AI shopping agents now fail far more often than they buy. Quantum Metric's 2026 Peak Benchmark, released November 5, found that retail AI agents abandon 48 carts for every purchase they complete, against about 3 for all retail traffic. Agent visits to brand sites grew 25x in under a year. More of your next customers will send a bot to check you out before they ever call, and most of those bots are hitting walls.
What did Quantum Metric find about AI shopping agents?
Quantum Metric found that AI agents are the fastest growing visitor on brand websites and also the worst at finishing a purchase. Agent sessions went from 46,000 in October 2025 to 1.14 million in September 2026, and retail agents abandoned 48 carts per completed order.
The data comes from a survey of 1,500 consumers in the US and UK, combined with anonymized behavior data from Quantum Metric's platform, according to the company announcement. The report splits AI traffic three ways: people who click a link from an AI tool, people who browse with an AI assistant and then take over, and agents browsing a site alone for someone.
| Finding | Number |
|---|---|
| Consumers who used AI to shop, book or pick a service in the last six months | Nearly 2 in 3 |
| Growth in agent sessions, Oct 2025 to Sept 2026 | 25x (46,000 to 1.14 million) |
| Retail carts abandoned per completed purchase, agents | 48 |
| Retail carts abandoned per completed purchase, all traffic | About 3 |
| Travel carts abandoned per completed purchase, agents | 26 |
| Consumers who would finish a failed agent purchase on the same site | 45% |
| Forbidden (403) errors in retail, often a sign of bot blocking | Up 2.7x |
Why do AI agents abandon so many carts?
A lot of agents get blocked or confused before they reach checkout. Quantum Metric reported that forbidden (403) errors, which often mean a site is blocking bots, rose 2.7x in retail, and retail agents spend 75% less engaged time per session than people (about 30 seconds instead of two minutes).
An agent reads the page, looks for the price, the options and the button, and moves on if it can't find them fast. A human will scroll, squint and maybe call. CEO Mario Ciabarra put it this way in the release: "Traditional analytics often can't tell those visitors apart, so a blocked agent looks like any other drop-off." You see a bounce in your analytics. You never learn it was a buyer's assistant that gave up.
What happens to the customer after an AI agent fails?
Most of them go somewhere else. Only 45% of consumers said they would finish the purchase themselves on the same site if their agent couldn't complete it, according to the report. The other 55% are a sale you paid to attract and then handed to a competitor.
The people who browse alongside an AI assistant are a different story. Those mixed sessions last about 19 minutes compared with 7 for typical human traffic, and in retail they convert about 30x higher than agent-only sessions. A person with an assistant at their elbow is a hot lead. Treat them like one.
Does this matter if you run a local service business?
Yes, because the survey covers booking and choosing services, not only online shopping. Nearly 2 in 3 consumers used AI to help them shop, book or select a service in the past six months, which means someone's assistant is already comparing you with the plumber, dentist or med spa down the street.
I've been running ads since 2013, and the pattern is old even if the visitor is new. You pay to get someone to your page, then you lose them to a slow form, a missing price range or a phone nobody answers. An AI agent just finds those gaps faster than a person does, and it doesn't leave a voicemail.
Here is what I would check this week:
- Open your booking page on your phone and time how long it takes to book. If it takes more than a minute, an agent will likely give up first.
- Put your service area, hours, starting prices and what happens on the first visit in plain text on the page. Agents read text, not text baked into images.
- Ask whoever runs your site whether a firewall or bot blocker is refusing AI assistants. A 2.7x jump in 403 errors says a lot of sites are turning buyers away by accident.
- Make sure every lead that does come through gets an answer in minutes, by text or call, at any hour.
That last one is where most local businesses bleed. If you want a number for it, our missed call calculator shows what unanswered calls cost you in a month based on your own call volume and job value.
KenjiAI was built for this kind of shift. Our Voice AI agents answer calls and book appointments around the clock, and Conversation AI handles texts and chat from your site, so the person or the assistant on the other side gets a real answer and a time on the calendar instead of a dead end. Every booking lands in the CRM with reminders already set.
What should you watch over the next few weeks?
Watch your own conversion rate from Thanksgiving through Cyber Monday and compare it with your traffic. Quantum Metric says AI-referred retail traffic in early October was already 3.1x higher than the same week last year, and 1 in 5 consumers have let, or would let, AI complete a holiday gift purchase.
If your visits climb and your bookings don't, the gap is probably friction you can't see in a normal report. Fix the booking flow first, then make sure every lead gets a fast reply. Our guide on why leads go cold covers the response-time side, and the full Quantum Metric benchmark is free to read.