Agentic commerce in B2B connects purchasing requests to the rules of a business relationship. An agent can find suitable items, prepare repeat orders or negotiate quotations. To be useful in daily operations, it needs to account for customer prices, responsibilities, technical requirements and approvals. Much of the value lies in making the knowledge held by sales and procurement available within the workflow.
The request starts before the cart
A buyer rarely just orders “twelve parts”. They need a replacement for a particular machine, a specific pack size or materials for a confirmed project. A product with a similar name may be entirely unsuitable. A low price helps little if the item arrives too late or lacks a required approval.
Start with the requirement. An illustrative request might be: “Reorder the approved filters for our equipment. The current item number is in the last order. Delivery must arrive before the scheduled maintenance.” This example is a working model. It shows the information an agent needs to connect: order history, technical suitability, quantity and delivery date.
Customer terms need the right context
A public list price and a negotiated contract price answer different questions. The agent needs access to the terms of the signed-in business customer. Those include volume pricing, pack sizes, currency and the delivery terms that apply to this order.
Permissions need to reach the person or role authorised to act. An employee may be allowed to search for products without being allowed to approve a purchase. A procurement team may have a budget shared by several orders. Those rules belong in the responsible systems. The agent works with them and makes outstanding decisions visible to people.
| Step | The agent's task | What the business defines |
|---|---|---|
| Understand the need | Combine history and requirements | Which data and customer accounts are accessible |
| Prepare the offer | Determine items, quantities and valid prices | Which terms and delivery commitments apply |
| Negotiate | Answer a price request within the permitted range | Discount limits and responsibility for exceptions |
| Obtain approval | Present the complete proposal | Who may approve each type of order |
| Order | Execute the approved request | Duplicate prevention and a binding completion |
| Follow up | Associate status and deviations with the order | Who takes over when something goes wrong |
A concrete Lab example: the Merchant Quote Agent
The Merchant Quote Agent from the Shopware Agentic Commerce Lab handles price requests on B2B quotations. The merchant sets the discount limit. Conventional code checks offers against that limit. Requests about shipping, payment terms or product changes go to people. Decisions can be traced through the dashboard.
The project is published as an alpha. For initial use, its repository describes a draft mode in which replies are approved before being sent. That provides a concrete starting point for deciding which sales tasks to delegate and where personal discussion retains its value. The prerequisites listed in the repository determine the choice of a test environment.
Negotiation needs an economic measure of success
An automatically granted discount does not by itself make a better deal. Compare completions, margin, handling time and the number of handovers. An agent that closes quickly but regularly concedes too much is pursuing the wrong goal. An inflexible agent can also lose useful business.
The Lab's RFQ Bench examines these questions in controlled negotiations. It records conditions and negotiation traces and considers captured value, agreement rates, cost and runtime, among other measures. Use this kind of comparison to assess a strategy under consistent conditions. Benchmark results and outcomes from actual customer orders remain separate observations.
Start with a familiar business relationship
A useful first application is reordering known products for an existing group of customers. Requirements can be traced, terms are available and people know the usual exceptions. That makes deviations easier to recognise.
Test a changed pack size, expired prices, a blocked product and two simultaneous purchase attempts in particular. The agent purchase acceptance guide provides a starting point. Plan capacity for human approvals as well: an agent can prepare more requests than the team can handle.
What changes for B2B merchants
A well-prepared catalogue becomes usable technical knowledge. Sound customer records enable appropriate offers. Reliable order processes make further delegation possible. Differentiation then also comes from how well a merchant understands and fulfils a complex requirement.
That creates room for distinctive services: assisted spare part selection, orders aligned with maintenance schedules or offers tailored to a customer's project. The protocols connect the systems. The Lab projects provide concrete experiments, and the handbook examines the business implications.