SAP and Google Cloud are building an agentic commerce architecture that links search, marketing, inventory, payment and post-sale service. The aim is to let consumers purchase through Gemini or Google Search while retailers continue using SAP systems behind the interaction.
The partnership addresses a familiar failure: an intelligent front end recommends a product that the warehouse cannot fulfil because customer, campaign and inventory data remain fragmented.
A common protocol lets agents enter the commerce stack
SAP Commerce Cloud is adopting the Universal Commerce Protocol to standardise exchange among merchants, payment services and agents. A consistent interface can reduce the custom work required for each new AI channel.
SAP and Google plan to make merchant products discoverable in Gemini and Google Search AI experiences. The consumer expresses intent conversationally while backend services confirm inventory, build the cart and process the transaction.
SAP’s Gemini-powered Shopping Assistant can operate through chat, voice and text while maintaining state across the journey. Live warehouse and marketing data constrain recommendations so the agent does not present an attractive but impossible bundle.
Retailers need clear control over price, availability, brand rules and consent. An external agent should receive only the data required for the transaction, and customers should know which company is responsible when the conversation crosses platforms.
Bidirectional data makes recommendations operationally real
SAP Business Data Cloud Connect links with Google BigQuery through governed, zero-copy access. Instead of duplicating large datasets, the systems can combine weather, location and advertising signals with consented profiles, transactions and service history.
Zero-copy architecture can reduce latency and storage, but it does not remove privacy obligations. Administrative policy must govern who can query which fields, for what purpose and how derived audiences are retained.
SAP Engagement Cloud uses the combined context to coordinate specialised agents for segmentation, content and delivery. Gemini models can generate localised messages and imagery, while interaction data influences later communication.
The feedback loop needs boundaries. Continuous optimisation can become relentless targeting if frequency, sensitivity and customer preference are not enforced outside the model.
Human marketers should set objectives, approved claims and escalation rules. They remain accountable for campaigns even when agents choose timing and variation.
Retailers must protect ownership inside third-party interfaces
Agentic commerce may complete a transaction without the customer visiting the retailer’s site. SAP says consented engagement and purchase history can still return to the merchant’s customer systems.
That continuity preserves service and loyalty, but the platform controlling discovery may gain substantial influence over which products appear. Retailers need transparent ranking, portable data and alternatives if channel terms change.
Measurement also changes. Page views become less informative than agent queries, accepted recommendations, fulfilment accuracy, returns and service resolution.
Security testing should include prompt injection in product data, manipulated customer context and agents attempting unauthorised discounts or actions. Payment and refund limits need deterministic enforcement.
Merchants should also rehearse disagreement between systems. If BigQuery, the warehouse record and a payment service report different states, the assistant must pause rather than choose the most convenient answer. A customer-facing explanation and rapid human resolution protect trust when real-time architecture encounters an ordinary data mismatch.
The architecture’s promise lies in connecting intelligence to the operational truth of commerce. A conversational suggestion has value only when stock, price and delivery can support it.
SAP and Google Cloud are trying to make agentic retail compatible with existing enterprise investment rather than requiring a new storefront. If governance travels across the same connections as data, retailers can reach customers through AI-native channels while retaining the records and responsibility needed to fulfil the promise after the agent completes the sale.
