Title: SAP Commerce AI: agents, ChatGPT, Claude, and Gemini | CCI Source: https://www.commercecloudintegrations.com/ai Publisher: Commerce Cloud Integrations # SAP Commerce + AI We connect your SAP Commerce Cloud shop to ChatGPT, Claude, and Gemini, and to Copilot, Perplexity, and SAP Joule. Shoppers find your products in the assistants they already use. Your service, merchandising, and engineering teams work on live orders, stock, and code from Slack, Teams, and the IDE. *Animation: One SAP Commerce shop answering in seven assistants: ChatGPT: a shopper asks for a rain jacket and gets two Meridian products with live stock. Claude in Slack: order 48213 explained and refunded in 52 s. Gemini: the same catalog answers a comparison, with stock for the Vienna store. Copilot in Teams: PR #812 reviewed, tested, built, and on d1 in 3 h 40 min. Meta AI in WhatsApp: order 48213 tracked and sent to the office address. GitHub Copilot in VS Code: the N+1 price lookup on /c/jackets batched into one query. Joule in Backoffice: a slow checkout traced to a cronjob and fixed. p95 0.6 s.* ## AI services for SAP Commerce Cloud For shoppers ### Get found and bought in AI search - #### SEO for ChatGPT, Claude, and Gemini We make product and category pages readable to answer engines: robots rules for GPTBot, ClaudeBot, Google-Extended, and PerplexityBot, server-rendered price and stock, and llms.txt. Every month we run 50 buyer questions in ChatGPT, Claude, Gemini, Google AI Overviews, and Perplexity, and report which answers cite you. - robots.txt - llms.txt - SSR - citation report - #### Structured data and product feeds for assistants We generate Product, Offer, AggregateRating, and FAQPage JSON-LD from OCC on every PDP, and publish product feeds with live price and availability for ChatGPT shopping and Google Merchant Center. - JSON-LD - OCC v2 - Merchant Center - feed sync - #### Agentic checkout We let an assistant place an order on your shop: cart, delivery mode, and payment through OCC, with the Agentic Commerce Protocol for ChatGPT and Google's Agent Payments Protocol. Your promotions, tax, and fraud rules still run. - ACP - AP2 - OCC carts - Stripe, Adyen - #### Semantic and vector product search We add dense vector fields to your Solr 9 index next to the keyword fields, so "jacket for a wet week in Scotland" returns rain shells and not zero results. Boost rules, facets, and stock filters keep working. - Solr 9 - hybrid ranking - embeddings - zero-result rate - #### Shopping assistant on the storefront We build a chat assistant into Composable Storefront that searches the catalog, compares products, checks store stock, and adds to cart through OCC. It answers only from your catalog and hands over to a person in ASM. - Composable Storefront - OCC - ASM handover For service and merchandising ### Answers from live orders, not from a ticket queue - #### AI agents for customer service A rep asks Claude in Slack or Copilot in Teams about an order and gets the status, the cause, and the fix. Refunds, address changes, and job reruns wait for one approval and leave an audit entry. - Claude in Slack - Copilot in Teams - ASM - approvals - #### SAP Joule on Commerce Cloud We set up Joule for customer service and merchandising tasks in Backoffice, and connect it to the same tools as your other assistants, so every team works in its own window. - Joule - Backoffice - SAP BTP - #### Product content at catalog scale We generate descriptions, attributes, alt text, and translations into the Staged catalog. A merchandiser approves each batch in a Backoffice workflow before the catalog sync puts it online. - Staged to Online - workflows - translations - #### Order and customer messaging Shoppers ask where an order is on WhatsApp or on your site. The assistant reads the consignment and the carrier status from SAP Commerce and changes the delivery address before dispatch. - WhatsApp - consignments - carrier tracking For engineering ### Shorter cycles from story to d1 - #### AI-assisted SAP Commerce development We set up Claude Code, GitHub Copilot, and Cursor with rules for your extensions, items.xml, ImpEx, FlexibleSearch, and Spring configuration, so the generated code fits your platform and passes review. - Claude Code - GitHub Copilot - Cursor - project rules - #### Test generation and review gates Agents write junit tests for populators and services, and Playwright tests for checkout. The CCv2 pipeline runs them with static analysis that fails a pull request when a query runs inside a loop. - junit - Playwright - CCv2 pipeline - #### AI-assisted upgrades and migrations Agents scan the codebase for APIs removed in 2211-jdk21, rewrite javax to jakarta for Spring 6, and draft ImpEx mappings from legacy exports. People review every change and the reconciliation counts. - 2211-jdk21 - javax to jakarta - ImpEx For operations ### Problems found, explained, and closed - #### AI incident triage and monitoring When Dynatrace opens a problem, an agent reads the PurePaths, the CCv2 logs, and the cronjob history, names the cause, and proposes the fix in the on-call channel, with the runbook step attached. - Dynatrace - CCv2 logs - cronjobs - runbooks - #### Guardrails, security, and audit We map each agent tool to a Backoffice role, keep writes behind an approval, mask personal data before it leaves the platform, and log every call with the user, the tool, and the result. - role scopes - PII masking - audit log - GDPR ## What the assistants can read and change Every assistant gets the same tools. Reads run on demand. Writes wait for a person to approve them in the chat. | Data | Source in SAP Commerce | Access | | --------------------------- | ---------------------------------------------------------- | ----------------- | | Products, prices, stock | /occ/v2/{site}/products/search, stock per point of service | read | | Orders and consignments | OCC orders, ASM customer session | read | | Carts and checkout | OCC carts, delivery modes, payment | shopper confirms | | Refunds and address changes | Returns, order process, consignments | approval | | Cronjobs and Solr indexes | Cronjob history, indexer jobs | approval to rerun | | Product content | Staged catalog, Backoffice workflow | approval to sync | | Code and builds | Git repository, Cloud Portal builds | pull request | ## Fixed-scope ways to start 1. 2 weeks ### AI readiness review We map your SAP Commerce version, OCC exposure, data, and roles, and rank the AI use cases by value and risk. You get the plan and the first three backlog items. 2. 3 weeks ### Answer engine audit We run your 50 most valuable buyer questions in ChatGPT, Claude, Gemini, and Google AI Overviews, then fix robots, rendering, and structured data on the top routes. 3. 4 weeks ### Agent pilot for one team One assistant, one team, one channel: for example Claude in Slack for customer service, read-only on production and approvals for writes, with the audit log live. ## Questions about AI on SAP Commerce Can ChatGPT, Claude, and Gemini read our SAP Commerce catalog? Only what their crawlers can fetch and render. If product pages load price and stock after hydration, or robots.txt blocks GPTBot, ClaudeBot, and Google-Extended, the assistants cite someone else. We fix the crawl rules, server rendering, and structured data first, then measure citations every month. [SEO for ChatGPT and LLMs](https://www.commercecloudintegrations.com/services#llm) Does this work with SAP Joule? Yes. We set up Joule for Commerce Cloud tasks and give it the same tools as the other assistants, so a merchandiser in Backoffice and a rep in Slack see the same answer. [AI agents on SAP Commerce](https://www.commercecloudintegrations.com/services#agents) Is our customer data sent to OpenAI, Anthropic, or Google? Only the fields a tool returns, after masking. We use the business and enterprise tiers of each provider, which do not train on your data, choose EU regions where they exist, and log every call. [Security hardening for SAP Commerce](https://www.commercecloudintegrations.com/guides/security-hardening-for-sap-commerce) Which SAP Commerce versions do you support? SAP Commerce Cloud 2211 and 2211-jdk21, and on-premise 2205 and earlier releases that expose OCC under /occ/v2, which arrived in 2005. Older estates get the integration through a small extension. [Composable Storefront OCC contracts](https://www.commercecloudintegrations.com/guides/composable-storefront-occ-contracts) How do agents change data without risk? Every write tool maps to a Backoffice role and waits for a named person to approve it in the chat. The approval, the tool call, and the result go to the audit log. [AI agents on SAP Commerce](https://www.commercecloudintegrations.com/services#agents) Do we need Composable Storefront for AI search? No, but it helps. Accelerator pages can be made readable too. Composable Storefront with server-side rendering puts the product data in the first HTML response without extra work. [Composable Storefront SSR incidents](https://www.commercecloudintegrations.com/guides/composable-storefront-ssr-incidents)