Drive-Through Chatbot

Half a decade before the AI hype took hold a mid-market fast-food franchise attempted a voice-activated drive-through. We were retained to help build it, and the protype worked.

The primary journey was a converstional flow between customer and chatbot. Conversational flows and rules engines were required to model to deliver successful customer experiences.

The information architecture required a establishing an ontology and taxonomy. Additionallyh we documented mind maps and conversational flow diagrams.

We needed a way to bring a human into the loop, which meant understanding sentiment and giving customers an obvous method of ejecting. This is where many other human factors emerged.

Signage

In addition to notifying customers of an escape mechanism we also needed to provide a privacy notice. This project happened as GDPR was rolling out in Europe and as the California Consumer Privacy Act (CCPA) was being legislated.

While the system would be documenting the conversation to train the model, we were obligated to ensure customers that no personally identifiable informaiton (PII) would associate a specific person to an order or conversation. This could best be done with a series of yard signs placed at drive-through entrances.

Menu Boards

Some restaurants had modern, digital LED menus, while others used physical menu boards. There were a handful of variations.

A specialist in outdoor advertising was retained to help understand the new customer expereince with menu boards. It could have been a massive shift or a non-event. We concluded that digital menu boards would be best as they could provide dynamic messaging. This led to another issue: who would design that expereince and where would it be managed?

Employee Expererience

Our research concluded that in most cases the most talented team member on any shift is often tasked with the drive-thorugh. Over 70% of orders are through this channel, so this person is usually handling the majority of the business. Additionally this person is doing at least four other activities in addition to speaking with customers in their cars.

This chatbot would help simplify this person's alreay complex workday. This is how automation can improve both quality and worklpace safety.

The person tasked with the drive-thorugh would then become the human-in-the-loop. We therefore derrived an alert that would allow every employee to passively monitor the drive-through and quickly act when a human needed to intervene.

Restaurant Environment

Fast-forward to 2025. I am enjoying breakfast at another fast-food chain that has recently introduced a drive-though chatbot. For over 30 minutes the only audio in the store is the chatbot repeating the same conversation over and over.

While it's important for the crew to be monitoring the conversations, the in-store customers do not need to hear any of it. Our prototype channeled all of this through the headsets.

This was built on Amazon Web Services. We also needed to integrate with the menu, which was a homegrown database, and the headset/intercom system.

• Interviewed dozens of humans who would be using the product to understand how they could benefit
• Engineered a $300 audio bypass to separate internal communications from the learning model
• Authored thousands of test cases for proving out all permutations