sayhi@zarasa.io
English
Agentforce Commerce AI agents connecting storefront, chat and ChatGPT
8 min read · July 9, 2026

Your Customer No Longer Searches Google — They Ask ChatGPT. And Now They Can Buy From You Right There

Salesforce just shipped its biggest commerce release in years: AI agents that staff your online store, handle B2B orders over chat, and put your catalog directly inside ChatGPT. Here is what US businesses need to understand right now.

What Salesforce announced (minus the jargon)

On July 6, Salesforce published Agentforce Commerce, and the core idea fits in one sentence: the purchase no longer starts on your website — it starts in a conversation.

Until now, an e-commerce chatbot was exactly that: a robot that answered FAQs and, the moment things got interesting, handed you off to a human or a form. These new agents are different. They see your real inventory, know the customer's order history, and can close the sale right there in the conversation — no handoff, no friction.

There are three agents, and each one targets a specific bottleneck:

  • Shopper Agent (your storefront). A customer asks, "Can these trail shoes arrive before the weekend?" The agent checks live stock, confirms the carrier cutoff, offers in-store pickup as a fallback, and completes the purchase in that same chat. Your best store associate, running 24/7 with zero call volume.
  • Buyer Agent (B2B self-service). A wholesale buyer texts, "I need 40 cases of what I ordered last quarter," and gets a product image to confirm the SKU, their contract price, and a completed purchase order — no portal login, no rep involvement, no back-and-forth email.
  • Merchant Agent (your operations team). Instead of navigating twenty admin screens, your merchandising team types "promote the summer line and suppress anything with fewer than 5 units in stock" — and it happens instantly. Salesforce reports early users cut time-on-task by 88%.

The numbers that should get your attention

The data in the announcement comes from Salesforce's analysis of the 2025 holiday season, drawn from 1.5 billion shoppers — a significant portion of them in the United States:

20%of global online holiday sales were AI-influenced — roughly $262 billion in revenue
59%faster revenue growth for retailers running their own AI shopping agents vs. those that weren't
higher conversion rate for traffic referred by AI assistants vs. social media

The 8× conversion figure is worth sitting with. Social media has been the dominant acquisition channel for US retailers for a decade. AI-referred traffic is already converting at eight times that rate — and the channel is just getting started.

Starting this month, through the OpenAI integration, a catalog connected to Salesforce can appear — and sell — directly inside ChatGPT. Google Search (AI Mode) and the Gemini app follow in the coming months. Salesforce made one thing clear: you remain the merchant of record. The order comes to your platform, through your fulfillment, with your loyalty program intact. This is not a new marketplace taking a cut of your customer relationship.

Why this is the right moment for US businesses

The US is where this shift hits hardest, for a few concrete reasons.

American consumers adopted AI assistants faster than any other market. ChatGPT crossed 100 million US users in 2024. A growing share of those users are now asking it product questions that they used to type into Google or Amazon. When the answer surfaces your catalog with a buy button, you are in the conversation. When it doesn't, you aren't.

Holiday is decided on AI referrals, not ad clicks. The 2025 holiday season data is the proof of concept. Twenty percent of online holiday revenue — across Black Friday, Cyber Monday, and the full December window — was already shaped by AI. That number will be higher in 2026. US retailers who are not indexed in AI shopping results will feel it in Q4.

B2B commerce in the US runs on portals that buyers hate. Most enterprise procurement portals in the US are clunky, require logins, and force buyers to re-enter information they have already given you. The Buyer Agent replaces that friction with a chat interface over the channel the buyer already has open — whether that is Slack, Teams, SMS, or a website widget. Buyers get answers in seconds. You get cleaner orders with fewer errors.

The frame that matters: this is not a chatbot upgrade. It is a new sales channel. The retailers and distributors who treat it that way — with dedicated inventory, pricing logic, and escalation paths — will see results. The ones who bolt it onto an existing FAQ bot will not.

Is this only for enterprise retailers?

No. Storefront Next is included in every B2C Commerce plan at no additional cost and gets a production-ready store live in under 30 minutes. Mid-market US retailers with an existing Salesforce org can move fast.

The honest caveat: an agent is only as good as the data underneath it. If your product catalog has gaps, your contract pricing lives in a spreadsheet outside Salesforce, or your inventory feed has a 24-hour lag, the agent will surface those problems — quickly and publicly. Every implementation we run starts with a data audit for exactly this reason. Clean data is not optional; it is the prerequisite.

How to get started without overcomplicating it

The same phased approach we use for Agentforce projects applies here:

A 4-step launch framework

  • Step 1 — Pick one use case (weeks 1-2). For B2C retailers, the Shopper Agent on your highest-traffic category is the fastest path to a measurable result. For B2B distributors, the Buyer Agent on repeat orders has the clearest ROI story. Do not try to cover everything at once.
  • Step 2 — Audit your data (weeks 2-4). Product catalog completeness, pricing accuracy in Salesforce, and inventory feed latency are the three things that make or break the agent's first impressions with customers.
  • Step 3 — Run a sandbox test against real order history (weeks 4-8). Pull your last 500 customer inquiries or B2B orders and replay them through the agent. Measure containment rate and error rate before going live.
  • Step 4 — Launch with a human review layer (weeks 8-12). The agent handles and drafts; a rep reviews before confirmation. Expand automation incrementally as confidence builds. Trying to go fully autonomous on day one is the most common reason pilots stall.

The question your leadership team will ask

When your CMO asks ChatGPT to find your product category, will your catalog appear? When a buyer asks their AI assistant for a reorder, will your storefront be the one that responds? The 2025 holiday data shows this is already happening at scale. The window to get indexed before your competitors do is closing. That is the decision on the table.

How Zarasa can help

Zarasa is a Salesforce Consulting Partner headquartered in Miami. We work with US-based businesses on Agentforce deployments from initial assessment through production launch — including the data cleanup, integration work, and change management that most vendors skip over. The first conversation costs nothing.

Ready to put Agentforce Commerce to work?

We'll assess your current Salesforce setup and tell you exactly what a first deployment would take — timeline, cost, and expected return.

Talk to our team

Frequently asked questions

Do I need Salesforce to use these agents?

Yes — the agents run on Salesforce's Commerce platform. If you already run Sales Cloud or Service Cloud, part of the foundation is in place since customer data is shared. The exception is Agentic Commerce Search, which can run on non-Salesforce storefronts.

When is selling inside ChatGPT available?

The OpenAI integration is generally available as of July 2026. Google Search (AI Mode) and the Gemini app are announced for the coming months. Full details are in the official Salesforce announcement.

Does this replace my sales team?

It automates the repetitive parts — answering product questions, confirming stock, processing repeat orders. Consultative selling, relationship management, and complex negotiations stay human. Most teams use it to free up reps for higher-value work, not to reduce headcount.

What happens when the agent can't handle an order?

It escalates to a human rep with the full conversation context attached — similar to how a Service Cloud case works. Defining those escalation thresholds clearly is one of the most important parts of the implementation.