The Most Successful Product Amazon Ever Killed
On May 13, Amazon retired Rufus.
Not because it failed. Rufus reached more than 300 million customers in two years and drove roughly 12 billion dollars in annualized incremental sales. Shoppers who used it were 60% more likely to complete a purchase, per Flipflow's analysis of the launch data.
Amazon killed it anyway. In its place stands Alexa for Shopping, a single agent built on Alexa+ that runs across the Amazon app, Amazon.com, and Echo devices. It is free for every Amazon account. No Prime subscription, no Echo, no separate app required.
The difference is not a better chat window. The difference is authority. Rufus answered questions and pointed you toward checkout. Alexa for Shopping holds your payment details, watches prices on your behalf, and completes transactions without a final tap.
Chatbots recommend. Agents buy. Amazon just handed transaction authority to the busiest product search box in US retail.
Ten Years of Voice Commerce That Never Converted
The Echo launched in 2014. The voice shopping predictions started about a week later, and they have missed every year since.
The market is real but thin. Ringly puts US voice commerce at 22.4 billion dollars for 2026, a sliver of online retail. Adoption is the blunter number. Capital One Shopping research finds that only 23.3% of Echo owners have ever shopped by voice. Google Home owners do it more, at 36.8%, despite a smaller installed base.
The standard excuse blamed the microphone. Nobody wants to recite a card number to a speaker, the argument went, and nobody trusts a gadget to pick products sight unseen.
The better explanation: there was never an agent on the other end. Legacy Alexa matched a phrase to a reorder or a top search result. It had no memory of your preferences, no view of price history, no judgment. Voice shopping failed because talking to a cash register is worse than tapping one.
Early Alexa+ behavior suggests the brain was the bottleneck, not the voice. Daniel Rausch, Amazon's VP of Alexa and Echo, told CNBC in February that Alexa+ users hold two to three times more conversations than legacy users, and shop through it three times more often.
Fix the agent and the channel starts working.
What an Agent Does That a Chatbot Could Not
Alexa for Shopping moves into the Amazon search bar itself. Ask a question where you used to type keywords. The assistant answers in place, with comparisons, buying guides, and a running memory of what you told it last week. Rajiv Mehta, Amazon's VP of conversational shopping, describes it as "an expert personal shopper who already knows you and remembers your preferences."
The concrete capabilities matter more than the framing:
- Price memory. Up to a full year of price history on hundreds of millions of products, visible on demand.
- Price alerts. Ask for a notification when the face cream drops to 75 dollars.
- Conditional purchases. Tell it to buy the headphones when they hit 30% off. It watches, then completes the order with your default payment method.
- Scheduled Actions. Standing instructions, like adding sunscreen to the cart if it falls under 10 dollars and you have not bought any in two months. That example is Amazon's own.
- Cross-device memory. A conversation that starts on an Echo Show resumes in the app or on the web, mid-thought.
| Capability | Rufus (2024 to 2026) | Alexa for Shopping (2026) |
|---|---|---|
| Where it lives | Chat panel inside the app | Search bar, app, web, Echo devices |
| Memory | Largely per session | Persistent, synced across devices |
| Price tools | Could answer deal questions | One-year history, alerts, target-price auto-buy |
| Purchasing | Handed you to checkout | Completes checkout itself |
| Off-Amazon reach | None | Buy for Me on third-party sites |
| Access | Amazon app shoppers | Every Amazon account, free, no Prime |
Buy for Me Makes Amazon the Customer of Every Store
The feature that should keep retail executives up at night is the quiet one. When a product is unavailable on Amazon, the agent locates it on a third-party retailer's site and completes the purchase there. It pays with the customer's encrypted Amazon payment details and ships to the stored address. The shopper never leaves Amazon's interface.
Several external retailers told reporters covering the launch that they never authorized Amazon to transact on their sites.
