Mavent logo — white wordmark with orange accent
Abstract network of glowing nodes and connecting lines on a dark background

M4AI Commerce · Commerce Cloud

Agents that
buyers
actually use.

Mavent packages Agentforce agents for the two ends of commerce — a shopper-facing concierge and an operator-facing merchandiser — built on a catalogue and order platform that agents can read and act on.

Packaged Agentforce agents for Salesforce Commerce Cloud

Two sides of the counter

Selling and being
sold to.

Commerce has two audiences with very different tolerances for error. The agent on your storefront is judged in seconds by a customer. The agent in your merchandising team is judged against a spreadsheet. Treating them as one deployment is where agentic commerce programmes go wrong.

The agent your customers meet

Discovery, recommendation, configuration and order status. This is judged in seconds, so it has to be grounded in a clean catalogue and behave inside your pricing and entitlement rules — not improvise around them.

  • Natural-language product discovery
  • Recommendations from real purchase history
  • Order tracking, returns and changes
  • Contracted pricing held per account

M4AI Commerce agents

Four agents across
the buying journey.

Salesforce ships merchant, buyer and personal shopper agents with Commerce Cloud. M4AI Commerce packages them with the post-purchase work, the guardrails and the catalogue groundwork that make them dependable.

01

Commerce Cloud · Storefront

Personal shopper agent

A conversational concierge on your storefront that narrows a large catalogue using the shopper's own words, recalls past purchases and similar customers' choices, and adapts when they say it is the wrong colour.

02

Commerce Cloud · B2B

B2B buyer agent

Gives business buyers a consumer-grade way to find, configure and reorder contracted products, with their negotiated pricing and entitlements enforced by default.

03

Commerce Cloud · Service

Post-purchase agent

Answers where-is-my-order, returns and modification requests inside the commerce experience, so the buyer is not handed off to a separate service queue.

04

Commerce Cloud · Operations

Merchandiser agent

Works the operational side — building storefront changes in natural language, spotting products bought together, and proposing promotions for slow-moving inventory.

What an agent needs to work

Not a chatbot
in a checkout.

An agent that sells has to know the product, be allowed to act, and be stopped from acting outside your rules. In commerce those three constraints are unusually concrete, which is why agentic commerce is moving faster than most AI use cases.

The five things an agent needs

Salesforce frames an agent by role, data, actions, guardrails and channel. In commerce all five matter at once: a role, trusted machine-readable product data, API-driven actions, a trust layer, and a channel the buyer already uses.

Product data as the foundation

Structured catalogue data — including schema.org and GS1 style attributes — is what makes a product machine-readable. An agent that cannot tell a width from a weight cannot recommend anything useful.

Actions, not just answers

The difference between a chatbot and an agent is that it completes the action: adds to the order, changes the address, starts the return. That means the APIs behind each action have to already exist and be safe to call.

How we start

Narrow first.
Then earn the rest.

Agentic commerce is easy to demo and hard to make dependable. Most of the work is not the agent — it is the catalogue, the entitlements and the actions the agent is allowed to take.

  1. 01

    Choose the side of the counter

    Start with an internal merchandising agent or a bounded post-purchase case — rarely open-ended sales on the storefront first.

    One scoped agent

  2. 02

    Make the catalogue machine-readable

    Attributes, variants, availability and pricing cleaned up so retrieval returns the right product every time.

    Trusted product data

  3. 03

    Wire the actions, not just the answers

    Cart, order, returns and account APIs surfaced as agent actions with permission checks at the boundary.

    Safe action layer

  4. 04

    Measure on conversion and deflection

    Basket value, search-to-cart, returns handled without a human. Widen the role only when those hold.

    Evidence to scale

Questions we get asked

The short
answers.

Yes. Commerce Cloud is the storefront and order platform. Salesforce also ships embedded agents with it — the merchant, buyer and personal shopper agents. M4AI Commerce is the packaged layer around those: the catalogue and data work, the extra agents for returns and post-purchase, the guardrails, and the channel integrations.

Selling across
too many channels?

Tell us how your customers and partners buy today. We will map where the process breaks, what an agent could take on, and what has to be true in your catalogue first.

Ask a Mavent