B2B Pulse: B2B eCommerce Industry Updates

This week's AI commerce news is about everything being built around the shopping assistant: product data, purchasing authority, negotiated pricing, implementation, payment rails, and the physical distribution network that still has to move the order. The demos are getting easier, while the operating model is becoming the real work.

That is particularly true in B2B eCommerce, where a buyer may have permission to purchase one product, from one supplier, for one location, within one budget, at a price that does not appear on the public website. “Add to cart” was never the whole transaction.

B2BEA partner watch: Oro goes after implementation economics

OroCommerce launched OroMomentum, an AI-enabled program focused on implementation, custom development, and major-version upgrades. Oro says the program can cut implementation and custom-feature time by up to half and make upgrades up to twice as fast. Those are vendor-reported results and will need to be tested against real projects.

This is one of the more interesting platform announcements of the week because it applies AI to the cost of delivering B2B eCommerce across implementation and ongoing platform management, as well as the storefront.

Implementation effort and upgrade debt are often bigger barriers than the license. Manufacturers and distributors need account hierarchies, pricing rules, workflow, ERP connections, and custom logic. Then someone has to carry all of that through the next upgrade.

If OroMomentum works as described, buyers will expect clearer answers about test coverage, code review, security, and support across future versions. Agencies will also face more scrutiny of effort-based pricing for repeatable engineering work. AI may change the services model before it changes the storefront.

Product data has become the AI readiness test

New research from Akeneo found that 95% of 1,000 senior IT decision makers in the United States and Europe believe their organizations are ready for AI-driven commerce. At the same time, 44% of U.S. respondents said more than half of their AI project effort goes to data preparation.

The distance between that confidence and the amount of preparation required says more about AI readiness than either number on its own. AI brings missing attributes, inconsistent taxonomies, conflicting product information, and brittle integrations into the same workflow at once. A recommendation can be slightly wrong and still look clever, while an agent that uses the wrong price, compatibility detail, or regional compliance information creates a much more expensive problem.

For manufacturers and distributors, product data now sits inside search, service, purchasing, and the control layer for AI.

Partner search updates get closer to how industrial buyers actually buy

HawkSearch announced that an industrial filtration supplier selected its platform to combine account-level purchase history with product discovery, replacement-part matching, alternatives, manuals, and technical documentation. The customer is unnamed, so this is a vendor-reported adoption signal rather than independent proof.

The use case is still worth paying attention to. Industrial buyers often arrive with an old part number, an incomplete description, or the memory of what they bought three years ago. Search that understands account history and compatibility is more useful than a polished natural-language demo sitting on top of generic keyword results.

VTEX also changed the fallback ranking in Intelligent Search. When a query cannot match every term, its OR search now gives more weight to rare, distinctive words and less to common terms such as units of measure. VTEX says OR fallback accounts for about 5% of searches and as much as 15% in some stores. The update is automatic.

The change sounds small until a technical catalog ranks a “200-tablet organizer” above the ibuprofen a buyer asked for. Search quality is often won in these unglamorous edge cases.

Commerce agents are becoming reusable architecture

Anthropic released working blueprints for shopping and merchant agents, including integrations for catalogs, carts, checkout, customer preferences, inventory, and order history. The shopping agent builds the cart but hands the transaction to merchant-controlled checkout. The merchant agent can recommend price and promotion changes, but a person approves them before they go live.

The architecture matters more than the reported retail results. Anthropic says retailers using Claude agents have seen carts up to 35% larger and purchase completion improve by as much as 60%. Those figures are company-reported retail results, so their relevance to B2B still needs to be demonstrated.

The separation of responsibilities is more useful: the agent can sit across an existing commerce stack while catalog data remains a source of truth, checkout stays with the merchant, and price changes retain human approval.

That is a credible starting point for B2B. The next layer includes account pricing, order permissions, inventory by location, approved substitutes, credit status, and reliable order services. The agent layer can travel across platforms, while the commercial context remains specific to the business.

Two B2BEA partners are approaching the same shift from different sides. commercetools' AgenticLift is intended to make products discoverable and purchasable in AI channels while preserving the existing commerce infrastructure. Optimizely introduced purpose-built AI models for marketing, along with an open benchmark covering 285 tasks and more than 6,000 criteria. Optimizely says its post-trained models produced 10 times greater cost efficiency than general frontier models in early testing.

Platform competition is moving past “we have an agent.” The useful questions are what the agent can see, what it can change, how its work is measured, and whether it can operate inside the stack a business already has.

Purchasing authority is becoming infrastructure

EMVCo opened a draft agentic-payments framework for public feedback through September 30. Its proposed Intent Services would let participants register, retrieve, and manage the purchasing authority granted to an AI agent before, during, and after a card transaction. The current scope includes recurring purchases, cumulative budgets, and post-transaction activity.

EMVCo helps set the technical specifications used across the card ecosystem, which gives this framework unusual weight among agentic-payment announcements.

Autonomous purchasing needs a durable answer to a basic question: who authorized what, within which limits, and for how long?

In B2B, that authority may include a buyer role, an approved supplier, a cost center, a contract price, a shipping location, and an escalation threshold. Permission becomes part of the order itself.

Visa's new Trust Index makes the adoption problem plain. Seventy-two percent of U.S. consumers have used an AI assistant, but only 23% trust generative AI to handle a payment transaction for them. When Visa was named as the payment provider, the figure rose to 61%.

Mastercard's new report on AI agents and commerce reaches nearly the same conclusion. Trust is becoming machine-readable infrastructure made up of identity, authentication, authorization, and limits.

The B2B transaction is expanding in both directions

Airwallex added direct-debit collection and early-access French e-invoicing capabilities. Airwallex Billing now supports reusable bank mandates across ACH, SEPA, Bacs, and EFT. Its French release adds structured invoices and credit notes, receipt of vendor e-invoices through Bill Pay, and separate tracking for compliance and payment status.

This is where B2B payments are headed, toward a complete invoice-to-cash or invoice-to-pay workflow that includes mandates, failed payments, reconciliation, ERP posting, and regulatory reporting.

On the other side of the transaction, Vertice launched Ana, an agent that builds negotiation strategies, communicates with software vendors, and tracks purchasing and renewal activity. Vertice says Ana has handled more than 4,000 negotiations covering $500 million in spend, with average savings of 18% and purchasing cycles shortened by 15 days. These are vendor-reported results.

Buyer-side negotiation agents could have a material effect on B2B sellers because they will find inconsistent pricing, weak discount governance, and slow approval workflows quickly. Sellers have spent years preparing for digital buyers, and automated buyers add another set of expectations.

Airwallex is automating what happens after commercial agreement, while Vertice is automating part of what happens before it. Agentic B2B commerce is stretching across negotiation, contract, authority, invoice, payment, and auditability.

Order automation is becoming visible, and the back office is moving too

Conexiom launched Relay for RFQs, orders, acknowledgements, shipping notices, and invoices received through email threads, spreadsheets, handwriting, images, and text. It prepares validated data for 40 ERP systems and includes editable playbooks, exception handling, and automation analytics.

Manufacturers and distributors still receive a large volume of orders outside the portal. Document extraction creates value, but visible rules and reusable corrections determine whether the automation can be trusted. A useful evaluation should cover price and quantity conflicts, exception rates by document type, auditability, and rollback behavior.

Microsoft's Dynamics 365 Commerce roadmap also lists September availability targets for multiple-outlet B2B ordering, on-behalf-of support through Entra External ID, asynchronous commerce payments, and pay-by-link for call-center orders. Microsoft notes that planned dates can change, so customers should confirm tenant availability before treating the roadmap as a release note.

Identity, payment handling, outlet structure, and assisted ordering are the daily machinery of B2B commerce, which makes these operational releases worth watching.

Architecture is separating experience velocity from the transactional core

SAP and Vercel announced a new storefront path for SAP Commerce Cloud. Vercel is developing Next.js templates that connect to discovery, content, cart, checkout, and customer journeys. SAP remains responsible for commerce data and transaction processes, while Vercel supplies frontend deployment, scaling, and observability.

The familiar “headless versus monolithic” debate now leads to a more useful operating-model question: who owns frontend uptime, API performance, release governance, integration testing, and the seam between the experience and the transaction?

A faster storefront shifts complexity toward the seams between systems, teams, and release processes.

Security remains an operating task

Adobe issued a priority-one security update for Adobe Commerce, Adobe Commerce B2B, and Magento Open Source. The critical vulnerability, CVE-2026-75650, carries a CVSS score of 10.0, requires no authentication to exploit, and has been used in active attacks.

A commerce platform sits close to customer records, pricing, payments, and operating systems, which gives this security update the same operational importance as a major product release. When a vulnerability allows arbitrary code execution, the response includes both installing the hotfix and assessing whether the environment was already compromised. Innovation and maintenance receive different stages, but both determine whether the platform works.

One thing worth listening to

The latest Inside Commerce podcast episode features Centra founder and Chief Innovation Officer Martin Jensen on the “agent experience” era. The useful premise is that the advantage will not come from access to code or models alone. It will come from the tools, playbooks, guardrails, and organizational context built around them.

That is the throughline across this week's news. Product data gives agents something accurate to work with. Authority tells them what they can do. Platforms connect them to transactions. Payments, order systems, and distribution turn those transactions into business outcomes.

The companies building the product data, authority, payment, and operational infrastructure will decide how much of the order an agent can be trusted to handle.

Disclosure: Akeneo, commercetools, HawkSearch, Optimizely, OroCommerce, and VTEX are B2BEA partners. These relationships did not determine the conclusions in this article.