CAI Automated Drive-thru | Proof of Concept

Objective
Reinvent the Drive-thru Experience by enabling the voice of ‘The Quick Service Restaurant’ with an increasingly capable virtual cognitive agent (VCA) that begins with the drive-thru and evolves to greater store and online features/capabilities.

Approach
1: Research customers' current ordering behaviors and restaurant fulfillment processes.
2: Stand up a “Quick Service Restaurant” collaboration team to design and develop a POC.
3: Govern the program and manage the technology foundation and brand integration. This includes partner technology stacks, analytics and AI components, creative teams, and future state virtual experience.

Team Structure

Innovation & Design Team Roles:
*One person may perform multiple roles.
Program Navigator, Digital Strategist, UX Researcher, Conversation AI Designer, Visual Designer, Client, and Client Partners.

Tasks:
Research, ideate, and design proof of concept (POC) and draft minimum viable product (MVP).

Development & Delivery Team:
Project Manager, Platform Architects, Developers, Linguists and Testers.

Tasks:
Deliver and deploy the POC.

Research & Design - Process Summary

Step 1: Understand and Communicate Complexities

Building a Conversational AI (QSR) Quick Service Restaurant ordering system is very different from building a simple question and response system. Interactions are much more conversational, and many contain multiple intents. Recognizing “multiple intents” is an issue with current voice systems.

Example:
You can ask Alexa “What time is it?” and she will tell you.
You can ask Alexa “What the weather is like outside?” and she will answer.
But.. Ask Alexa “What time is it and what’s the weather like outside?” she will respond “Sorry I don’t know that one.”

Additionally is it more challenging than building an ordering app. Apps have defined touch/input points plus strict user flows and thus can easily control how input is received. Conversational interactions are “looser” and must be designed to “attempt to guide” by using audio and/or visual tools that prompt the user to input the data in a manner conversational AI systems can understand.

In the event the “attempt to guide” fails “safety net protocols” must be in place to take over the interaction.