An artificial intelligence (AI) strategy depends on information the organisation can use for its intended purpose. Before committing to a new service, leaders need to establish whether that information is available and whether its meaning and limitations are understood.
Data management and governance therefore belong in the initial investment discussion. Accurate data remains essential. The strategy must also establish when that data applies and whether it is suitable for the decisions AI will support.
A correct figure may refer to a different period from the document explaining it. A product description could apply to one market while the price comes from another. When AI combines sources into an answer, those differences must remain visible.
The strategy needs to provide for preserving that context as information is put to new uses.
The disciplines described in DAMA’s Data Management Body of Knowledge already include metadata, document and content management, integration, architecture and security alongside quality and governance. Context and unstructured information were never outside the scope of data management.
Where investment has concentrated on structured information in operational systems and reporting, documents and other business knowledge may have been managed separately. Their ownership and controls can differ from those applied to database records.
An AI service may draw on transaction records alongside written explanations that employees previously consulted separately. The organisation must decide how these sources can be used together and which takes precedence when they disagree.
The data strategy must cover those relationships across the organisation’s systems and document collections.
Consider an AI service helping employees understand the organisation’s products.
Structured data might provide product codes, prices, specifications and availability. Unstructured information might include product guides, technical manuals, training material and approved descriptions.
The correct product record must be connected to the document version for the relevant market. A current price combined with an outdated specification can produce a convincing but misleading answer, even though each source accurately records something.
The strategy must therefore address more than access to these sources. It needs consistent identifiers, shared definitions, document ownership and rules for resolving conflicting information. It must specify how updates and withdrawals will reach the AI service.
Simply collecting the material in one place does not establish those relationships.
Metadata provides descriptions of information and the context needed to interpret it. This includes definitions, ownership, dates, versions, classifications, permitted uses and relationships with other information.
For an AI-enabled service, these descriptions need to be available and usable within the processes that select and supply information. A catalogue entry that nobody maintains, or that the service cannot use, offers limited protection against a misunderstanding.
Lineage adds another essential dimension. It helps establish where information originated and how it was transformed or combined. Where an AI service retrieves organisational sources, retaining a record of the material used can support investigation and review. It does not, on its own, prove that the resulting answer is correct.
The strategy needs to fund this work and assign responsibility for keeping the descriptions and source histories current. Teams should establish these requirements while designing the service.
Context also includes the rules governing how information should be interpreted and used. Business rules establish conditions such as eligibility, approval limits and exceptions. Data rules define valid values, relationships, calculations and checks.
In the product example, knowing the current price is insufficient if a discount applies only to a particular customer category or requires approval. An AI-generated answer needs to respect those conditions.
These rules may already exist in application code, procedures, spreadsheets or employees’ knowledge. Developing an AI strategy requires the organisation to identify the relevant rules, resolve inconsistencies and establish which versions are authoritative. It must also determine how those rules will be enforced. Including a rule in a document or prompt does not guarantee that AI will follow it.
The familiar warning “garbage in, garbage out” describes the consequences of poor inputs.
With generative AI, the concern can become “garbage in, gospel out”. An answer assembled from incomplete, outdated or misunderstood information may be expressed with enough confidence that people accept it as authoritative.
The weakness is not necessarily a visibly incorrect value. It may be missing context, an unidentified exception or a relationship the system has inferred incorrectly.
This changes what the strategy must provide for. Improving the sources is essential, but the organisation must also evaluate the outputs, make relevant evidence available and establish when a person should question or escalate an answer.
Reliable inputs reduce risk. They do not remove the need to assess what AI produces from them.
The business purpose should determine which information and controls take priority. For each proposed AI use, the organisation needs to identify the structured records and unstructured content involved, the context required to interpret them, and the consequences of getting that interpretation wrong.
That assessment should determine the order of investment. Before adding more sources or upgrading a model, the organisation may need to agree which documents are authoritative. Resolving conflicting definitions can also support several proposed uses of AI.
Governance must identify who approves sources and maintains their context. It must also establish who investigates a disputed output and can authorise a correction or stop the service. The investment plan needs to cover these continuing responsibilities.
Education grounded in DAMA remains valuable. People now also need practice in applying these disciplines together when developing and using AI services.
The learning priorities should include:
Meaning and context. Understanding business definitions, metadata, relationships and the limitations of information, alongside conventional quality measures.
Structured and unstructured information. Learning how records, documents and other content can be connected without losing version, authority or access restrictions.
Business and data rules. Building on established skills in defining rules, with training adapted to work involving AI. This includes making assumptions and exceptions explicit, assigning ownership, managing changes and testing whether a service applies the rules correctly. People also need to distinguish decisions that can follow a defined rule from those requiring human judgement.
Lineage and evidence. Understanding how to trace sources and transformations, and why a source reference does not automatically validate an AI-generated claim.
Governance in use. Applying ownership, stewardship and permitted-use rules to the information supplied to AI, with clear responsibility for reviewing its outputs.
Strategic judgement. Assessing information dependencies before committing to an AI use case, and recognising when preparation, a narrower scope or further testing is required.
The depth will differ by role. Leaders need to understand the investment implications. Data practitioners need to implement the supporting disciplines. Business specialists must contribute the meaning and rules that technology cannot be assumed to know. Employees using AI need to recognise when an answer warrants scrutiny.
Training should use practical exercises that connect established data management knowledge to these responsibilities.
Before approving an AI investment, leaders should be able to explain which information it will use, what context and rules must accompany it, and who will maintain them. Those answers belong in the strategy and the learning plan for the people who will put it into practice.
Reference
DAMA International. What is Data Management? dama.org/about-dama/what-is-data-management/
Data Foundations for the AI Era | Datanacity Thought Leadership Series | Article 3
Published by Datanacity with permission from Jill Flynn.
If you'd like to stay up-to-date with all of our latest training courses and promotional webinars please subscribe to our newsletter.
