Agentic AI in E-Commerce: How Xentral Is Using AI Agents Today

Agentic AI (AI systems that act autonomously) shifts the logic of e-commerce. Instead of guiding people to a product page, you have to make your data and processes machine-readable. According to Gartner, around 40% of all enterprise applications will include task-specific AI agents by 2026, up from under 5% in 2025. In this article, we walk through what agentic commerce means for your business, which benefits are realistic, and how to lay the technical foundation.

What is agentic AI in e-commerce? 

Picture your customer telling their voice assistant: "Order me everything for a vegan barbecue for six people, organic quality, budget of 80 euros." An AI agent understands the instruction, compares products across shops, checks availability, reads reviews, and completes the purchase on its own. No human clicks required.

That's agentic commerce. Agentic AI describes AI systems that plan, decide, and act independently toward a clear goal. In e-commerce, these agents take over the entire buying process. Xentral sees agentic commerce as the next stage of online retail, where shops need to be optimized not just for humans but for machines too.

Different from classic chatbots 

Classic chatbots (rule-based dialog systems) only react to predefined inputs. Agentic AI plans proactively, chains steps together, and acts across multiple systems. These systems are built on large language models (LLMs) and follow a loop of thinking, planning, and acting. For retailers on Xentral, that means the agent isn't a service feature anymore. It's the channel itself.

Why does this matter now? 

AI shopping has arrived in the German market. According to the BEVH, 19.3% of respondents have already actively sought product recommendations from AI applications (survey from 2025). At the same time, Gartner predicts that around 60% of brands will use agentic AI for direct customer interactions by 2028. What we see at Xentral: retailers without a clean data foundation today won't be found in AI channels tomorrow.

Why is agentic commerce already changing German online retail? 

Agentic commerce shifts three layers at once: the channel, the payment, and the data requirement. In 2025, OpenAI and Stripe released the Agentic Commerce Protocol (ACP) as an open standard. The protocol lets agents buy directly inside ChatGPT, starting with US sellers on Etsy and Shopify. For Xentral customers, the signal is clear: checkout is moving out of your own shop and into the agent channel. Your product and inventory data have to be consistently accurate everywhere.

New infrastructure: MCP as the standard 

Alongside ACP, the Model Context Protocol (MCP) is emerging as the technical bridge between agents and enterprise systems. Anthropic reported over 97 million monthly SDK downloads and more than 10,000 active public MCP servers in January 2026. Through MCP, agents read product data, inventory, and prices directly from the ERP. Xentral's API-first architecture delivers exactly that connectivity.

What does it mean for the German market? 

German online retail is growing again. The BEVH forecasts nominal revenue growth of 3.8% in goods trade for 2026. Competition on marketplaces and price comparison platforms is getting tougher. Retailers who build a clean single source of truth with Xentral can serve AI channels like ChatGPT or comparable agents on top, without duplicating their processes.

Benefits of agentic AI for retailers and customers 

The benefits often get oversold. At Xentral, we separate what's realistic today from what's still just a future vision.

What do retailers gain? 

1. Process efficiency: Many repetitive tasks like order entry, reordering, or return processing run automatically on rules. Xentral customers cut manual work in daily sales and order fulfillment this way.
2. New channels: Agent-ready product data opens your shop to AI interfaces like ChatGPT Instant Checkout.
3. Higher data quality: If you want to serve agents, you have to keep product information, inventory, and prices clean. Your shop, marketplaces, and marketing benefit from that too.
4. Scalable personalization: Agents learn user preferences and surface better-matching products, without you optimizing separately for each channel.

What do customers gain? 

Buying behavior and expectations are shifting on the customer side too. Xentral sees three clear patterns in audiences already using AI shopping.

1. Less effort per purchase: Complex orders are handled with a single instruction.
2. Better fit: Agents factor in budget, preferences, and past purchases.
3. Time back: Price comparisons, review reading, and availability checks are handled by AI.

A realistic take 

Agentic AI doesn't fix the underlying problems of bad master data. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, mainly because of unclear business cases and missing risk controls. That's why Xentral takes a pragmatic approach: operational automation first, agents layered on top.

Three steps to agent-ready e-commerce 

Step 1: Build a single source of truth 

AI agents are only as good as the data they get. Excel lists, outdated product fields, and inconsistent inventory are the most common blockers.

Modern inventory management software pulls all the relevant data into one place: products, customers, orders, and inventory. Everyone, human or AI agent, works from the same correct information.

Step 2: Automate operational processes 

In Xentral, automation runs on two layers in parallel. Rule-based processes like inventory syncs, document generation, and DATEV export are handled by the Xentral Flow Editor with deterministic if-then logic. Wherever context needs to be interpreted, AI agents in the Agent Hub take over.

What's live in the Agent Hub today:

• Incoming invoice agent reads invoices from emails and PDFs, recognizes items even without clean SKUs, and generates a booking suggestion automatically.
• Delivery inquiry agent pulls delivery status from the ERP and drafts a ready-to-send customer reply.
• Payment reconciliation agent matches cryptic or incomplete payment entries against open items and suggests the correct assignment.
• Product question agent answers customer inquiries based on product data, spec sheets, and website content. Handles images and attachments too.
• Order agent reads orders from emails, PDFs, and CSV files, and creates them in the ERP automatically.

A higher-level orchestrator understands incoming requests and routes them to the right sub-agent. If an email contains several different requests, the orchestrator identifies each topic and distributes them in parallel. Every action currently requires confirmation from a team member. Human-in-the-loop isn't an optional feature here, it's an architectural principle.

Related reading: We explain the difference between rule-based flows and AI agents, and when to use which, in the Xentral Flow Editor guide.

Step 3: Expose open interfaces

Agentic commerce runs on APIs. AI agents need to talk to your shop to pull product information and trigger purchases. A system built on modern, API-based architecture is the prerequisite.

Xentral offers open, integration-friendly infrastructure with a REST API, webhooks, and the Xentral Connect middleware. On top of that, Xentral is actively building its own MCP server, so you can plug in your own LLMs and connect external tools like Claude directly to the ERP, for example to query BI dashboards by chat or pull analyses directly from ERP data.

How much control do you keep over AI agents? 

Every retailer asks three questions before letting AI take over operational processes: Who's liable for mistakes? Where does my data live? And can the AI do things I didn't approve?

At Xentral, the answers are built into the architecture, not buried in the fine print.

The entire AI infrastructure runs on European servers and follows GDPR. Every customer gets their own data silo. It's architecturally impossible for one tenant's data to leak into another tenant's responses. The connection to external language models runs through secure APIs that prevent model providers from either storing your data or training on it.

All AI activity is traceable and auditable, no black boxes. AI permissions always mirror the general system permissions of the user in question: an agent can't do more than the person using it. And until a suggestion gets executed, a team member confirms it. No amount is debited automatically. No email goes out without approval.

Xentral uses a best-of-breed approach with several language models: Anthropic (Claude), Google (Gemini), and OpenAI (GPT-4o). For every ERP task, the best-fit model gets picked. Open-source and EU-based models are also in the pool.

The bottom line: agentic commerce isn't theory for Xentral customers anymore 

If you have a clean data foundation in your ERP today, you can use the Agent Hub right away. No implementation project, no IT department, no waiting on an enterprise rollout. The question isn't whether AI agents will take over operational processes. The question is whether your ERP has the data structure they can actually work on.