SharkNinja Rolls Out Agentforce and Shopper Agent Globally: AI Agents Handle 20,000 Chats a Week Before and After Purchase
SharkNinja has deployed Salesforce Agentforce and Shopper Agent worldwide, with AI agents handling around 20,000 chats a week from product selection to setup support. How it works, how to read the numbers, and what it means for e-commerce operators.
Key Takeaways
- Home appliance maker SharkNinja rebuilt its commerce stack on Salesforce Commerce Cloud, Agentforce, Shopper Agent, Data 360, and Service Cloud in about nine months, then rolled it out from North America to every country it operates in. AI agents now handle around 20,000 customer chats a week, from choosing a product before purchase to setup guidance that starts from a QR code on the box and finding replacement parts
- The key point is that pre-purchase and post-purchase service run on the same agent and the same customer data foundation. Customers who bought at big-box retailers, Amazon, or TikTok Shop can be recognized as SharkNinja customers through warranty registration and setup support, and guided toward their next purchase
- Most of the published results come from Salesforce customer stories, and the definition and baseline of the 11% chat conversion rate, as well as implementation costs, are undisclosed. For e-commerce operators, the first step is to unify product and order data and design the path to a human before deploying an agent
SharkNinja rebuilt its stack in nine months and put AI agents at the front of customer service
CIO Velia Carboni is helming an agentic race forward with Salesforce.
diginomica.comSharkNinja, the company behind Shark vacuums and Ninja kitchen appliances, has replaced its entire commerce foundation and put AI agents at the front of its customer touchpoints. The details come from an interview with CIO Velia Carboni published by enterprise IT outlet diginomica on September 25, 2026.
"In nine months we did a full re-write of everything," Carboni says. The company went live in the U.S. and Canada first and has since rolled out to every country where it operates. As part of the same program it also went live with Service Cloud, Data Cloud (now Data 360), Agentforce, and Shopper Agent, the conversational shopping agent for online storefronts.
SharkNinja sells products across 41 consumer categories, launches around 25 new products a year, and adds 13 or 14 countries annually. According to diginomica, Agentforce agents now handle around 20,000 customer chats each week.
When Carboni joined two years ago, the legacy platform had limits on customization, social integration, and global scalability, and order management grew more complex as distribution centers were added. According to a Salesforce blog post, the implementation kicked off in January and went live in September, in Canada first and then the U.S. Those roughly nine months correspond to the "re-write," and the rollout to the rest of Europe was completed in the second quarter of 2026 (see below).
The pre-purchase agent answers "which one should I buy?"
The starting problem was that the breadth of the catalog itself confused shoppers. "We innovate 25 products a year but don't retire many, so consumers can get overwhelmed by choice," Carboni says, describing the shopping agent as a "co-shopper."
Salesforce's customer story describes how the agent works. When a shopper asks "Which blender is best for smoothies?", the agent first asks clarifying questions. It then searches the live product catalog through a "shopper search action" and ranks the top three most relevant products. It also checks purchase history so it does not recommend items the customer already owns, suggesting complementary products and accessories instead.
The same agent handles post-purchase service. It autonomously tracks orders, registers warranties, kicks off returns, and pinpoints compatible replacement parts, giving product-specific answers to questions such as which replacement filter fits a given Shark vacuum. The Salesforce blog emphasizes that a shopper can move from a delivery status question to discovering a new product within one conversation. A customer contacting support about an old vacuum is also a prime candidate for an upgrade.
This shopping capability was added to an existing FAQ agent. SharkNinja took eight weeks to add guided shopping to its FAQ agent and went live in early November 2025. Founder Mark Rosenzweig explained that the business generally has a cutoff of no new capabilities after October 1, and said the launch was possible "because the foundation and data were already there." According to the customer story, the agent handled 280,000 chats in its first four months, with an 11% conversion rate.
Salesforce is turning this kind of agent into a product. In the Agentforce Commerce release that became generally available in June 2026, Shopper Agent is positioned as an agent that carries the customer conversation on the brand's own storefront from discovery through checkout and into service. Salesforce argues that native connections to inventory, order management, and customer data are what separate it from a bolt-on chatbot.
The post-purchase agent starts with a QR code on the box
In the diginomica interview, Carboni singled out the "unboxing agent" as something the company has really been pushing.
The first product was the Ninja Luxe Café Premier espresso machine, which has more than 20 parts and accessories. Its instructions were horizontal PDFs that were hard to read on a phone. According to Salesforce's customer story, this one product drove thousands of setup-related inquiries in 2025, with average handle times reaching 18 minutes per call.
Under the new approach, launched in April 2026, a customer scans a QR code on the box to open a model-specific experience. They choose a starting point such as setting up the machine, dialing in espresso, or exploring recipes, then follow a preset, step-by-step flow. To make sure customers do not miss critical steps, this part is deterministic rather than left to free-form AI generation. At any point, though, customers can go off-script and ask questions in their own words, such as how to clean the milk frother, and the agent answers from articles, images, and videos tied to that exact model.
What stands out in the design is that it decides in advance when the agent should not answer. If a customer asks to speak to a person or raises a safety concern such as "my machine is sparking," the agent does not try to resolve it and directs the customer to human support. To keep answers grounded in model-specific content, Salesforce engineers built a custom retriever.
The agent supports seven languages, including Japanese. SharkNinja anticipates that it will autonomously resolve 30% of inquiries and assist around 1,000 customers a day. Both figures are expectations, not results. Plans include adding purchase completion, account updates, and real-time handoffs to service reps, and extending the experience to other products.
