Databases Are Evolving in the AI Era… With Huge Implications For ERP

Observations

Table 1 lists major database products that have added foundational features (new ways of indexing or structuring data) that support AI. With these new features, the information currently stored in these databases can be accessed immediately through techniques that would previously require data to be replicated into a specific AI database. They effectively enable AI-like capability over existing data and enterprise applications.

Over the next two to three years, all core business applications (ERP systems (ERPS), customer relationship management systems (CRMs), human capital management systems (HCM), etc.) will likely enable such AI capabilities in the databases that drive their products, allowing them to rapidly build AI services natively into their solutions. Such capabilities will quickly move from a competitive advantage to a must-have.

This will be a major driver of what IBRS calls invisible AI, in which the vast majority of AI reasoning, information collection, and summarisation, as well as decision support, will happen automatically and natively within the core business solutions’ processes.

The current excitement over user-facing Agentic AI will quickly fade as more agentic tasks will be integrated into the core of business solutions, powered by their AI-evolved databases. Vendors will also experiment with new ways for staff to interact with business solutions, as detailed in the IBRS paper, ‘The Future of End-User Computing: The Biggest Change Since the Mouse is Upon Us’.

What Could Slow This Change?

While the new AI-evolved database technology will be available, the actual adoption rate of the features it enables may be slow to take off. Significant challenges exist in terms of migrating from legacy core systems what is likely to be a Software-as-a-Service (SaaS) solution. However, a lack of skilled personnel to introduce, manage and help staff adopt new ways of working will be a significant barrier. Change management relating to ERP migrations (or even upgrades) is a complex and failure-prone activity, even without fundamentally new ways of working with the core solution!

Adopting the new invisible AI capabilities may also be complicated by the need for highly specialised AI-enabled tasks, such as advanced machine learning (ML) model training for specific organisations and complex natural language processing to retain brand. These specialised solutions will still require dedicated AI platforms and services outside the core ERP database. Unfortunately, like ERP customisations of the last decade impacting migrations today, these custom AI services are likely to become the next decade’s legacy challenge.

Table 1: State of AI-Evolved Database – 2025

This table lists traditional databases that now support vectors and/or graphs. It excludes databases that are solely aimed at AI solutions, such as Pinecone (vector) or Neo4j (graph).

Explanation of AI Features:

• Vector Indexes: these are specialised indexes designed to store and search high-dimensional vector embeddings efficiently. Vector embeddings are numerical representations of data, such as text, images, or audio, that capture semantic meaning, enabling similarity search and other AI/ML workloads.
• Graph Indexes: while not always referred to as graph indexes in the same way as vector indexes, traditional databases can support graph capabilities through various means. This can include native graph data types, extensions, or integrated graph engines that allow for the efficient storage, querying, and traversal of relationships between data entities.

Next Steps

• Engage with your vendor(s) to explore the roadmap and the likely added costs associated with leveraging an AI-evolved database. If and when they adopt such capabilities, they should prioritise the processes they plan to transform with this invisible AI. If feasible, work with the vendor to identify and prioritise such processes.
• As detailed in the IBRS paper, ‘Generative AI in the Enterprise: Buy, Build or Platform?’, do not rush towards creating custom AI agents that are likely to become standard offerings within the organisation’s core business solutions in the next three years.
• If the core solution is provided as SaaS, expect vendors to increase either licensing fees to cover their increased costs for indexing information in multiple forms. Vendors may also add consumption costs to cover the execution of AI capabilities (which Salesforce has already done with its Atlas reasoning agentic AI solution).
• If your core solution is self-hosted, expect to need increased infrastructure to support the new capabilities provided by AI-evolved databases.
• If digital sovereignty is a requirement, confirm with vendors exactly where the vectors or graphs of their databases will be located, where vectorisation will be processed, and where search computing will be located. In short, it is not enough to have your data stored in a local database; it will also be essential to understand where and how that database leverages the new AI structures.

Footnotes

1. pgvector is an extension for Postgres that enables efficient storage and similarity search of high-dimensional vector data, commonly used for ML models, recommendation systems, and natural language processing applications.