> What a B2B shop must expose so buyer-side procurement agents can find, compare and order: product data, price truth, quotes, identity and protocols.
>
> Web page: https://balazscsorba.com/blog/procurement-agent-buyer-side · Language: English · Also available in: [Deutsch](https://balazscsorba.com/de/blog/procurement-agent-buyer-side.md) · [Magyar](https://balazscsorba.com/hu/blog/procurement-agent-buyer-side.md)
> Author: Balázs Csorba · Published: 2026-10-15 · Keywords: agentic commerce readiness, buyer-side procurement agents, what B2B shops must expose to AI agents, UCP integration, ACP agentic checkout, AP2 mandates supplier, machine-readable quotes B2B, product data for AI agents, punchout vs agentic commerce, AI agent traffic ecommerce

[Blog](https://balazscsorba.com/blog)/Web engineering

# Agentic commerce readiness: what buyer-side procurement agents need from your B2B shop

What a B2B shop must expose so buyer-side procurement agents can find, compare and order: product data, price truth, quotes, identity and protocols.

[Balázs Csorba](https://balazscsorba.com/about)·October 15, 2026·9 min read

-   Agentic commerce
-   Procurement
-   B2B shop
-   Product data
-   UCP

![Three cards slide toward a basket, a gate with a check lets one through.](https://balazscsorba.com/images/blog/procurement-agent-buyer-side/cover.webp?v=8d8d5bcb87)

## Key takeaways

-   Buyer-side agents are already in procurement suites: SAP planned a Joule agent for bid analysis in SAP Ariba from Q1 2026, and Amazon Business reports $60B in annualized gross sales with AI buying tools.
-   Agentic commerce readiness is a stack: clean IDs and specs, markup and feeds, truthful price and stock, a quote API, punchout or order APIs, agent identity, and only then protocol endpoints.
-   UCP, ACP and AP2 cover discovery, checkout and payment mandates for list-price goods; negotiated B2B quotes have no adopted standard as of October 2026.
-   Adobe measured US retail product pages as the least machine-readable page type at 66%, so the product page is where most suppliers lose agent visibility.
-   Do not expose contract prices publicly to look agent-ready; a structured, authenticated quote endpoint keeps B2B price rules intact.

On this page

1.  [Are buyer-side procurement agents real yet?](https://balazscsorba.com/#buyer-side-agents-are-real-now)
2.  [What do procurement agents actually do?](https://balazscsorba.com/#what-procurement-agents-actually-do)
3.  [The supplier readiness stack](https://balazscsorba.com/#the-supplier-readiness-stack)
4.  [UCP, ACP and AP2 on one page](https://balazscsorba.com/#protocols-in-one-page)
5.  [What product data can agents read?](https://balazscsorba.com/#structured-catalogs-agents-can-read)
6.  [What about negotiated quotes?](https://balazscsorba.com/#machine-readable-quotes-and-the-b2b-gap)
7.  [What is the business case for agentic commerce readiness?](https://balazscsorba.com/#the-business-case)
8.  [When should you not rush agent protocols?](https://balazscsorba.com/#when-not-to-rush)
9.  [An agentic commerce readiness checklist](https://balazscsorba.com/#readiness-checklist)
10.  [Sources](https://balazscsorba.com/#sources)

Listen to this article

0:000:00

**Agentic commerce readiness** is the state in which software agents working for a buyer can find your products, read correct specs, prices and availability, ask for a quote or build a cart, and pass the result into the buyer’s approval process. In B2B, the demand side is moving first: procurement suites are adding agents that source, compare and prepare purchases, and suppliers whose data those agents cannot read drop off the shortlist without noticing.

This article is the supplier-side checklist: what buyer agents do today, the seven layers a shop needs in dependency order, a one-screen map of UCP, ACP and AP2, the honest gap around negotiated quotes, and what the business case can and cannot claim as of October 2026. The protocol details themselves are in [the UCP, ACP and AP2 guide](https://balazscsorba.com/blog/agentic-commerce-protocols-ucp-acp-guide).

## Are buyer-side procurement agents real yet?

Yes, inside the procurement platforms large buyers already use, and with clear limits. Contracts and purchase orders are still approved by people, protocol checkout started in the US, and agent traffic in B2B grows from a small base.

