At SAP Sapphire 2026, a digital display on the show floor cycled through the names of AI agents built by a select group of SAP partners. DataXstream appeared prominently among them, a visible acknowledgment of work that had been months in the making. During partner day presentations, SAP executive Philip Herzig singled out DataXstream by name as an example of the kind of partnership driving SAP’s vision for the autonomous enterprise forward. For a company with deep roots in SAP SD, sales execution and order management, that recognition reflected the sustained technical commitment to building AI-native solutions within the SAP ecosystem at a time when the ecosystem itself is undergoing fundamental change.
SAP’s AI Platform: A Unified Foundation for Agents
To understand the significance of DataXstream’s recent AI work, it helps to understand where SAP is taking its platform. SAP made several announcements during Sapphire 2026. One significant announcement was the SAP Business AI Platform, a consolidation of SAP BTP, business AI capabilities, and cloud data services into a single, unified foundation. The strategic intent: by unifying disparate SAP domains under one platform, AI agents can traverse business processes end to end, pulling data from finance, supply chain, sales, and customer management without being constrained by domain boundaries. An agent handling a customer retention workflow, for example, can access order history, pricing, material availability, and promotional data in a coordinated sequence because the underlying platform connects those domains in a coherent data fabric.
Alongside platform unification, SAP has moved to adopt what it describes as a bring-your-own-AI strategy. The platform now supports integration with external AI systems under a governed framework, extending beyond SAP-native tooling to accommodate the AI investments that customers and partners have already made. For SAP’s partner ecosystem, this is a meaningful shift. Organizations with existing AI investments, proprietary models, or preferred large language model providers can connect those systems into SAP workflows without forfeiting the security and compliance standards that enterprise deployments require.
Hack2Build: A Multi-Agent Solution for Customer Retention
SAP Hack2Build 2026 was a global competition inviting development teams to design and build agentic workflows using Joule Studio and SAP BTP. Selected from a field of more than 100 competing teams, DataXstream’s team advanced to the build phase and delivered a working prototype of its Churn Risk Agent, earning second place in the final rankings among some of the largest names in the global SAP ecosystem.
John Kane, Senior AI Engineer at DataXstream, was careful to frame the Churn Risk Agent as a multi-agent workflow from the outset. The solution was designed as a coordinated sequence of specialized agents, each responsible for a discrete piece of the customer retention problem. The first agent applies a machine learning model built by DataXstream’s team to assess churn probability for each customer, drawing on order history and recent purchasing behavior to generate a risk score. Customers above a defined risk threshold are passed to a second agent, which evaluates their buying patterns and identifies materials likely to generate renewed engagement. A third agent checks current material availability so that recommendations are grounded in actual inventory. A fourth agent retrieves applicable promotional pricing for those materials. From there, the workflow creates a sales inquiry transaction in SAP and generates a pre-formatted email for the sales representative, summarizing the risk assessment, the proposed materials, and the available promotions. The only human action required is the decision to send that email. Every step upstream of that touchpoint is handled by agents operating in sequence, with no manual intervention needed to move between them.
The architecture enabling this workflow is an MCP server, a model context protocol implementation that functions as a generic, tool-agnostic layer between the agents and the underlying systems. Because the MCP server is front-end agnostic, the same tool set is accessible through Joule Studio, through OMS+, or through a third-party interface. That modularity means the solution is portable across presentation layers, a design choice with implications well beyond the Hack2Build context.
Race to Sapphire: Twenty Agents, Enterprise-Ready
Following Hack2Build, DataXstream received an invitation to participate in SAP’s Race to Sapphire, a separate challenge designed to identify partners building production-quality, enterprise-ready AI agents for the Sapphire 2026 showcase. SAP’s requirements were specific: each agent needed to operate across multiple tool sets, handling the kind of multi-step, context-sensitive workflows that characterize real business problems. Single-function tools did not meet the bar.
Megan Herring, Developer at DataXstream, participated in much of the Race to Sapphire build effort. The team constructed 20 distinct agents, each covering a different business domain, spanning order management, customer service, materials, pricing, and churn analysis. Some of those agents, including the Churn Risk Agent, carried over from the Hack2Build work. Others were built specifically to meet the competition’s requirements.
