The News
On October 7, SAP’s customer experience organization held its own SAP Connect keynote to put products and substance behind what it calls Autonomous CX.
CMO Jessica Keehn opened with a survey SAP conducted among 4,800 customers, CX practitioners and CX decision makers. In the survey, comparing price and value is customers’ top priority for AI in buying but ranks seventh in business investment. More than nine in ten CX leaders surveyed said they were prepared for agentic buying, while 22% said they could fully connect customer tools to the operational systems behind them and 3% reported a true end-to-end connection. One of her opening messages was that businesses need to invest to deliver on their customers’ priorities to be successful. Hold that thought.
CX President and Chief Product Officer Balaji Balasubramanian described Autonomous CX as built “the other way around”; starting from enterprise context, based on the Business AI Platform, then assistants and agents on top. “If you only provide data from your CRM systems and AI does not know your orders, your inventory, your subscriptions, and your financials, anything that AI does is practically a guess.”
The announcements, detailed in SAP’s Innovation News Guide range from new agents for the Content and Campaign Assistants, new assistants for quoting and incentive compensation, to B2B support in SAP Order Management Services, all planned for general availability in Q1 2027. On stage, SAP added B2B and account-based marketing in Engagement Cloud and general availability of Commerce Cloud ERP edition for midsize companies.
A live demo took a fictional brand, Apex Athletics, from a supplier PDF to product catalog, campaign, sales visit, quote and S/4HANA order, then through a quality problem that a case management assistant resolved with alternate stock, a replacement order and a customer email drawn from S/4HANA records.
CX Chief Revenue Officer Mark Niemiec then presented Ericsson. Quotes for its network equipment had taken up to 14 days, with process variation across more than 175 countries. Ericsson standardized commerce, pricing and ordering on SAP and built a Quote Creation Agent that sellers address by email from Outlook. Global Product Owner Aaron Thomas said Ericsson started with quoting as “something that is basic and pragmatic” and will test SAP’s order management next. Keehn closed: “Pick one place to start, commit to it”, closing the loop to the start of the keynote.
The Bigger Picture
The Enterprise Titans agree that a generic model alone does offer differentiation. Context does. They differ on where this context, and therefore the differentiator, lives.
Salesforce builds context around the customer record. Data 360, MuleSoft and Informatica supply metadata and master data, and AIforce exposes Salesforce data and permissions to Claude, Slack and other interfaces. With Koa, Salesforce added a CRM reasoning model post-trained from NVIDIA Nemotron on synthetic data, which it says matches or exceeds leading models on its CRMBench benchmark.
Microsoft builds context from work. Microsoft IQ combines Work IQ from Microsoft 365 activity with Fabric IQ, a semantic layer over Fabric data, and Foundry IQ for retrieval. Dynamics 365 is one source among many.
ServiceNow builds context from operations. Its Context Engine maps people, roles, assets, services and policies on top of the CMDB.
Oracle builds context in the database, with AI Database 26ai serving the data Fusion Applications run on.
Adobe builds context from profiles and brand: Adobe Experience Platform is the contextual layer of CX Enterprise.
Zoho builds context by owning the whole stack. More than 55 applications share one platform and one permission model, so agents built in Zia Agent Studio draw on more than 700 actions across Zoho products and run as digital employees within the organization’s existing access rights. Its own Zia LLM, three models, trained in-house for business tasks, run in Zoho’s data centers in the U.S., India and Europe; Zoho says its generic models are not trained on customer data. Its MCP server opens data and actions to outside agents under the same permissions.
SAP builds context from the transaction. Its knowledge graph spans more than seven million data fields, half a million tables, 50,000 APIs and 400 data products. Business Data Cloud, Reltio and Dremio extend this to non-SAP data; RPT and Prior Labs’ tabular AI supply predictions trained on business data. In the executive Q&A, CEO Christian Klein called APIs and MCP servers “table stakes” and located the intelligence in the ontology.