-   **SAP Ariba:** SAP announced in October 2025 a next-generation Ariba rollout from Q1 2026 with a Joule agent for bid analysis that compares supplier bids on total cost, including unit prices, shipping and payment terms, plus AI summaries of supplier responses and intake management that routes requests across SAP and non-SAP systems ([SAP News](https://news.sap.com/2025/10/sap-connect-spend-management-procurement/)).
-   **Amazon Business:** $60 billion in annualized gross sales in Q2 2026, more than 11 million organizations, 1.8 million of them new in the first half of 2026, with AI tools such as Amazon Business Assistant, Savings Insights and Spend Anomaly Monitoring ([Amazon, 21 Jul 2026](https://press.aboutamazon.com/2026/7/amazon-business-reaches-60-billion-in-annualized-gross-sales-serving-over-11-million-organizations-worldwide)).
-   **Open protocols:** Google launched the Universal Commerce Protocol on 11 January 2026 with Shopify, Etsy, Wayfair, Target and Walmart, starting with checkout on eligible listings in AI Mode and the Gemini app ([Google](https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/)); Stripe and OpenAI publish the Agentic Commerce Protocol; Google’s AP2 defines payment mandates.

## What do procurement agents actually do?

They do the reading and comparing that buyers used to do in spreadsheets: find suppliers, compare specs and total cost, prepare a request for quotation, summarize bids, watch spend for anomalies, and place routine orders inside policy. Each activity depends on a specific kind of supplier data.

![Buyer-agent purchase flow with human gates Five steps from left to right: a requisition approved by a human; the agent discovers suppliers through feeds, UCP or punchout catalogs; it compares specs, prices and lead times; it requests a quote or builds a cart; and the order goes through human approval in the buyer’s ERP before a purchase order is issued. The first and last steps are human gates.](https://balazscsorba.com/images/blog/procurement-agent-buyer-side/diagram-7.webp?v=32954afe06)

Buyer-agent purchase flow: the agent does the search and comparison in the middle, while requisition approval and the purchase order stay with people.

Discovery needs canonical product identifiers and structured attributes. Comparison needs unambiguous price semantics: list or contract price, unit of measure, pack size, tax and freight. Quote preparation needs a machine-readable answer with validity and lead time. Ordering needs a path that ends in the buyer’s system, which for enterprise customers usually means punchout. The two ends of the flow stay human, which matches Gartner’s prediction that by 2030, 75% of B2B buyers will prefer sales experiences that prioritize human interaction ([Gartner, Aug 2025](https://www.gartner.com/en/newsroom/press-releases/2025-08-25-gartner-says-by-2030-that-75-percent-of-b2b-buyers-will-prefer-sales-experiences-that-prioritize-human-interaction-over-ai)).

## The supplier readiness stack

Readiness is seven layers built from the bottom up, and each layer is worthless if the one below it is wrong. A protocol endpoint that serves an incorrect price or a part number the buyer’s ERP cannot match does not make a supplier agent-ready; it makes the error machine-readable.

![Supplier readiness stack for buyer-side agents Seven layers from bottom to top: product IDs and specs; schema.org markup and Merchant Center feed; truthful price and stock including list versus contract price and unit of measure; a quote API with lines, validity and lead time; punchout and order APIs; agent identity and request signing; and agent protocols UCP, ACP and AP2 on top. An arrow on the left shows the build order from bottom to top, and a note says no layer fixes a broken layer below it.](https://balazscsorba.com/images/blog/procurement-agent-buyer-side/diagram-11.webp?v=b326a239dd)

The supplier readiness stack: build from the bottom up, because no protocol endpoint fixes wrong product data, prices or stock below it.

1.  **IDs and specs:** canonical part numbers, GTINs where they exist, manufacturer part numbers, and attributes as fields rather than prose. In DACH B2B the exchange format for this is still BMEcat, an XML catalog standard published by the German association BME, licence-free, with 2005.2 as its latest version ([ECLASS](https://eclass.eu/support/technical-specification/data-model/bmecat)).
2.  **Markup and feed:** product pages with schema.org Product and Offer markup, plus a Merchant Center feed where Google surfaces are relevant.
3.  **Price and stock truth:** published list prices that match the checkout, availability that matches the warehouse, and explicit units of measure.
4.  **Quote API:** an authenticated endpoint that returns customer-specific prices, validity and lead time as structured data.
5.  **Punchout and orders:** cXML or SAP OCI punchout for enterprise buyers, and an order API for direct integrations.
6.  **Identity and signing:** a way to tell a signed buyer agent from an anonymous scraper, and to sign your own agent requests.
7.  **Agent protocols:** UCP, ACP and AP2 endpoints, once the layers below are reliable.