Hold that against the court calendar. In March, a federal judge granted Amazon a preliminary injunction against Perplexity, blocking its Comet browser from logging in and shopping on users' behalf. Amazon's winning argument was that a user's permission is not the platform's authorization. Nine weeks later, Amazon shipped Buy for Me, an agent that shops other retailers' sites with exactly that: user permission. The injunction is paused while the Ninth Circuit hears the appeal.
Andy Jassy has argued that outside agents lack personalization, purchase history, and reliable stock and delivery data. Buy for Me is the inverse claim. Amazon's agent has all three, so it should shop everywhere.
Whatever the courts decide, retailers face the choice now:
- Block the bot. Fingerprint and refuse agent traffic. You keep control and lose every sale the agent would have routed to you.
- Feed the bot. Publish clean structured data: live availability, pricing, variants, shipping terms. The agent transacts reliably, and you become a fulfillment endpoint inside Amazon's customer relationship.
- Ignore it. The agent buys through brittle scraping anyway. Checkout errors land on your support team while Amazon owns the customer.
None of these are comfortable. One is worse than the others, and it is the default.
The Answer Is the New Ad Slot
Inside Amazon, the shift is quieter but bigger for sellers. When the assistant answers inside the search bar, that answer competes with sponsored placements for the same glance. An AI overview that names two products has functionally sold the top of the page without charging anyone for it. Yet.
Sellers spent a decade optimizing for a ranking algorithm: titles, backend keywords, bids. The agent does not read ads the way shoppers do. It weighs structured attributes, review substance, return rates, availability, and that newly visible year of price history.
Call the new discipline agent answer optimization. The question is no longer whether you rank. It is whether you are the answer, and whether you make the comparison table the agent builds.
Three practical consequences:
- Attribute completeness becomes revenue. Missing size, material, or compatibility fields silently drop you from comparisons.
- Price games stop working. A visible year of history exposes the inflated list price. The agent knows the shirt was 15 dollars in March.
- Reviews get read, not counted. The agent summarizes themes. A 4.6-star product with recurring durability complaints loses to a 4.4 with consistent praise.
Most brands have no view of what agents say about their catalog, and you cannot optimize what you cannot see. That visibility gap is a data problem, the kind our commerce intelligence APIs exist to close. Tracking how your products surface in agent answers is the new rank tracking.
Spoken Intent Is a New Data Layer
A keyword reveals what someone typed. A conversation reveals why.
"Running shoes" is a query. "I need wide-fit running shoes for a first marathon in October, under 100 dollars" is a demand forecast with a deadline attached. Multiply that by hundreds of millions of sessions and the conversational layer becomes the richest intent dataset commerce has produced.
Target prices may be the most valuable exhaust of all. When thousands of shoppers set auto-buy thresholds at 75 dollars for the same moisturizer, that is a live price elasticity curve no survey could draw. Scheduled Actions expose household replenishment cycles, product by product.
Amazon captures all of it. Brands see none of it by default.
That asymmetry explains why retailers are building conversational layers of their own. Kate Spade put an AI gift concierge in front of shoppers in April, and more will follow. The only way to own conversational intent data is to host the conversation. That means building, evaluating, and monitoring an agent yourself, the work we partner on through Build With Us.
What to Watch Before the Holidays
Alexa for Shopping launched about six months before Black Friday. That is the test that counts.
Watch three numbers between now and then. Whether the 23.3% of Echo owners who have ever voice-shopped finally moves. Whether sellers report AI answers cannibalizing sponsored clicks in the search bar. And whether the Ninth Circuit lets the Perplexity injunction stand, because that ruling decides if blocking a shopping agent is even enforceable.
The deeper shift does not need a court's permission. Voice commerce spent ten years as a demo because nothing on the other end could remember, compare, or commit. Now something can. It carries 300 million shopping histories, a year of price memory, and stored payment credentials.
Every store on the internet is about to meet its newest customer. It does not browse, it does not see ads, and it never forgets a price.