AI gets most of the attention in current technology conversations, but the foundation underneath it gets less attention than it deserves. AI systems are only as good as the data foundations they are built on. The organisations that have produced reliable, valuable AI capability have done so on top of strong data management. The organisations that have struggled to translate AI investment into business value have usually been working with weaker data foundations than they realised.
This piece walks through the data management foundations that modern AI actually needs, why each matters, and the patterns that distinguish organisations that have built strong foundations from those whose foundations are weaker than their AI ambitions require. It is written for technical and business leaders thinking about their own data and AI strategy.
Master Data as the First Foundation
Master data management is the first foundation that AI work depends on. Customer records that are duplicated across systems, product hierarchies that are inconsistent, account structures that differ between operational and analytical platforms all create the kind of data ambiguity that AI cannot resolve through model sophistication alone. The model trained on data with master data issues inherits those issues and produces outputs that reflect them.
Per Gartner – Data Management, the discipline of master data management has matured substantially over the past decade, with established patterns for resolving the kinds of issues that affect AI work. Organisations that invest in master data work before launching AI projects produce more reliable AI outputs than organisations that skip this foundation and try to address master data issues after the fact.
Data Quality as Ongoing Discipline
Data quality is the second foundation. AI work surfaces data quality issues that operational use of the same data may not have surfaced, because AI is more sensitive to certain kinds of inconsistency than human users are. Patterns of missing values, inconsistent formatting, drift over time, and structural changes in how data gets captured all matter for AI in ways that may not have mattered for the operational uses the data was originally collected for.
Strong data quality discipline is not a one-time clean-up. It is an ongoing practice that includes monitoring, alerting, and remediation as part of normal operations. Organisations that treat data quality as something to fix once and then forget tend to find that the issues come back as new data flows in. The work of AI Data Management includes this ongoing discipline rather than treating data quality as a one-time engagement.
Data Lineage and Governance
The third foundation is data lineage and governance. Knowing where data came from, what transformations it went through, and what versions of data fed into what models matters for several reasons. Reproducibility requires lineage. Compliance and audit responses require lineage. Debugging when models behave unexpectedly often requires lineage. Organisations without strong lineage capability have to do more work each time these needs come up.
Governance frameworks formalise who has authority over which data, how data classification works, and what controls apply to different kinds of data use. AI work often involves data that is sensitive in ways the operational uses had not surfaced, including personal information, commercial sensitivity, and regulatory considerations. Governance frameworks that anticipate these considerations support AI work better than ad hoc approaches.
Storage and Access Patterns
Modern AI work has specific requirements for how data is stored and accessed. Training requires being able to read large volumes efficiently. Inference requires being able to access specific records quickly. Feature stores have emerged as a specific pattern for serving the data needs of AI systems. Data warehouses, data lakes, and lakehouse architectures each support different patterns of AI work, and the right choice depends on the specifics.
Organisations that have built modern data infrastructure can support AI work straightforwardly. Organisations whose data infrastructure is older or more fragmented sometimes need to invest in modernisation before AI work can run efficiently. The infrastructure decisions are foundational and affect every AI project the organisation runs, which makes the investment in getting them right particularly worthwhile.
Connection to Other Modernisation
Data management foundations connect to broader modernisation efforts in ways that organisations should plan for explicitly. ERP modernisation often involves data work that overlaps with what AI requires. UI modernisation, including work like that done by Sprinterra on Acumatica platforms, often surfaces data structure decisions that affect what is possible analytically. The cross-cutting nature of data work means that organisations modernising in multiple directions can sometimes accomplish foundational data work as part of those efforts rather than as separate projects.
Recognising these connections is part of the strategic conversation around modernisation. Organisations that plan modernisation efforts to share foundational data work where possible spend less than organisations that approach each modernisation effort in isolation and rebuild similar data foundations multiple times.
Building Foundations Incrementally
Strong data foundations rarely get built all at once. The pattern in organisations that have ended up with strong foundations is usually incremental investment over time, with each project adding capability that subsequent projects can build on. The first AI project might invest in master data work for the specific domain it operates in. The second might extend that work and add lineage capability. The third might build on what the first two established, adding less foundation work to its own scope because foundations from earlier work are already in place.
This incremental pattern produces strong foundations more reliably than approaches that try to build comprehensive foundations as a separate effort before any AI work begins. It also distributes the cost across multiple project budgets in ways that are easier to justify than one large foundation project. Organisations that adopt this incremental approach tend to end up with the foundations they need for serious AI capability, while organisations that defer foundation work entirely tend to remain at the level of pilot projects that never quite become production capability.
Where Most Organisations Should Start
For organisations early in this journey, the practical starting point is usually master data and quality work in the domain where the first AI project will operate. This produces immediate value for the AI project itself and creates a foundation that subsequent work can build on. Other foundation areas, including lineage, governance, and infrastructure modernisation, can come into focus as the AI portfolio expands.
Organisations further along should be evaluating which foundation areas are now the binding constraint on what they can do next. The constraints shift as foundation maturity grows. The early constraint is often master data. As that strengthens, lineage and governance become the limiting factor. As those mature, infrastructure and access patterns can become the next set of investments. The progression is gradual, and organisations that stay attentive to where the current constraints are tend to invest in the right areas at the right times rather than building capability that is not the bottleneck.