AI in PIM 2026: What vendors actually deliver today

Almost every PIM vendor is promoting AI in 2026, and many now talk about “agents”. From the outside, it is hard to tell what is usable in the product and what exists only in a press release. We reviewed public information from 17 vendors: help centres, documentation, release notes and press releases, as of October 2026. Our wiki article AI in PIM explains what AI can do in PIM in principle. Here, we look at the current state of the market.

Key takeaways

• Almost all PIM systems can now generate and translate text, classify products and tag images. These capabilities are no longer a differentiator.
• Agents that plan and execute multiple steps are publicly documented by Akeneo, Pimcore and Bluestone. At Syndigo, inriver and Stibo Systems, they are mainly at the announcement stage.
• At least eight vendors have documented an MCP server through which external AI assistants can access product data. This is becoming standard.
• Hardly any vendor publicly states whether an AI feature is generally available or still in beta. You should specify this in the contract.
• The transparency obligations under the EU AI Act have applied since 2 August 2026. Generated product images and text without meaningful editorial review may require labelling.

Three levels: assistance, workflow, agent

In PIM, the term AI covers very different capabilities. For this comparison, we distinguish three levels:

• Assistance: A feature performs a single task at the click of a button, such as writing product copy from attributes, translating it or tagging an image. A person initiates the task and reviews the result.
• AI in workflows: AI steps run automatically within imports, rules or approval processes, for example extracting attributes from supplier PDFs or triggering translations after every change.
• Agent: An agent is given a goal, plans the steps itself, reads and modifies data, and reports the result. The key question here is who approves changes.

What vendors deliver today

The table shows what vendors publicly document. “Announced” means there is a press release or release note, but no information on availability. Where an entry is missing, we found no public evidence. That does not mean the feature does not exist.

Information based on vendor websites, documentation and press releases as of 26 October 2026. Please verify in a proof of concept before making a decision.

Five findings from the analysis

1. Availability often remains unclear

Only SAP and Informatica explicitly describe individual agents as generally available. Syndigo mentions beta customers, while Struct refers to a preview. Most other vendors use “introduces” or “launches” in their press releases without specifying when the feature will be available or which package includes it. Stibo Systems asks users to register their interest in AgentWorkx. For your selection process, this means asking vendors to demonstrate what is running in the product today and ensuring the quote specifies which AI features are included.

2. MCP is becoming standard

The Model Context Protocol (MCP) enables AI assistants such as Claude, ChatGPT or Copilot to access product data. MCP servers are documented for Akeneo, inriver, Stibo Systems, Pimcore, Bluestone, Sales Layer, Viamedici and Informatica. Salsify, Struct, Centric and ATAMYA have announced one or offer a preview. What matters is whether the server has read-only or also write access, and what permissions it has. Akeneo recommends a dedicated API user with minimal permissions. For more background, read our article MCP and Agentic Commerce.

3. Approval processes vary considerably

Pimcore documents its approval process most clearly: every write action by an agent is proposed, reviewed and approved. Akeneo states that Agentic Ziggy changes nothing without consent. Bluestone asks before making any change. Other vendors, by contrast, promote workflows without manual intervention. For product data with legal implications, such as safety instructions, hazardous substance information or technical specifications, you should insist on an approval step.

4. Using your own AI model often costs extra

Many companies want to use an AI model that their IT department has already approved, for example through Azure OpenAI. This is possible with Pimcore, ATAMYA, Bluestone and Struct, and with Stibo Systems if you have your own Azure subscription. With Akeneo, using your own model is a paid add-on, while the standard model is included in the package. Clarify where your data is processed and whether it is used for training.

5. AI pricing is rarely made public

Plytix is the most transparent: every plan includes 500 AI credits per month, with one credit per text and 10 to 30 per image. Bluestone states that its subscription includes one million tokens per day for the agent, while Sales Layer provides the MCP server at no extra charge. SAP charges AI Units for the Catalog Optimization Agent. For the other vendors, AI is tied to the package or edition, and prices are not public. Ask for AI costs to be itemised separately in the quote, including a scenario in which you significantly scale up usage.

Benefits documented by customers

Public metrics linked to named customers are rare. Here is what we found:

Most other figures on the market come from vendor surveys or events without named customers. During the selection process, ask for reference customers using an AI feature in production and for details of their starting point.