As Herring described it, the challenge involved more than engineering. Designing 20 agents with genuine business utility, each covering distinct functionality without overlapping tool sets, required sustained creative effort alongside the technical work. “SAP wanted enterprise-ready AI agents for real business problem solving,” Herring said. “They wanted agents that had access to a variety of different tools. So not only did we need 20 agents, but we needed agents that would cover multiple tool sets.” SAP reviewed and validated all 20 agents, confirming that each met the enterprise-ready standard the competition required.
DataXstream was among a limited group of partners selected to showcase its agents on the Sapphire show floor, with the IA team’s work displayed alongside SAP’s broader messaging about the autonomous enterprise and the direction of AI in business process execution.
The Agents in Practice
Two of the agents Herring demonstrated illustrate the practical range of what the team built. The Churn Risk Agent, takes a customer identifier as input and returns a churn probability score along with the key factors driving that score. In a live demonstration, the agent identified a customer as high risk, cited the number of days since her last order, evaluated that figure against her historical order cadence, and provided a reasoned assessment. The model underlying the prediction is DataXstream’s own, built to incorporate customer-specific purchasing patterns so that a 20-day gap is evaluated in the context of that customer’s history rather than against a static threshold.
The Order Assistant agent demonstrates a broader scope. During a demonstration, it successfully retrieved detailed sales order information, confirmed delivery status, and created a new order for a specified customer with specified materials, generating a linked transaction in SAP without the user navigating the system directly. Herring noted the agent’s visibility into its own process as one of its more useful characteristics: rather than presenting a loading indicator, it surfaces the tools it is calling in real time, giving the user a running account of where the query stands.
The Technology Horizon: Agent-to-Agent Coordination
Both Kane and Herring pointed to the same direction when discussing where agent architecture is heading. The current generation of agents, connected to tool servers through MCP, represents one tier of functionality. The tier taking shape now is what SAP and the broader AI industry are calling agent-to-agent, or A2A, coordination.
In an A2A model, a smart middleware server acts as an intelligent router between agents. The server is itself connected to a language model, which means it receives a request, determines which tools are available, calls those tools, evaluates the response, and decides whether the task has been completed or whether additional tools are needed. Kane described A2A as a fundamental change in how systems integrate with one another. What makes A2A significant is where the intelligence resides. Traditional integration requires both systems to agree on exact data structures in advance; a misaligned field is enough to break the exchange. An A2A server, connected to a language model, absorbs that complexity into the middleware itself: it receives a request, determines which tools to call, evaluates whether the response is complete, and manages the handoff without either system needing prior knowledge of the other’s schema. A precisely formatted API call with a rigid field structure gives way to an intelligent handoff between systems that can negotiate the parameters of a request dynamically.
For DataXstream, the practical implication is that the 20 agents built for Race to Sapphire are a foundation, not a ceiling. Herring described a direction toward fewer, more capable agents organized in hierarchies, where a primary agent coordinates secondary agents as the workflow requires. An Order Assistant agent, for example, could invoke a Churn Risk agent when order patterns suggest retention risk, without the user initiating that secondary query. The coordination would happen because the agents are structured to collaborate, each operating within its domain while contributing to a broader workflow.
From Competition to Commercial Availability
The Churn Risk Agent that DataXstream built for Hack2Build is now available in the SAP Store as an add-on to OMS+. The transition from competition prototype to commercial offering reflects the maturity the DataXstream team brought to the build phase. The solution entered the competition as a functioning, deployable workflow designed to address a genuine business problem that SAP customers in distribution and manufacturing face on an ongoing basis.
The availability of the Churn Risk Agent in the SAP Store also demonstrates the practical benefit of DataXstream’s architectural choices. Because the solution was built on SAP BTP using Joule Studio and Kyma runtime, with connectivity managed through an MCP server, it fits cleanly into the SAP ecosystem that customers already operate within. Bringing it to market as an OMS+ add-on means customers can extend the capabilities of a platform they already use rather than integrating a standalone tool.
DataXstream’s Position in the SAP AI Ecosystem
The work DataXstream’s team completed over the course of Hack2Build and Race to Sapphire reflects a considered approach to AI development in the SAP context. The solutions are grounded in real operational data, designed to meet SAP’s governance requirements, and built to extend the capabilities of existing customer environments. As SAP continues to evolve the Business AI Platform and expand the reach of Joule Studio through MCP and A2A integration, the partners positioned to help customers realize that value are the ones who have been building in that environment, under those constraints, with those standards, for some time.
DataXstream is one of those partners.
