Give commerce its voice back

I have spent my career inside ecommerce. I built and sold an ecommerce marketing agency that worked with hundreds of online brands. I then co-founded Ecomma, where we acquired and operated more than 70 ecommerce brands ourselves.

Operating that many stores teaches one lesson on repeat: the brands that win have the tightest feedback loop between what customers say and what the business does. I watched that loop break everywhere. Customers wanted to talk. No brand could afford to listen, because a human phone line never penciled out at scale. So every brand, including ours, fell back on emails, flows, tickets, and chat widgets. Reactions instead of conversations.

In June 2025 I started Loop Voice to fix this. Nine to ten months of building and hard iteration got us to a product that works. Version 2 launched in August 2026 with strong merchant feedback. Today over 10,000 conversations a month run through Loop. We ship a new version every six to nine months.

Commerce began as a conversation. Two people at a market stall, talking. The seller learned what the buyer wanted. The buyer learned what the seller had. Price, fit, doubt, and trust all got settled in the same exchange. The feedback ran both ways, in real time, and both sides walked away with an answer.

That was not a feature of commerce. It was the interface. For most of human history, to buy something was to talk to someone.

The internet scaled commerce and dropped the conversation. In its place we got channels that provoke a reaction instead of starting an exchange. Email. Retargeting. SMS. The chat widget in the corner.

None of these ask a question and wait for the answer. They push a message and hope. The customer cannot ask the one thing holding up the purchase. The business never hears why the cart was left. Neither side gets an answer.

There is research behind why this feels thin. Media richness theory (Daft and Lengel, 1986) ranks communication channels by how much they carry: immediate feedback, tone, the ability to clear up confusion on the spot. Media synchronicity theory (Dennis, Fuller, and Valacich, 2008) adds that some work needs both sides present at once. Buying is that kind of work. A voice conversation is the richest channel short of standing in the same room. Text is the leanest.

Customers know this. In a ServiceNow survey of more than 27,000 consumers, 87 percent ranked the phone as their preferred channel for customer service. A live conversation settles in five to fifteen minutes what an email thread drags across days.

We did not remove the conversation because it stopped working. We removed it because we could not staff it. Labor, like in most exchanges, doesn't scale very well.

Three curves crossed at once.

Humanness. AI voices now sound close to human. On Vapi's Humanness Index, which scores models in blind tests against a real human baseline of 100, xAI's Grok TTS ranks first at 96, with ElevenLabs Eleven v3 and MiniMax Speech 2.5 close behind near 92. The gap a listener can hear is nearly closed.

Cost. A production AI voice agent runs roughly $0.07 to $0.25 per minute all in. A human contact center call runs about $6.47. Labor can represent up to 95 percent of contact center costs, per Gartner. The economics inverted.

Latency. Human turn-taking is fast: the modal gap between speakers falls between 0 and 200 milliseconds, a pattern that holds across all ten languages tested (Stivers et al., PNAS, 2009). The best voice systems now respond in under 400 milliseconds per turn. Fast enough that the pause no longer gives the machine away.

There is a fourth signal. Character.AI users spend an average of 80 minutes a day talking to its chatbots, per the company's own CEO. The behavior is already here.

There is a caveat about speaking with an AI. Most consumers dislike AI phone calls today, because big companies deployed them badly: stiff scripts, dead ends, no way to reach a person. That is a real objection, not a marketing problem. I believe trust is earned one good conversation at a time. Every call we place has to be worth answering. That is the standard.

Every voice interaction is bound by three things: accuracy, latency, and humanness. Today they trade off. Push for the most human voice and you often pay in speed. Push for speed and accuracy slips.

Model progress is collapsing that trade-off. Each new generation buys more of all three at once. Loop Voice is built to capture each improvement the moment it ships, so the triangle keeps shrinking underneath our merchants without any work on their side.

Every communication channel got regulated once it mattered. Email got CAN-SPAM, effective January 1, 2004, nearly a decade after commercial email exploded. Calls and texts got the TCPA in 1991, then the FCC's prior express written consent rules, effective October 16, 2013. On February 8, 2024, the FCC ruled unanimously that AI-generated voices count as artificial or prerecorded under the TCPA. States moved too: California's bot disclosure law in 2019, and the Utah AI Policy Act in 2024.

Voice AI will be regulated. We treat that as an advantage, not a threat. Loop Voice is built for disclosed AI, consented calls, and consumer control from the first minute. When the rules tighten, compliant operators gain ground and the rest scramble. We intend to be the ones already standing.

We build the whole loop, not just the agent. We build the simulation environment the agent is tested in, and the critic that judges a good conversation from a bad one. As Waymo co-CEO Dmitri Dolgov said at Y Combinator's Startup School 2026, to make a physical AI agent work at scale you have to build the entire ecosystem: the agent, the simulator, and the critic that rigorously evaluates and judges performance.

We do not train foundation models. Frontier labs keep shipping better models every few months. An applied company wins by riding those curves, not racing them, with superior context, orchestration, and integration depth.

LoopIQ is the compounding engine. Every call produces outcome data: which phrasing converts, and whether the consumer felt heard. The system learns from every conversation. More calls make better conversations. Better conversations make more calls.

We integrate downward over time, from the application layer toward voice hosting and telecom. The goal is to become the best-priced voice platform with the highest outcome value, because we own more of what we depend on.

And the agent always speaks as the merchant. The merchant's brand, the merchant's phone number, the merchant's voice. Loop Voice is invisible. The brand is not.

I've learned throughout my career that time kills value, so operating with a sense of urgency is key to provide the best product and the best price possible.

We start with retail, because retail sets the customer experience standard that other industries copy. Global ecommerce is roughly $6.4 trillion a year. Contact center and outsourcing is over $100 billion a year. Gartner projects that by 2026, conversational AI deployments in contact centers will reduce agent labor costs by $80 billion.

The sequence is deliberate. Shopify SMB first. Then mid-market and enterprise commerce. Then every business-to-consumer interaction that should have been a conversation all along.

Commerce began as a conversation. I intend to give it back.