Salesforce · Filed Feb 25, 2026 · Published Sep 10, 2026 · verified — real USPTO data

Salesforce Patents a Voice AI That Talks Store Workers Through Checkout Tasks

Salesforce has filed a patent for an AI voice assistant designed to sit on a store employee's phone or tablet and talk them through retail tasks, from completing a sale to closing out the register at the end of the night.

A person speaks to an iPhone, which converts the voice to text, classifies it, and performs an action based on the input. Drawing from patent filing US 2026/0268907 A1.
A person speaks to an iPhone, which converts the voice to text, classifies it, and performs an action based on the input.
See all 4 drawings from this filing ↓
Publication number US 2026/0268907 A1
Applicant Salesforce, Inc.
Filing date Feb 25, 2026
Publication date Sep 10, 2026
Inventors Nitin MANGTANI, Sandilya GARIMELLA, Viswanth CHADALAWADA, Madhav SAHU
CPC classification 704/275
Grant likelihood Medium
Examiner CENTRAL, DOCKET (Art Unit OPAP)
Status Docketed New Case - Ready for Examination (Jun 8, 2026)
Parent application is a Continuation of PCTUS2024043437 (filed 2024-08-22)
Document 16 claims

What Salesforce's retail voice AI actually does for store workers

Imagine you're working a closing shift at a retail store and you can't remember the exact steps to reconcile the cash register before shutting everything down. Instead of hunting through a binder or calling a manager, you just ask your phone.

That's the idea behind this Salesforce patent. It describes an AI assistant built specifically for store workers, one that understands spoken questions and commands in the context of a retail environment. It knows, for example, that you have to balance the till before you can close the store, not after.

The AI runs as an app on a regular smartphone or tablet, which means stores wouldn't need special hardware. It's trained on support tickets and product manuals so it can answer the kind of questions that come up during a real shift, and it can have a back-and-forth conversation rather than just responding to single commands.

From the filing · CLAIM 1
… storing a small language model (SLM) which, when executed by the processor, causes the processor to convert the spoken input to text, classify the text with the SLM, and identify and perform an action in response to classification of the text.

Translation: A compact AI stored on the device translates speech into text and figures out what physical action to take next.

How the on-device language model handles spoken store commands

The patent describes a point-of-sale (POS) conversational AI that combines four capabilities: it listens to spoken input through a microphone, converts that speech to text, understands the meaning of the request in a retail context, and then responds in a natural, two-way conversation.

At the center of the system is a small language model (SLM), which is a compact version of the kind of AI that powers chatbots like ChatGPT. Because it's smaller, it can run directly on the device, such as a phone or tablet, without needing a constant cloud connection. The model is trained specifically on retail support data, including help-desk tickets and product documentation, so it understands the vocabulary and workflows of a store environment.

A key feature is dependency awareness. The system understands that certain tasks must happen in a specific order. If a worker asks to close the store before balancing the register, the AI knows to flag that sequence as incorrect.

The claim covers:

  • A microphone capturing spoken commands
  • A processor running the language model
  • The model converting speech to text, classifying the request, and triggering the appropriate action
From the filing · THE ABSTRACT
The AI understands dependencies (e.g., the tills should be reconciled before closing the store) and predict appropriate responses to conversational inputs.

Translation: The system knows the correct order of operations, such as counting cash registers before locking up.

What this means for retail workers and point-of-sale software

Retail has one of the highest employee turnover rates of any industry, which means stores are constantly training new workers on complicated, multi-step processes. A voice assistant that can answer procedural questions in plain speech, without requiring workers to pause and search a manual or wait for a manager, could cut down on errors during high-pressure moments like end-of-day closing.

Salesforce's growing interest in industry-specific AI agents points toward a strategy of embedding AI into operational software rather than selling it as a separate tool. For retailers already using Salesforce's commerce or CRM products, a voice assistant that ties into the same system could be a meaningful addition to a platform they're already paying for.

That makes this Salesforce's 20th filing we've tracked since May on AI agents that act for you, following one on asking questions before building and one on agents built from old transcripts.

Editorial take

Retail workers get stuck on procedural tasks constantly. Closing a register the wrong way, skipping a step in the end-of-day routine, or mishandling a return can trigger hours of cleanup work, and those mistakes repeat across every shift, every store, every week.

A voice assistant on a phone that actually knows store procedures, and understands that some steps have to happen before others, targets something with real financial weight. Training it on real support tickets rather than generic manuals is what gives it a chance of answering the questions employees actually ask, not the tidy ones that show up in a training binder.

The honest uncertainty is whether it holds up in a loud, busy store where someone is talking fast under pressure. That is where the value either proves itself or falls apart.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings

4 drawing sheets from US 2026/0268907 A1 · click any drawing to enlarge

Patent filing page

Source. Full patent text and figures from the official USPTO publication PDF.