Burger King's "Patty" AI is listening for "please" and "thank you." Here's what HR should do next
Burger King is piloting an AI assistant called Patty across roughly 500 US locations, with plans to expand by the end of 2026. Built on OpenAI tech and embedded in employee headsets, Patty flags phrases like "welcome to Burger King," "please," and "thank you," rolling the results into a broader BK Assistant platform. Leaders say it's a coaching tool, not a word-by-word checklist for performance management. Still, in a climate of rising workplace monitoring, HR can't take that promise at face value-policy and guardrails come first.
What Patty actually does
According to The Verge, Patty listens to employee-customer interactions for specific "friendliness" phrases and can surface a score when managers ask. Burger King says it's also experimenting with capturing tone, not just words. The assistant ties into kitchen equipment, inventory, and a cloud POS to answer on-the-spot questions (like ingredients or cleaning steps) and auto-update menus when items are unavailable-changes propagate across digital channels within about 15 minutes. Promotional materials also show Patty nudging staff on tasks like bathroom cleaning, while AI drive-thru ordering is being tested in fewer than 100 locations.
Why employees are uneasy
Surveys show broad scepticism of digital tracking. Pew Research Center has reported worker concerns around privacy and fairness with AI-based monitoring tools. The American Psychological Association's 2023 Work in America findings indicate 51% say their employer uses tech to monitor them; monitored workers report higher rates of discomfort, micromanagement, and emotional exhaustion.
Pew Research Center | APA 2023 Work in America
Key risk areas HR should assess before rollout
- Consent and notice: Clear disclosures that conversations may be analyzed; signage for customers and acknowledgements for employees.
- Scope creep: Lock the purpose to coaching and service quality; prohibit use for discipline, pay, or scheduling unless re-approved through governance.
- Data minimization: Define what is captured (audio, transcripts, metadata), retention limits, and deletion timelines.
- Individual vs. aggregate reporting: Prefer team/shift/store-level insights; avoid one-to-one "script compliance" scores.
- Fairness and inclusion: Account for accents, multilingual interactions, neurodiversity, and cultural variations in politeness.
- Goodhart's law: If you make "please/thank you" the target, people will game it. Balance with qualitative feedback and outcomes (compliments, complaint rates).
- Transparency and feedback loops: Let employees see how the system works, how to challenge errors, and how improvements are measured.
- Security and vendor controls: Who has access to raw audio and transcripts? Is data encrypted? Is it used to train third-party models?
- Labor relations: Engage unions and, where applicable, works councils early; align with CBAs and notice requirements.
- Accessibility: Ensure headsets and voice prompts work for hearing-impaired or speech-impaired staff; provide alternatives.
Practical guardrails to put in policy
- Purpose statement: Coaching, operations support, and service quality only; explicitly ban disciplinary use without separate approval.
- Aggregation by default: Store-level dashboards, not individual leaderboards. Individual coaching only via human observation + context.
- Short retention windows: Keep raw audio as short as legally and operationally feasible; log summaries, not verbatim recordings, when possible.
- Human-in-the-loop: Any manager action requires human review; no automated nudges that feel punitive.
- Bias and accuracy testing: Pre-pilot tests across accents and languages; monitor error rates and false positives by location and demographic group.
- Clear communications: Plain-language FAQs for employees and customers; visible opt-out pathways where legally required.
- Incident handling: Define how to report misuse, data leaks, or harmful prompts; set response SLAs.
- Training: Teach managers how to use insights for recognition and coaching-not policing.
Questions to ask any "Patty-like" vendor
- What exactly is collected (audio, transcripts, timestamps, location) and where is it processed (device vs cloud)?
- How is tone detected and validated? What's the documented error rate across dialects and background noise?
- Can we disable individual-level scoring and export only aggregated metrics?
- What are the retention defaults, deletion controls, and audit logs?
- Does the model fine-tune on our data? If so, can we opt out?
- How are models updated, and how will we be notified of changes that affect behaviour or accuracy?
- What controls prevent the data from being used in performance or termination decisions?
- How do we provide employees access to their data and a dispute process?
The upside-if HR sets the rules
Used well, a headset assistant can reduce cognitive load, speed up training, and cut menu or equipment errors. Real-time stock updates prevent customer frustration, and on-demand SOP guidance keeps shifts moving. The benefit shows up in fewer escalations and faster recovery-not in forcing workers to recite magic words.
If you're building your internal playbook for AI in people processes, see our AI Learning Path for HR Managers.
Bottom line
Ambient AI in frontline work is here. Treat it like any high-stakes people system: define purpose, minimize data, keep humans in charge, and measure impact on both morale and guest experience. If the tool can't meet those standards, it's not ready for your workforce.
Your membership also unlocks:
