The ERP bottleneck most CIOs overlook when discussing AI

When organizations evaluate AI opportunities, the focus is most often on Copilots, intelligent assistants and customer-facing automation. Yet some of the most practical opportunities can be found in the work surrounding ERP delivery.

Documentation and testing consume significant resources, create dependencies on key specialists, and become harder to maintain as systems evolve. They rarely receive much attention in strategic AI discussions, but they influence how quickly and safely an organization can upgrade, scale and change its ERP environment.

Recent developments in AI agents and Model Context Protocol, or MCP, could make these disciplines more efficient. For CIOs, the opportunity is not simply faster content generation. It is a delivery model that can connect AI to business applications, development environments and governed deployment processes.

The bottleneck hiding in plain sight 

During an ERP implementation, organizations spend considerable time documenting business processes, configurations and system decisions. Once the project is completed, priorities shift. New requirements emerge, teams move on and documentation gradually falls behind the live solution.

The consequences often appear later, during audits, upgrades, regulatory changes or personnel transitions. Missing configuration records and outdated process descriptions can increase costs, slow decisions and make it harder to adapt systems to new business requirements.

Testing presents a similar challenge. As ERP environments change, teams must maintain test cases, prepare test plans and coordinate regression and acceptance testing across IT and the business. Much of this work is repetitive, but it remains essential to operational stability.

Documentation and testing may not delay a project on day one. Over time, however, they can constrain the speed and safety of change.

Why the technology matters now 

Large Language Models (LLMs) can generate and analyze text, but generic language capabilities are not enough for reliable enterprise delivery. AI also needs access to relevant application data, operational context and instructions for interacting with systems correctly. 

This is where MCP becomes relevant. 

An MCP server provides a standardized connection-layer between an AI model and a business application. It can give an AI agent access to application data and functionality, information about how the application works and instructions governing how it should interact with the system. 

For a CIO, the significance is practical. Instead of responding only to generic prompts, an AI agent can work with the context of a specific business application and its processes. This creates a stronger foundation for generating configuration documentation, preparing test cases or supporting other delivery activities. 

The technology does not stand alone. A governed delivery framework can combine: 

• AI-assisted development in Visual Studio Code 

• AI agents and prompt-based development workflows 

• GitHub repositories, reusable components and shared standards 

• Azure DevOps backlogs and deployment pipelines 

• Security, compliance and governance controls 

Together, these elements connect experimentation with established development and deployment practices. The business value lies in moving faster without separating AI innovation from the controls applied to other enterprise solutions. 

Where AI can create practical value 

The five areas below offer an initial view of how this model could support ERP delivery: 

1. Training material creation

An AI agent can use process descriptions, solution configurations and Microsoft Learn content to produce structured training materials. This can reduce manual preparation and improve consistency across projects. 

2. Configuration documentation 

An agent connected to a live Microsoft Dynamics 365 environment can read configuration data and generate corresponding documentation. This helps organizations keep documentation aligned with the solution rather than relying on periodic manual updates. 

The value is not simply reduced effort. It is improved visibility, better governance and reduced dependence on institutional knowledge that is held by a small number of specialists. 

3. Test-case generation 

Using process descriptions and application knowledge, AI can prepare structured test cases for human review. Specialists remain responsible for confirming that the cases reflect business requirements, exceptions and risks.

4. Test plans and test suites 

Generated test cases can be organized into process-based plans and suites. This provides a more consistent testing framework while reducing administrative effort. 

5. Automated test execution 

AI-supported workflows can also execute predefined tests, document results and update test records. The potential value is lower effort in regression and acceptance testing while maintaining human oversight and validation. 

These use cases matter because they address work that ERP teams already do. AI does not create the need for documentation and testing, but it can make both easier to maintain as operational capabilities rather than one-off project deliverables.

Faster delivery still requires governance

AI-assisted development can shorten the path from idea to prototype, but speed introduces its own management questions.

A prototype is not the same as a production-ready service. Before an AI-supported solution enters an ERP delivery model, organizations still need defined ownership, security controls, validation criteria, deployment processes and ongoing monitoring. 

The same applies to the source material used by AI. If process repositories are outdated or configuration data is inaccurate, the output will reflect those weaknesses. 

So AI increases the importance of familiar disciplines: 

• Reliable documentation 

• Structured data and process information 

• Standardized delivery practices 

• Clear ownership 

• Human validation 

• Security and compliance controls 

Without this foundation, organizations may produce isolated prototypes rather than sustainable operational capabilities. 

Questions CIOs should ask 

Before investing, CIOs should ask: 

• Which recurring activities consume the most specialist time? 

• Where do documentation gaps create operational or compliance risk? 

• How much effort goes into preparing and maintaining test assets? 

• Which systems and data sources would an AI agent need to access? 

• Who owns the quality of the source information and generated output? 

• How will prototypes move through security, validation and deployment? 

• Where must human judgment remain mandatory? 

These questions connect the technology to the delivery model and help distinguish a useful operational capability from an attractive demonstration. 

Improve the disciplines ERP already depends on 

The strongest AI business case in ERP may not come from introducing an entirely new capability. It may come from improving the documentation and testing practices that determine how safely an organization can change its systems. 

AI agents, MCP, GitHub and Azure DevOps provide the technical foundation. Strong governance, reliable information and human expertise determine whether that foundation creates lasting value. 

For CIOs, that is the real opportunity: not simply producing documentation or tests faster, but making essential delivery disciplines more consistent, scalable and easier to govern. 

Learn more 

Interested in exploring how AI agents, MCP-based solutions and modern delivery practices can support ERP documentation and testing in your organization? Discuss your priorities and delivery model with a Cepheo expert.