that need to be reset and send them to the system for resetting. These jobs are batched before they are sent to the satellites, so it may take a few minutes to get enough requests. A user may tell the chatbot that they don’t see the ID of the malfunctioning radio in the list, which would get them escalated to an agent. An escalation may also occur if the reset fails.

The outcome:
The radio_reset workflow was successful, but because the rest of the intents were FAQs, the client didn’t realize the expected cost-savings and the project ended after a few years.