What B2A is
B2A (business-to-agent) describes a deal in which an AI agent stands between the store and the person. The person sets the task and the limits — budget, timing, preferences — and the agent searches, compares and places the order. The store’s direct counterpart changes: it is no longer a site visitor but a machine customer that reads data and calls APIs.
The model applies both to retail and to corporate purchasing: an agent can top up office supplies just as it can build a family’s grocery basket.
B2C, B2B and B2A compared
| Parameter | B2C | B2B | B2A |
|---|---|---|---|
| Who chooses | A person | A company’s buying group | An AI agent on behalf of a person or company |
| What drives the choice | Brand, reviews, convenience, emotion | Price, contract, supplier reliability | Total price, availability, delivery time, return terms, reviews |
| Point of contact | Website, app, advertising | Account manager, tender, customer portal | Feed, API, agentic protocol |
| How loyalty shows up | Habit, loyalty programme | Contract | A rule the person gave the agent |
| What kills the deal | An awkward checkout | Opaque terms | Missing data, a CAPTCHA, an unstable API |
What you sell to an agent
- A machine-readable catalogue. Complete attributes, current prices and stock in the feed and in structured markup — see agent-ready product feed.
- Protocols. A standard way to check out without clicking through the page layout: ACP (OpenAI and Stripe, September 2025), UCP (Google, NRF 2026), YCP (Yandex, 27 February 2026, for Yandex’s ecosystem in Russia).
- Transparent terms. Total price including delivery, delivery time, returns and warranty as separate fields, not small print in the terms and conditions.
- Interaction quality. The agent has to finish the task without dead ends — that is agent experience.
What is left for marketing
An agent needs no emotion — it needs data that lets it compare offers. But the decision about whom to trust with a purchase is still made by a person: they pick the agent, allow or block brands and stores, and set limits. So the brand works at the level of human trust, while at the agent level the winner is whoever has more accurate data and clearer terms.
Important: in B2A you do not lose because your creative is weaker, you lose because the agent could not read your data or judged it unreliable.
First steps for a store
- Look at a product through an agent’s eyes: which attributes, price and terms are available without rendering the page.
- Compare your total price including delivery with competitors — that is the number an agent compares, not the shelf price.
- Pick a protocol by audience: ACP or UCP for international assistants, YCP for the Yandex ecosystem.
- Check that bot protection does not block legitimate agents on the way to checkout.
- Track agent orders as a separate channel in your analytics.