Voice commerce has faced the same obstacles for a decade, including turning spoken requests into accurate orders, verifying customer identity and completing payments without delays.
In 2016, PYMNTS CEO Karen Webster wrote about the potential of voice-enabled payments through Amazon’s Alexa. What Alexa could not yet prove was that the device knew who was talking or could close a sale without a screen. Ten years later, accuracy, authentication and transaction speed remain central to making voice commerce work.
Block’s latest artificial intelligence research could help with part of that. On Friday (Oct. 23), Jack Dorsey’s company unveiled Dessa, a speech-to-speech AI model designed to listen and respond simultaneously. Nine days earlier, its Square business introduced an AI-powered restaurant kiosk that lets customers place orders conversationally and complete purchases.
Block’s Dessa Model Listens While It Talks
Dessa processes sound directly instead of turning speech into text, running it through a language model and reading the answer back, Block said in its announcement of the model. It’s audio in, audio out, trained from scratch on recorded human conversation with no text and no pretrained model. Dessa is full-duplex, so it listens while it speaks and learned from real conversations when to cut in and when to signal it is following along.
In a restaurant line, that could mean a customer who says, “A large latte, actually make that a medium,” does not have to wait for the assistant to finish before correcting. Block also said inference runs faster than real time, and the model runs locally on a MacBook Pro M4 Max or better.
A model that runs on-device could power business and consumer experiences at lower cost, TechCrunch reported Saturday (Oct. 24).
What Block has not published matters as much. The Dessa announcement carried no word error rate, no latency figure, no benchmark against competing voice models and no model size. It said nothing about recognizing who is speaking.
Block is offering early access to a small group of AI safety researchers, while a public release remains pending, according to the announcement.
Sounding natural and getting an order right are different tests, and Block has shown only the first.
Square Ties Voice Orders to Menu Data
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
Square Kiosk handles the second test by reading the seller’s own catalog. Menus, prices and modifiers transfer from the seller’s Square account, and sold-out items and price changes reach every kiosk automatically, Square said on its product page. A guest can ask what is in the spanakopita or what is good for kids, and the menu filters to match. Suggestions draw on what sells at that location and time of day, according to an Oct. 14 press release.
The hardware shows where voice still breaks down. Three directional microphones and AI noise filtering separate the person ordering from the room, according to a video on the product page.
Square recommended placing kiosks at least 24 inches apart, said voice may not cope with music venues and keeps touch available at every step, according to an FAQ on the product page. Voice works only in English and only in the United States. As for results, self-ordering kiosks are designed to take more orders during busy periods, and outcomes vary by business.
Square has disclosed no accuracy, service-time or order-value data.
Voice Recognition Still Does Not Authorize Payments
Neither product uses a voice to confirm a payer. At the kiosk, guests pay with tap, dip or PIN on a built-in reader and check in to Square Loyalty or a Neighborhoods profile on the screen, according to the product page.
The card and the account do the identifying. The voice only does the ordering. That split keeps disputes and unauthorized orders inside systems merchants already trust, and it limits how much of the transaction a model like Dessa can take over.
For Square, the revenue model is already in place. The kiosk costs $1,999 with a counter stand, plus a $99 monthly software fee per device, according to the Oct. 14 press release. That charge is separate from the required Square Plus subscription at $49 per location per month or Square Premium at $149. Payment processing costs 2.5% plus 15 cents per tap, dip or swipe with Plus, compared with 2.4% plus 15 cents with Premium, according to Square’s pricing page.
Dessa, by contrast, is open only to safety researchers and has a waitlist, and Block has not said when or whether it will reach Square hardware.
For all PYMNTS AI coverage, subscribe to the daily AI newsletter.
Add as Preferred Source
