Your existing data estate may already support valuable AI applications. The challenge is knowing whether the information a particular use case needs is current, accessible, reliable and properly controlled.
A housing provider has a dashboard showing outstanding repairs, contractor performance and changes in demand. The reporting works. Managers can see where workloads are growing.
Now a service team wants an AI assistant to answer a tenant’s question:
When will my repair be completed?
The assistant needs the latest appointment, the repair history, notes from the contractor and perhaps a change recorded this morning. That information may be held across a repairs system, correspondence and external contractor updates. Some records may be restricted.
The dashboard can be correct without containing everything the assistant needs to answer that individual question. A production AI application has to retrieve the relevant information, apply the right permissions and handle gaps without inventing an answer.
The same data estate, different requirements
Reporting generally relies on defined measures, agreed business rules and established refresh schedules. An AI assistant answering an operational question may need record-level information that has changed since the last report refresh.
The difference becomes more important when an application can act. An assistant drafting a response for a colleague to review has different requirements from an agent authorised to update a repair record or send an appointment confirmation.
The data work should follow the task. Some use cases can rely on a well-governed existing analytical platform. Others require direct integration with operational systems, additional data-quality work or more frequent updates.
What the research says
Gartner surveyed data management leaders in 2024 on whether their organisations had the data management practices needed to support AI.
Gartner also made a separate prediction: that through 2026, organisations would abandon 60% of AI projects unsupported by AI-ready data. That is a forecast for a specific group of projects, not a measured failure rate for AI projects generally.
The practical concern is familiar. A pilot can succeed with carefully selected examples. In production, records change, sources disagree, access rights differ and staff expect the answer to reflect the current position. The application needs a way to deal with those conditions.
Four checks before an AI application uses your data
For any particular use case, these four questions show whether the data is ready, and what work remains.
In the repairs example, the assistant should say so, or route the enquiry to a person, if the appointment cannot be confirmed. It should not guess.
Some of this work will already be done. If a well-governed platform supplies current, permissioned data for the use case, the assistant can use it. The point is to check, not to assume that either answer is true.
What we’ve found in customer data estates
Servent’s Data & Analytics work includes assessing fragmented estates and designing the changes needed to make their information easier to manage and use.
Two types of engagement are relevant here. One establishes the condition of an estate and what should change. The other delivers the Azure data platform. Both matter when an AI use case depends on information spread across existing systems.
Start with the use case, then decide what to change
An organisation does not have to assess or rebuild every dataset before starting AI.
Take one worthwhile use case. List the information required, where it comes from, who owns it, how current it must be and who is allowed to access it. Test whether the existing data and application environment can supply that information reliably.
The findings may justify a focused integration, a change to access controls, better data-quality checks or a broader Data & Analytics programme. The amount of work should follow the evidence.
Where use-case priorities and wider readiness are still unclear, Servent’s AI Launchpad provides a structured, paid assessment. Where a specific data constraint is already known, a focused Data & Analytics discovery can define the necessary changes. Once the requirements are understood, AI Platform provides the governed environment for developing and operating the AI workload.
For your highest-priority AI use case, can you name the authoritative data sources, their owners, the required freshness and the permissions that must apply?
Gateway to AI is an eight-part Servent series examining what organisations need to develop, deploy and operate enterprise AI.
Research: Gartner, Lack of AI-Ready Data Puts AI Projects at Risk, 26 February 2025.
Previous in the series: Gateway to AI: what it takes to get AI into production
More from the blog:
