Taco Bell
drive-thru voice ordering agent integrated with the POS
Taco Bell deployed drive-thru order-taking voice AI (Omilia) in about 890 restaurants across 38 states, with transaction times on par with or faster than human order-taking.
Key points
- Voice AI for order-taking at the drive-thru, integrated with the POS and the digital menus.
- Omilia Voice AI solution connected to Yum!'s Poseidon POS.
- About 890 restaurants across 38 states, transaction times on par or faster.
- Evidence A, mixed-signals status (viral incident summer 2025, rollout slowed then resumed).
Objective
Take drive orders through a voice AI to relieve crew members, keep service fast and consistent at peak hours, and free staff for greeting and preparation.
The deployment
Taco Bell is deploying a voice AI, provided by Omilia, that takes orders at the drive-thru. The system understands natural language, handles order modifications without a fixed script, adapts to stock in real time and to limited offers, filters road noise, and handles different accents. It is connected to Yum!'s proprietary Poseidon POS and the digital menus. Starting from more than 100 restaurants in 13 states in the summer of 2024, the deployment reached about 890 restaurants across 38 states. The rollout hit a publicized snag in the summer of 2025 (aberrant orders pushed by customers to trip up the system), which led Yum! to pause before resuming the expansion.
Results Proof A
Scale deployment announced by official Yum! Brands press release (100+ restaurants, 13 states in 2024) and confirmed as continuously expanding by established restaurant press (about 890 restaurants, 38 states). Primary brand source plus concordant press coverage.
How it works
Documented architectureThe stack in detail
- plateforme Omilia Voice AI Voice order-taking solution at the drive: ASR, natural language understanding, handling of order modifications, filtering of noise and accents, voice synthesis.
- infra Yum! Poseidon POS Yum! Brands's proprietary POS to which the voice order is sent to the kitchen.
- infra Menus digitaux drive-thru Screens for visual confirmation of the order, synchronized with real-time stock and limited offers.
How it runs, concretely
For ops teams-
1Customer speaks Customer
The customer places their order at the drive microphone.
-
2Understanding and handling the order AI
The voice AI transcribes, understands, handles modifications, and adapts to stock and current offers.
-
3Visual confirmation Digital menu
The order appears on the digital menu for the customer to check.
-
4Sending to the kitchen AI / POS
The order goes to the Poseidon POS.
-
5Human escalation Crew member
In case of a block or a complex order, a crew member takes over the conversation.
Voice recognition accuracy and the escalation rate to a crew member. If the AI does not handle accents, noise, or complex modifications, the experience degrades and staff must take over.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- structured menu and personalization rules
- stock state and limited offers in real time
- audio recordings to train and evaluate the ASR
Org prerequisites
- integration with the existing POS
- human escalation process
- franchise buy-in
- network of digital menus
Possible stack
- voice AI vendor (Omilia, Google Cloud)
- integrated POS
- connected digital menus
The plan, step by step
- Step 1Structure the menu and the personalization rules, map the POS integration and the real-time stock state.Deliverable: Structured menu usable by the voice AI vendor and POS integration spec.
- Step 2Integrate the voice AI with the POS and digital menus on 2-3 pilot restaurants, with systematic human escalation.Deliverable: Functional pilot: end-to-end voice ordering, crew member as safety net.
- Step 3Measure order accuracy, service time, and escalation rate against human order-taking.Deliverable: Comparative dashboard per pilot restaurant.
- Step 4Harden the system (noise, accents, out-of-norm or sabotage orders) and lock the escalation rules.Deliverable: Operations playbook per restaurant, documented incidents.
- Step 5Deploy in waves with franchise buy-in, tracking the same metrics.Deliverable: Multi-site rollout plan and weekly indicator tracking.
First step: Pilot on a few restaurants with systematic human escalation, measuring order accuracy and service time before expanding.
The same pattern, proven in other industries
Amazon
genAI shopping assistant
Booking.com
genAI trip planning assistant
Burberry
genAI copilot for consultants (search, image recognition, complete-the-look)
Sources
- S1 Yum! Brands to Expand Voice AI Technology to Hundreds of Taco Bell U.S. Drive-Thru Locations in 2024 Primary archive pending
- S2 Taco Bell's drive-thru voice AI expands to nearly 900 restaurants Established press archive pending
- S3 Taco Bell to roll out AI drive-thru ordering in hundreds of locations by end of year Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.