The second divide is how vendors measure progress. Agent counts remain a headline number; SAP reported more than 200 generally available agents since Sapphire, with more than 400 targeted by year end while Salesforce boasts more than 18,000 agents built by Salesforce partners. In the Q&A, Klein qualified this figure: counting agents is “important to increase the scope” of SAP’s agentic layer, “but more important are the A2A use cases,” agents working across processes, such as matching customer demand with supply.
My Analysis and Point of View
After the main keynote, I reiterated my stance that SAP treats CX as its Cinderella. Th CX keynote showcased why this needs to change. SAP CX is absolutely competitive, which is not seen by a wide enough audience.
SAP CX delivered the best keynote I have attended in quite some time. It was aimed at the right audience, with the right message. It was sharp, concise, relevant and delivered well. It spoke to the people who matter in the enterprise, the users. The marketer, the account manager and the service manager, the people who use these systems on a day-by-day basis, and left the executive as the punchline: “Boom, you’re welcome, CRO.” It sums up the pitch: make the user’s day work, then the revenue follows. SAP’s own customer research makes the same point from the buyer’s side, though its percentages should be treated as directional.
Two strong use cases. Apex Athletics is fictional, but it is a rare CX demo in which marketing, sales and service hand off through the order, inventory and quality inspection, not a CRM record. Ericsson is the even stronger proof because it is real, global and unglamorous. Note the sequence: Ericsson standardized on SAP first and built the agent second. “No integration project, no connectors, no middleware” is true because that project already happened.
Start with one problem. This is the best advice that came from the customer. Ericsson did not automate lead to cash. It picked the step touching the most people in the most markets, that had high friction. It built a business case, listened to sellers, and only then planned the next agents. That is how autonomy is earned, and it is the opposite of the agent-count race that Klein rightly played down.
SAP’s CX strategy is far stronger than its main keynote suggested. In the Q&A, the outline is coherent: CX as the front end of order-to-cash, agents built where the transaction lives, partners such as Parloa where it does not, and third-party agents calling SAP through its gateway instead of writing into it. Klein’s remark that CX deals have moved from availability-to-promise checklists to agent outcomes, “now we play another game,” belongs on the main stage, not only the CX one.
The direction of AI plays to SAP’s strengths. Frontier models converge and get cheaper; differentiation moves to context and domain models. Koa is validation from the CRM market leader that general-purpose models are not enough for its domain although Salesforce’s research does not show that it is significantly better, if at all. SAP makes the same bet with RPT and Prior Labs, and it holds a deeper context graph: the order, the inventory, the price, the contract and the ledger, plus soon workforce intelligence. A CRM-trained model reasons well about opportunities, but it still has to ask somebody whether the shoes are in stock. Every CX agent that keeps a promise ends in a back office, and in large enterprises that back office is often SAP.
The bottom line. SAP’s CX strategy is considerably better than the main keynote coverage suggests: coherent direction, a context advantage competitors now validate, and a credible reference. Ericsson’s result presumes an SAP-standardized landscape that many prospects with Salesforce or Microsoft in the front office lack, and most of the new capabilities arrive in Q1 2027. Still, buyers should not discount SAP CX because its top executives do not position it well; judge it on its merits, on your data. Cinderella has her own ball now; she is ready to dance. It just needs the fairy godmother to make her shine in a main keynote.
What buyers should do now:
1. Pick your Ericsson step. Find the CX step with the most people, markets and friction, often quoting, order status or returns, and pilot it on your data first.
2. Audit your context before your agents. Map which systems hold the truth on price, availability, contracts and orders. If it is SAP, its agents start with an edge. If not, budget the harmonization before comparing agent demos.
3. Ask every CX vendor where the agent gets its answer and where it writes it. Demand the call direction, audit trail and a production reference at your scale, and test domain models such as Koa and RPT on your tasks, not on vendor benchmarks
Comments
Post a Comment