Most of the attention on AI shopping agents goes to consumers asking a chatbot for a gift idea. The place agents are more likely to buy at scale first is less visible and far more predictable. Business purchasing.
B2B buying is recurring, follows rules and is usually dull for the people doing it. Those are exactly the conditions where AI agents work well. For B2B and omnichannel brands, the question is not whether customers will send agents to place orders, but whether your systems will be able to answer them.
Why B2B ecommerce suits AI agents
Federal Reserve Governor Christopher Waller made the point in a September 2026 speech, calling business purchases "fertile ground" for agent-led buying because they repeat and follow preset rules such as approved suppliers and budget limits. Gartner has gone further, forecasting that AI agents will intermediate more than $15 trillion in B2B spending by 2028.
Forecasts are forecasts, but the logic holds. Consider what a typical B2B order involves.
- The same products, ordered again, on a schedule
- Prices already agreed in a contract
- Approval rules that depend on amount, category or budget
- Purchase orders, payment terms and invoices
- Buyers who would rather do almost anything than log into another supplier portal
Every one of those steps is rule-based. That makes it a good fit for an agent and a poor use of a purchasing manager's time.
What a buying agent needs from your business
A customer's purchasing agent will not browse your catalog the way a person does. It will try to ask your systems direct questions. Whether it gets answers depends on what you expose.
- Account-specific pricing. The agent needs that customer's contract price, not your list price.
- Real-time stock and lead times. A reorder is only useful if the agent knows when it will arrive.
- Order history. Reordering starts with knowing what was ordered last time.
- Quote requests. For larger or custom orders, a way to request and receive a quote without email back and forth.
- Approvals and purchase orders. Support for the customer's approval steps, PO numbers and payment terms.
- Product documentation. Specifications, safety data and compliance documents an agent can retrieve and pass on.
The integration is the project
In most B2B businesses, that information is spread across several systems. Pricing and terms live in the ERP, such as NetSuite. Stock and production live in inventory or manufacturing tools such as Katana or Cin7. The online catalog lives on a commerce platform such as Shopify Plus or BigCommerce, both of which support B2B buying.
If those systems only talk to each other through a nightly sync, an agent will get yesterday's answer. If pricing logic lives in a spreadsheet a sales rep keeps, the agent gets no answer at all. That is why the work that makes a B2B business ready for agents is mostly integration work, the same work that makes self-serve ordering better for human buyers. We cover the architecture side in why AI agents need a composable architecture.
Three places B2B brands can start
- Make reordering effortless. Give existing accounts a fast path to reorder with their pricing and terms, and expose the same capability through an API. It helps human buyers now and agents later.
- Use agents on your side first. An internal agent that drafts quotes from a customer's request for a sales rep to approve is lower risk than letting outside agents transact, and it teaches you where your data is weak.
- Make order status and documents self-serve. Where is my order and can you send the spec sheet are some of the most common B2B service requests. Answering them automatically frees your team and is exactly what a buying agent will ask.
This is the problem we built Compass Commerce to address, connecting the systems a mid-market brand already runs so agents can work across them with permissions in place.
The risks to plan for
- Pricing errors at scale. A wrong price shown to one buyer is a phone call. A wrong price served to every agent is a margin problem.
- Permissions. An agent acting for one customer must never see another customer's pricing, orders or terms.
- Credit and terms. Orders need to respect credit limits and payment terms automatically.
- Approval thresholds. Decide which orders an agent can complete alone and which need a person to approve.
For more on verifying agents and handling disputes, read when an AI agent buys from your store.
Common questions
Will AI agents replace B2B ordering portals?
Not entirely. Portals will still serve buyers who want to browse, compare and manage accounts. But much of the routine reordering that happens in portals today is likely to move to agents and integrations over time.
Are our customers using purchasing agents yet?
Most are experimenting rather than relying on them, and adoption will be uneven by industry and company size. The integration work pays off now for human buyers either way, so there is little risk in starting early.
Do we need a new commerce platform to support AI agents?
Not necessarily. The bigger question is whether your current platform can expose account pricing, stock and ordering through APIs. If it cannot, that is when a replatform conversation becomes worth having.
Which integration matters most?
Usually the ERP, because it holds pricing, terms and credit. If pricing and stock are not reliable and current, nothing built on top of them will be either.
How do we keep contract pricing confidential?
With strict account-level permissions on every API and agent connection, tested before launch. An agent should only ever see what the buyer it represents is allowed to see.
Where to start
Map where your pricing, stock and order data actually live, and how current each one is. That map is your agent readiness plan. If you want help building it, our B2B and omnichannel commerce team works on exactly this. Get in touch.
