When software places the order instead of people
Imagine a customer's purchasing system triggering an order automatically. No phone call, no email, no sales visit. An AI agent compares availability, checks terms and sends the order straight to your B2B shop. This is no longer a distant scenario.
Gartner forecasts that AI agents will broker around 15 trillion US dollars in B2B purchases by 2028. At the same time, open standards are emerging, such as the Agentic Commerce Protocol from OpenAI and Stripe and the Universal Commerce Protocol from Google. First pilot projects are already running at banks and payment providers. And according to recent surveys, roughly one third of buyers already use AI chatbots for supplier research.
For manufacturers and wholesalers in the mid-market this means: the other side is changing. Anyone who does not prepare their B2B shop and product data for automated procurement will simply not be found. This article answers five concrete questions you should clarify now.
Question 1: Can an AI agent find our product data at all?
AI agents work differently from human buyers. They do not browse PDF catalogues and they do not call hotlines. They access structured, machine-readable data: standardised product names, unique article numbers (GTIN, EAN), normalised attributes such as dimensions, weight and material, and classifications following standards such as eCl@ss or UNSPSC.
If your product data is incomplete in your PIM system (Product Information Management) or exists only as free text, an AI agent is simply unable to identify your offering reliably. The same applies to your B2B shop: missing or inconsistent product attributes mean that automated procurement systems overlook your articles.
Practical note for Shopware-based B2B shops: Shopware offers native support for structured product attributes and API interfaces. Check whether your product data can be retrieved completely via the REST API and whether you map classification standards such as eCl@ss in your data set.
Recommended action: Run a product data quality audit before the end of 2026. Define mandatory fields for all product categories and make sure your articles can be identified unambiguously through standardised identifiers.
Question 2: Are prices and availability machine-readable?
In classic B2B sales, individual prices often end up in quotation PDFs sent by email. For a human buyer that is manageable. For an AI agent it is a dead end.
Agentic commerce, meaning automated purchasing by AI systems, requires prices and stock levels to be available in real time via interfaces. Customer-specific terms, volume prices and framework contract prices do not have to be publicly visible. But they must be programmatically available once the agent has been authenticated successfully.
Availability is just as critical as pricing. An AI agent that receives no reliable delivery date will move on to a competitor who provides that information.
Practical note for Shopware-based B2B shops: Shopware supports customer-specific price groups and stock displays via the API. Check whether your ERP connection returns real-time availability or whether you work with static stock levels that may be outdated.
Recommended action: Review your ERP shop integration for the timeliness of stock and price data. Aim for synchronisation intervals of no more than 15 minutes. For critical product ranges, a real-time connection is advisable.
Question 3: Who is allowed to order on behalf of a customer company?
This is the question most often overlooked in the technical discussion. When an AI agent orders on behalf of a company, it must be clear whether that agent holds the authority to do so, and who granted it.
B2B purchasing involves approval processes, budget limits and ordering permissions. These structures exist for good reason. An AI agent that automatically triggers an order worth 50,000 euros without a human in the customer company approving it is a compliance risk for both sides.
Modern B2B platforms therefore have to map role and permission concepts that also apply to machine actors. That means: an agent authenticates with defined credentials, acts within predefined limits and automatically triggers an approval workflow when those limits are exceeded.
Practical note for Shopware-based B2B shops: The Shopware B2B Suite offers role concepts, budget limits and approval workflows natively. These functions were originally designed for human buyers, but they can be transferred to agent access provided the API authentication is configured accordingly.
Recommended action: Define together with your key accounts which ordering processes may run automatically in future and which limits apply. Document these agreements and map them technically in your shop.
Question 4: Which orders should deliberately stay with people?
Not everything that can be automated technically should be. That is not a romantic view of field sales, it is a strategic decision.
Custom manufacturing, complex configurations, framework contract negotiations, consultation-intensive first orders: these processes require human judgement, contextual understanding and often trust that has grown over years. An AI agent can handle standard orders efficiently. It cannot build a relationship.
The question manufacturers and wholesalers now have to answer is: where is automation a gain, and where is it a risk to the customer relationship? Drawing that line is not a technical exercise, it is a sales strategy decision.
Practical note for Shopware-based B2B shops: Flag product categories or order types that require a human contact and route those enquiries specifically to your customer service or field sales team instead of pushing them into an automated checkout process.
Recommended action: Create an internal classification of your order types: what should run fully automatically? What requires approval? What stays with people on principle? This classification is the basis for your platform architecture over the next two years.
Question 5: How do we measure whether revenue comes through AI channels?
When an AI agent orders, it looks like a normal API order in your shop backend unless you make a distinction. That means you have no transparency about which share of your revenue is generated through automated procurement systems.
This is a measurement problem with strategic consequences. Without that data you cannot judge whether your investments in machine-readable product data and API infrastructure are working. Nor can you identify which customers have already moved to automated procurement and which product ranges run particularly often through AI channels.
Practical note for Shopware-based B2B shops: Implement a marker for API-based orders and differentiate between human users ordering via the API and automated systems. This can be mapped through specific API keys or user agent strings in the order source.
Recommended action: Add the dimension "order source" to your reporting. Define KPIs for the share of automated orders and track their development on a quarterly basis. This metric will grow in importance over the coming years.
What this means for your platform strategy
The five questions above are not isolated technical problems. They are symptoms of a deeper shift: B2B platforms are increasingly no longer built primarily for human users, but for hybrid environments in which people and machines act in parallel.
For manufacturers and wholesalers planning or developing their platform architecture today, this means in concrete terms:
Data sovereignty gains importance: who controls which data an AI agent may retrieve?
API quality becomes a competitive feature: clean, documented interfaces are no longer an IT topic, they are a sales topic.
Hybrid models are the realistic path: not everything will be automated, but everything has to be capable of automation.
Scalability decides: systems that handle ten manual orders per day today must process thousands of API calls tomorrow.
The good news: anyone who lays the groundwork now, meaning structured product data, clean API connections and clear permission concepts, is positioned considerably better for agentic commerce than competitors who continue to rely on PDF quotations and fax orders.
Frequently asked questions
How do I prepare my B2B shop for AI-driven procurement?
The first step is an audit of product data quality: are all articles maintained with standardised attributes and classifications such as eCl@ss? Prices and stock levels should then be available in real time through a clean API. In addition, you need a permission concept that defines which systems may perform which actions.
Which order types should not be automated?
Custom manufacturing, complex configurations, framework contract negotiations and consultation-intensive first orders should deliberately stay with people. These processes require contextual understanding and trust that AI agents cannot replace. A clear internal classification of order types is the basis for a sensible automation strategy.
How do I measure which revenue is generated through AI channels?
Marking the order source in the shop backend makes it possible to distinguish whether an order comes from a human user or an automated system. Specific API keys or user agent strings enable this differentiation. Add the order source dimension to your reporting and track its development as a standalone KPI.