After more than ten years of B2B commerce and PIM work with Spryker and Pimcore, my view is that layers one to three are a product data project with an API on top. Part numbers, units of measure, variant structure and the visibility of contract prices decide whether an agent can buy or only browse. If master data flows from ERP to PIM to shop already, the [Pimcore ERP delta sync](https://balazscsorba.com/blog/pimcore-erp-delta-sync) pattern is the place to make those fields reliable.

## UCP, ACP and AP2 on one page

UCP describes what a business can do and how an agent checks out, ACP is an agent-facing checkout that merchants accept or decline per agent, and AP2 carries the buyer’s authorization as signed mandates. All three assume list-price goods and a cart; none of them models a negotiated B2B quote.

-   **UCP:** a business publishes its capabilities in a JSON manifest at `/.well-known/ucp`; agents open sessions with `POST /checkout-sessions` and send `UCP-Agent`, `request-signature` and `idempotency-key` headers, over REST, MCP or A2A ([Google Developers Blog](https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/)). Google’s implementation requires an active Merchant Center account, reuses existing feeds, and keeps the merchant as merchant of record ([Merchant Center UCP guide](https://developers.google.com/merchant/ucp)).
-   **ACP:** an Apache-2.0 standard from Stripe and OpenAI; merchants stay merchant of record, choose which products are sellable, and can accept or decline per agent or per transaction. ChatGPT is the first implementing platform and Stripe the first payment provider, with its Shared Payment Token; discovery of participating merchants is still being built ([agenticcommerce.dev](https://agenticcommerce.dev/)).
-   **AP2:** an extension for A2A and UCP launched on 16 September 2025, in which checkout mandates and payment mandates are verifiable digital credentials, open for constraints and closed for a finalized checkout; standardization continues in FIDO Alliance working groups ([ap2-protocol.org](https://ap2-protocol.org/)).

Buyer channel

What the supplier must expose

Effort

Limits as of October 2026

Google AI Mode, Gemini (UCP)

Merchant Center feed, checkout-eligible products, UCP profile

medium

US-first, gated behind Merchant Center

ChatGPT (ACP)

ACP checkout endpoint, product selection, payment provider support

medium

onboarding per platform, merchant discovery still being built

SAP Ariba, Coupa and similar

punchout (cXML, OCI), catalogs, order and invoice documents

medium to high

per-buyer setup, agents work inside the suite

Direct buyer agents (API)

authenticated quote and order API, signed requests

medium

no standard for quotes, contracts per customer

## What product data can agents read?

Agents read what is explicit: structured data on the product page, feeds, and APIs. Google’s product documentation describes merchant-listing markup for pages where customers can buy, covering price, availability, shipping and returns plus variants, recommends providing both structured data and a Merchant Center feed, and asks for return policies under Organization markup ([Google Search Central](https://developers.google.com/search/docs/appearance/structured-data/product)).

The weak spot is the product page itself. Adobe’s analysis of more than a trillion visits to US retail sites scored machine readability at 75% for homepages, 74% for category pages and only 66% for product pages, so about a third of product-page content is not readable by language models ([Adobe](https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable)). That is retail data, but B2B product pages, with specs in PDFs and prices behind logins, are unlikely to do better. The B2B-specific markup issues are covered in [schema.org structured data for B2B shops](https://balazscsorba.com/blog/schema-org-structured-data-b2b-shop).

## What about negotiated quotes?

As of October 2026 no adopted standard exists for machine-readable negotiated quotes, and that is the main B2B gap. Agents can compare published offers, but contract prices, volume tiers, project pricing, validity, Incoterms, lead time and minimum order quantity live in the supplier’s ERP and in sales conversations.

