AI in Business: Convergence of ERP, Data, and Security

AI is moving from experimentation to becoming central to ERPs and CRMs. As use cases multiply, companies must simultaneously address application integration, data governance, and access control. Vendors like Microsoft and SAP are making progress, while the orchestration of business, IT, and security becomes essential for controlled deployment.

Deploying AI Requires Trade-offs and Coordination from the Start

Generalizing AI within a CRM, an ERP, and then management tools is not just about juxtaposing three projects. Business leaders must articulate the decisions or tasks to improve, while ERP and CRM teams verify processes and integration, data specialists prepare actionable information, and cybersecurity oversees access and actions. This coordination can begin as early as the choice of the ERP to anticipate future uses of BI and AI. Trade-offs are necessary regarding the validation of responses, tracking errors, testing changes before production, and defining performance indicators, and these responses must evolve with usage. Organizations that already connect their processes, explain their data, and control access have a stronger foundation, while initial projects elsewhere reveal inconsistencies in the information system.

Assistants and Agents: Increasing Capabilities According to Vendors

Management software is being equipped with AI for daily tasks, from summarizing customer interactions to anticipating supply chain disruptions, to agents capable of calling tools, triggering workflows, or preparing operations. Microsoft is developing these uses in Dynamics 365, and SAP is evolving Joule around business processes. However, the actual scope varies depending on versions, modules, and configurations, and the risk differs between a simple suggestion and an order actually placed by an agent. The challenge goes beyond adding an assistant to the interface to encompass the execution and orchestration of processes. In this area, TVH Consulting highlights a 360° approach that aggregates Microsoft and SAP ERP, CRM, Data, BI, and cybersecurity.

ERP-CRM Integration: Synchronizing Processes and Repositories

Confirming a delivery date via an assistant may require the CRM for exchanges and the ERP for the order status. Without synchronization, the response may be incorrect. Before expanding AI, it is essential to examine interfaces, common repositories, and management rules: which system holds authority for the billing address, how to reflect an ERP cancellation in the CRM, how to handle duplicates and status changes. The configuration of a validation process is as crucial as the technical connection, as poorly defined automation can propagate errors. Mapping flows, testing exception cases, and clarifying business responsibilities enhance the reliability and utility of AI functions.

Data Governance: Shared Definitions and Controlled Scopes

Teams may use different definitions of an active customer, a margin, or a delivered order, making reconciliations risky if AI ignores calculation rules. Data governance aims to harmonize definitions, ownership, and provenance, covering freshness, quality, and usage permissions. A database may be accessible without all its content being relevant for everyone. Anne Canu, director of the Customer Engagement BU at Microsoft for TVH Consulting, emphasizes that a successful AI project starts with the right data, needs, and objectives. However, it is not necessary to rebuild the entire heritage at once: a limited scope, identified sources, and defined indicators can serve as a starting point before gradual consolidation.

Security: Fine-Grained Rights, Least Privilege, and Human Safeguards

Connecting an assistant to management systems alters access paths and may expose sensitive data if it bypasses the authorizations of source applications. The risk increases when an agent can take action, hence the need to distinguish between reading, preparing, and validating operations. When an agent investigates a supplier delay, they should not change bank details or make a payment, and the principle of least privilege also applies to technical identities. It is necessary for cybersecurity teams to intervene from the framing phase to classify data, verify access rights, ensure logging, secure connections, and evaluate suppliers, while considering the possibility of malicious instructions hidden within content. Human validation remains required for sensitive or difficult-to-reverse actions.

From BI to AI: Enhanced Analyses but to be Closely Examined

Business Intelligence offers indicators and historical analyses; AI adds anomaly detection, forecasting, and natural language exploration. An industrial manager can compare material consumption variances and investigate factors, while a sales department can cross-reference CRM pipeline and ERP orders for its forecasts. These results require critical examination: correlation does not prove causation, and forecasts depend on historical data and assumptions. It is essential for users to access source data, identify key criteria, and correct interpretations. As multiple applications are mobilized, data quality, inter-system exchanges, and access security become central, bringing back to the forefront projects that organizations previously managed separately.