What the EU AI Act means for AI in PIM

The transparency obligations under Article 50 of the AI Act have applied since 2 August 2026. The Digital Omnibus adopted in July 2026 (Regulation (EU) 2026/1744) did not postpone this deadline, although it did postpone the obligations for high-risk systems. Three points are particularly important for product data:

• Chatbots and product advisers must be identifiable as AI.
• Providers of generative AI, including PIM vendors with built-in text generation, must mark generated content in a machine-readable format. Systems already on the market before August 2026 have a transition period until 2 December 2026. Ask your vendor how it is implementing this.
• Generated product images that depict a product differently from how it actually looks may qualify as deepfakes and must then be labelled accordingly. For text, the question is whether a human takes editorial responsibility for it. Under the Commission’s guidelines, a purely formal review is not sufficient.

The obligations for high-risk systems under Annex III do not apply until 2 December 2027 and generally do not affect typical product data processes. Seek legal advice on the classification of your use case. This article does not constitute legal advice.

Questions you should ask every vendor

• Which AI features are generally available today, which are in beta or announced, and which are included in our package?
• Can agents change data, and what does the approval step look like?
• Is there an MCP server, does it have write access, and how are permissions and logging managed?
• Can we use our own AI model or Azure subscription, and what does that cost?
• How is AI content labelled under the AI Act, and is that labelling retained when content is syndicated to retailers?
• Which reference customer is using the feature in production, and what measurable benefits has it delivered?

Frequently asked questions

Which PIM systems have AI agents?

Multi-step agents with approval workflows are publicly documented for Akeneo (Agentic Ziggy), Pimcore (Agent Bundle) and Bluestone PIM (AI Agent). SAP offers a generally available Catalog Optimization Agent for SAP Commerce Cloud. Syndigo, inriver and Stibo Systems have announced agents, with Syndigo already working with initial beta customers.

What is an MCP server in PIM?

An MCP server makes product data accessible to AI assistants such as Claude, ChatGPT or Copilot via the open Model Context Protocol. Depending on their permissions, assistants can either read data only or also modify it. MCP servers are documented for Akeneo, inriver, Stibo Systems, Pimcore, Bluestone and Sales Layer, among others.

Can I use my own AI model in the PIM system?

Yes, several vendors support this, including Pimcore, ATAMYA, Bluestone and Struct, as well as Stibo Systems via your own Azure subscription. Akeneo offers support for your own model as a paid add-on. Clarify costs, data processing and training before signing a contract.

Does AI-generated product copy need to be labelled?

It depends on the specific case. The transparency obligations of the EU AI Act have applied since 2 August 2026. As providers, PIM vendors must label content in a machine-readable format. Whether you also need to label it depends, among other things, on whether a human takes editorial responsibility for the copy. Generated product images that depict the product differently may qualify as deepfakes. Seek legal advice on your specific case.

Sources

• Akeneo: AI Configurations
• Akeneo: Security and privacy of AI features
• Akeneo: Agentic Ziggy (July 2026)
• Akeneo: MCP Server
• Salsify: SalsifyIQ (May 2026)
• Digital Commerce 360: Salsify Intelligence Suite
• Salsify: Kerry Group and VitrA
• inriver: Inspire
• inriver: Agents and workflow orchestration (March 2026)
• inriver: MCP
• Stibo Systems: ProductGen AI
• Stibo Systems: MCP Server (June 2026)
• Stibo Systems: AgentWorkx (September 2026)
• Stibo Systems: Lyreco
• e-commerce.news: Syndigo SynapseGo
• Pimcore: Agent Bundle
• Pimcore: Copilot
• Viamedici: EPIM 5 Update 5
• Viamedici: MCP servers in PIM
• ATAMYA: AI in PIM
• ATAMYA: Features by edition
• Centric Software: Centric PXM
• Bluestone PIM: AI Agent
• Bluestone PIM: MCP Server
• Boyum: AI features in Perfion
• Sales Layer: MCP Server
• Plytix: Pricing and AI credits
• Quable: AI in PIM
• Struct: AI Assistant
• Struct: Roadmap 2026
• SAP: Business AI Release Highlights Q1 2026
• Salesforce: Informatica agents (June 2026)
• Akeneo: Synchro Diffusion
• Akeneo: Elemis
• Regulation (EU) 2026/1744 (Digital Omnibus on AI)
• European Commission: Guidelines on transparency obligations (July 2026)

As of: 26 October 2026

Have any questions? We are happy to advise you.

Want to know which AI features make sense for your product data and which vendors already offer them? Get in touch.