The practical middle path is a structured quote endpoint behind authentication. It mirrors what a punchout session already does on the cart side: the buyer is identified, the price is resolved for that customer, and the answer is a document the buyer’s system can store. The shape below is a sketch, not a standard:

```
// Structured quote response (sketch, not a standard): authenticated buyer or buyer agent only
{
  "quoteId": "Q-2026-104233",
  "customer": "C-40017",
  "currency": "EUR",
  "validUntil": "2026-11-15",
  "incoterms": "DAP",
  "lines": [
    {
      "sku": "BRK-6204-2RS",
      "gtin": "4012345678901",
      "quantity": 500,
      "unitOfMeasure": "PCE",
      "unitPrice": 2.84,
      "priceBasis": "contract",          // list | contract | tier | project
      "minimumOrderQuantity": 100,
      "leadTimeDays": 6,
      "availableToPromise": 420           // live ERP figure, not a cached feed value
    }
  ],
  "audit": { "requestedBy": "agent:buyer-procurement", "onBehalfOf": "user:p.huber" }
}
```

Three rules matter more than the field names. Only authenticated buyers and their delegated agents get contract prices. Every quote response is logged with who asked on whose behalf. And `availableToPromise` comes from the ERP at request time, because a quote built on a cached stock figure is a promise the warehouse cannot keep.

**Do not publish contract prices to look agent-ready**

Exposing customer-specific prices in public markup or feeds breaks the confidentiality your contracts rely on and lets competitors read your price structure. Publish list prices where you have them, and give authenticated agents a quote endpoint. Request signing, for example Web Bot Auth with HTTP Message Signatures (RFC 9421) as described by [Cloudflare](https://blog.cloudflare.com/web-bot-auth/), helps tell a signed agent from a scraper, but it is not authorization.

## What is the business case for agentic commerce readiness?

The business case is protecting demand, not adding a guaranteed channel. Published numbers come from adjacent markets, so they are directional for B2B. No published study shows a B2B conversion lift from agent readiness, and none should be promised.

-   Adobe measured AI-sourced traffic to US retail sites up 393% year on year in Q1 2026 and 693% in the 2025 holiday period, and in March 2026 that traffic converted 42% better than non-AI traffic ([Adobe](https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable)).
-   Gartner forecasts, as reported by Digital Commerce 360, that AI agents will intermediate more than $15 trillion in B2B spending by 2028 ([DC360, Nov 2025](https://www.digitalcommerce360.com/2025/11/28/gartner-ai-agents-15-trillion-in-b2b-purchases-by-2028/)). It is a forecast, and the number your sales leadership has probably already heard.
-   Amazon Business at $60 billion annualized shows enterprise buying consolidating on platforms that can be instrumented with agents, which is the competitive pressure on independent B2B shops.

For a head of e-commerce at a distributor or manufacturer, the honest framing is that agent traffic in B2B is small today, catalog readiness is cheap relative to the channel it protects, and the same work improves on-site search, Google Shopping and punchout catalogs. Measure AI-referred sessions, signed agent requests, catalog completeness, availability accuracy and quote-endpoint uptime, and decide on protocol work when those numbers move.

## When should you not rush agent protocols?

Skip protocol checkout when your categories are low-value or heavily negotiated, when your buyers order through punchout already, or when the lower layers are not reliable yet. UCP checkout is US-first and gated behind Merchant Center, ACP requires onboarding per platform, and punchout remains the supported enterprise path today.

## An agentic commerce readiness checklist

1.  Audit identifiers: every sellable item has a canonical part number, and GTIN and manufacturer part number where they exist.
2.  Move key specs from PDFs and prose into attributes, and export them in your feed and BMEcat catalogs.
3.  Make markup and feeds truthful: price, availability, shipping and returns must match checkout and warehouse.
4.  Automate availability from the ERP and monitor the gap between published and actual stock.
5.  Design an authenticated quote endpoint with validity, lead time and an audit trail.
6.  Log and, where possible, verify signed agent requests separately from browser traffic.
7.  Run a protocol spike, such as a `/.well-known/ucp` profile against the spec samples, and test it end to end; I use Playwright MCP for this kind of round-trip check.
8.  Agree commercial rules for agent orders with sales: minimums, freight, returns, and who may see which price.

A readiness audit of this kind, read-only and followed by a short implementation spike, is a well-defined piece of work for a senior freelance engineer or a small studio. If you want help with the data and API layers, see my [B2B e-commerce work](https://balazscsorba.com/expertise/b2b-ecommerce-developer).

