How a Global Manufacturer Cut Reporting by 10+ Hours a Week

The business ran on several systems that didn't talk to each other. Operational data lived in Visibility ERP, customer and pipeline data in the CRM, and much of the project context was spread across Outlook and SharePoint. All reporting ran through SQL reports. That setup held up for standard reports but struggled with anything else. Analysts couldn't easily segment the data, isolate outliers, or combine information from different systems into one view. With more than 20 entities to report across, the team spent hours every week just producing reports. The company was also looking at adding two more business analysts to keep up.

Visibility built an AI data layer that connects all four systems and lets analysts work with them in plain English. It has three parts: An MCP server that gives Claude secure, live access to the Visibility ERP database, the CRM, Outlook and SharePoint. A semantic layer that maps the business's terms (entities, projects, customers, orders) to the underlying data, so a question in business language returns the right numbers. A set of custom Claude skills built for the customer's reporting, dashboarding and forecasting workflows, so results are consistent and repeatable. Most of the two months went into testing. The two teams worked back and forth to check that answers matched the source systems before anyone relied on them for decisions.

Visibility started with the customer's analysts, cataloguing which SQL reports they relied on and which questions they couldn't answer, and scoped the first build to flexible executive dashboards and project forecasting. It then built an MCP server giving Claude live access to the ERP, CRM, Outlook and SharePoint, so nothing had to be copied into a warehouse. A semantic layer defined entities, projects, customers and orders once for all 20+ entities, and custom Claude skills packaged reporting, dashboarding and critical-path forecasting so any analyst gets consistent results. Testing took most of the two months, with each round of checks against source systems refining the definitions and skills.

This approach suits manufacturers and project-driven organizations that have plenty of data but can't get answers out of it fast enough. Industry and workflow Engineer-to-order and make-to-order manufacturers, where every job is different and success depends on project timelines, critical paths and margins. Other project-based businesses (industrial services, fabrication, capital equipment) where forecasting project risk matters as much as reporting past results. Scale and structure Mid-sized to large companies, especially those with several entities, sites or regions that need both consolidated and segmented reporting. Organizations where leadership regularly asks for new views of the data, not just the same monthly reports. Conditions that make it work Data spread across systems. An ERP plus a CRM, email and document stores, with no easy way to combine them. A reporting bottleneck. Reports depend on SQL or a small group of technical staff, and the backlog is growing faster than the team. An existing analyst team. They know the business and can help define terms and check results. This approach makes analysts more productive rather than replacing them. Structured core data. An ERP database the AI layer can connect to directly. Visibility ERP customers are the quickest fit, because the data model is already well understood. Willingness to invest in validation. The customer has to agree on business definitions and test answers against trusted reports before rolling the system out.