Turning customers who bought through retailers into SharkNinja customers
Read as a case of agentic commerce, where AI agents handle commerce from product discovery through purchase and post-purchase service, the core of SharkNinja's approach is its channel structure. Carboni told diginomica that much of the business is not direct to consumer; customers buy through partners like big retailers, Amazon, or TikTok Shop.
We've really been pushing the unboxing agent, because we want people to enroll with us.Source: Velia Carboni, CIO, SharkNinja
SharkNinja became the largest brand on TikTok Shop in the U.S. in less than a year on the platform. But when customers buy through a retailer or marketplace, their names and contact details usually do not reach the manufacturer. Setup support that starts from a QR code on the box works as the first direct touchpoint between the manufacturer and the buyer. According to the Salesforce customer story, the agent can even register warranties for products purchased through third-party retailers.
Data 360, the customer data platform, ties these touchpoints together into one customer. It recognizes guest checkouts and purchases made with a different email address as the same person, linking browsing, purchases, and support history into a single profile. It also standardizes product names that differ by country, so the same product is recognized as one. This unification is a prerequisite for an agent that recommends only products the customer does not already own.
The approach is also extending beyond SharkNinja's own site. When Salesforce and Google Cloud announced an expanded partnership at Dreamforce 2026 on September 15, 2026, Carboni said that surfacing products across Google Search and the Gemini app, with checkout on the results page enabled by UCP (Universal Commerce Protocol, a common specification for purchases made through AI agents), shortens the path from discovery to purchase. Payments and order management stay on Commerce Cloud. When this will start for SharkNinja has not been disclosed. In the same announcement, Evan Gerber, SharkNinja's VP of AI and Emerging Technology, described combining Gemini Live with Agentforce to grow guided setup into a "product companion."
How to read the published numbers
| Metric | Value | Source | Caveat |
|---|---|---|---|
| Customer chats handled by AI agents | About 20,000 per week | diginomica (CIO interview) | Pre- and post-purchase breakdown undisclosed |
| Chats in the first four months of the shopping capability | 280,000 | Salesforce customer story | Cumulative since the early November 2025 launch |
| Chat conversion rate | 11% | Salesforce customer story | Definition and baseline undisclosed |
| Site conversion rate (YoY) | Up 6% | Salesforce customer story | Result of the overall Commerce Cloud rollout, not the agent alone |
| Items added to cart (YoY) | Up 14% | Salesforce customer story | Same as above |
| Customer churn (YoY) | Down 58% | Salesforce customer story | Same as above |
| Unboxing agent autonomous resolution rate | 30% | Salesforce customer story | An expectation, not a result. Assisting about 1,000 customers a day is also an expectation |
| Agentforce implementation cost | Undisclosed | None | Cost per chat and human handoff rate are also undisclosed |
Most of these numbers come from Salesforce customer stories, meaning a vendor presenting the impact of its own products. The 11% conversion rate is eye-catching, but there is no explanation of what counts as a conversion or how it compares with shoppers who did not use the agent. Shoppers who start a chat tend to have higher purchase intent to begin with, so the 11% cannot be read directly as the agent's effect. The year-over-year improvements are also results of the whole platform change.
On the Q2 2026 earnings call in August 2026, CEO Mark Barrocas said the Salesforce D2C launches in the rest of Europe had only just been completed in the second quarter, and that benefits would start to show in Q4 and accelerate in 2027. He also said the company does not break out the share of its D2C business. Agentforce implementation costs, cost per chat, and the share of chats passed to human agents are also undisclosed.
Human handoff is another open question. The shopping and service agent directs customers to a "Contact us" page if it cannot resolve a question after three attempts, and real-time handoffs to service reps with full conversation context are described as a future capability. In a Gartner survey of 3,566 customers conducted in February and March 2026, 87% said companies using GenAI for customer service must provide an option to reach a human agent. A Gartner analyst noted that customers forced through multiple unsuccessful AI interactions are less likely to use that tool again. How many of the 20,000 weekly chats end up back with humans will be a key factor in judging this case.
If it cannot show value in two weeks, change direction
Carboni says a three-to-five-year project would never have worked in the company's culture. AI work, including Shopper Agent, now runs in two-week sprints, and if an effort cannot show value in that time, the team revisits the problem itself.
To move past being "the typical company" with 200 AI pilots running, SharkNinja shut down most of its operations for about two weeks in the spring for a hackathon called "Jailbreak Live." Employees from functions including finance and HR, joined by 25 partners, took on 20 big problems, since narrowed to five.
What e-commerce operators can take from this
The first lesson is the order of work: data unification came before the agent. SharkNinja could add shopping capabilities in eight weeks because standardized product names, model-specific content, and purchase history unified across countries and channels were already in place. Any company considering AI-driven customer service should first check whether product and order data can be retrieved from one place.
The second is to design post-purchase touchpoints as an entry point for customer acquisition. If a QR code in the box leads to genuinely useful setup support and warranty registration, enrolling becomes a meaningful action for the customer as well.
The third is to decide up front what the agent should not handle and how to measure it. Hand safety issues and strong complaints to humans, and define in advance how chat users will be compared with non-users and what counts as a conversion. That comparison is exactly what SharkNinja's public information lacks.
Conclusion
SharkNinja is connecting pre-purchase product selection with post-purchase setup and parts lookup through the same AI agent and the same customer data, aiming to bring buyers who purchase through retailers into its own customer base. The scale of 20,000 chats a week is now public, but most of the results are still vendor case studies and expectations. The next things to watch are how smooth handoffs to humans become, whether UCP checkout on Google results actually goes live, and whether the benefits the CEO promised from Q4 onward show up in the numbers.