## Sources

1.  [Google: new agentic commerce tools and the Universal Commerce Protocol (11 Jan 2026)](https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/)
2.  [Google Developers Blog: Under the hood of the Universal Commerce Protocol (11 Jan 2026)](https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/)
3.  [Google Merchant Center: UCP integration guide](https://developers.google.com/merchant/ucp)
4.  [Agentic Commerce Protocol (ACP)](https://agenticcommerce.dev/)
5.  [Agent Payments Protocol (AP2)](https://ap2-protocol.org/)
6.  [Google Search Central: Product structured data](https://developers.google.com/search/docs/appearance/structured-data/product)
7.  [Adobe: AI traffic surges to retail sites, but many are not machine-readable](https://business.adobe.com/blog/ai-traffic-surge-retail-sites-not-machine-readable)
8.  [Cloudflare: Web Bot Auth, message signatures for bots and agents](https://blog.cloudflare.com/web-bot-auth/)
9.  [SAP News: SAP Connect, spend management and procurement (8 Oct 2025)](https://news.sap.com/2025/10/sap-connect-spend-management-procurement/)
10.  [Amazon: Amazon Business reaches $60 billion in annualized gross sales (21 Jul 2026)](https://press.aboutamazon.com/2026/7/amazon-business-reaches-60-billion-in-annualized-gross-sales-serving-over-11-million-organizations-worldwide)
11.  [Digital Commerce 360: Gartner, AI agents and $15 trillion in B2B purchases by 2028 (28 Nov 2025)](https://www.digitalcommerce360.com/2025/11/28/gartner-ai-agents-15-trillion-in-b2b-purchases-by-2028/)
12.  [ECLASS: BMEcat technical specification](https://eclass.eu/support/technical-specification/data-model/bmecat)

## Frequently asked questions

What is agentic commerce readiness for a B2B supplier?

It is the state in which software agents acting for a buyer can find your products, read correct specs, prices and availability, request a quote or build a cart, and hand an order to the buyer’s approval process. In practice it is mostly product data quality, price and stock accuracy, and an API or punchout path, with protocol endpoints on top.

Do B2B shops need to implement UCP or ACP now?

Not first. UCP checkout launched US-first and requires Google Merchant Center, and ACP onboarding runs per AI platform. For most B2B suppliers the earlier layers, clean product data, truthful availability and a quote or punchout path, bring more value now and are prerequisites for any protocol.

How do procurement agents handle negotiated B2B prices?

As of October 2026 there is no adopted standard for machine-readable negotiated quotes. Agents can compare published list offers, but contract prices, volume tiers, validity and lead times need an authenticated quote API or the punchout session the buyer already uses.

Should a B2B shop publish contract prices so agents can read them?

No. Contract prices belong behind authentication. Publish list prices and offer markup where they exist, and give authenticated buyers and their agents a quote or price endpoint that resolves the customer-specific price with an audit trail.

How can I measure whether AI agents reach my B2B shop?

Track AI-referred sessions and signed agent requests in your logs, catalog completeness for key attributes, the gap between published and actual availability, and uptime and error rates of your quote or order API. These are leading indicators; orders placed by agents come later.

Written by Balázs Csorba

Senior fullstack & AI engineer in Styria, Austria – 10+ years of Vue, Nuxt, Node.js and PHP, now building tooling for AI agents.

[B2B e-commerce & PIM →](https://balazscsorba.com/expertise/b2b-ecommerce-developer)[About me →](https://balazscsorba.com/about)

## More articles

-   [Schema.org structured data for B2B shops: prices behind logins, units and variants](https://balazscsorba.com/blog/schema-org-structured-data-b2b-shop)
-   [An AI assistant for B2B shop support: tools, permissions and the human handoff](https://balazscsorba.com/blog/ai-assistant-b2b-shop-customer-support)
-   [Streaming AI chat in Vue and Nuxt: accessible by design](https://balazscsorba.com/blog/vue-ai-chat-ui-streaming-accessibility)
-   [An LLM translation pipeline for i18n: glossary, placeholders and review](https://balazscsorba.com/blog/llm-translation-pipeline-i18n)

## Sounds like what you need?

Tell me about your project or role – I’d love to hear from you.

[Get in touch](mailto:contact@balazscsorba.com) [Connect on LinkedIn](https://www.linkedin.com/in/balazs-csorba)