AI promises to transform how your business finds, analyses and uses information. But before you give it access to your information estate, there's a fundamental question you need to answer: do you know what information you're making available, and whether it should be accessible in the first place?
Across businesses, AI is moving beyond experimentation into everyday operations. From searching corporate documents and summarising reports to supporting customer services and business decisions, the opportunities to improve productivity and efficiency are significant.
But while attention is focused on selecting AI tools, identifying use cases and demonstrating returns, an equally important question can be overlooked: is the information underpinning those AI systems ready to be used?
Your business may hold years of contracts, customer records, employee information, financial documents, emails and project files across Microsoft 365, SharePoint, document management systems and other repositories. Some of that information is current, accurate and useful. Other records may be outdated, duplicated, incorrectly classified or retained long after their original purpose has ended.
When AI begins searching and analysing that information, existing information management weaknesses can become more consequential.
AI doesn't automatically make your information more reliable, relevant or appropriately governed. If you don't understand and control the information it can access, you may be introducing new risks alongside the expected benefits.
Is your information estate really ready for AI?
Preparing for AI is often approached as a technology challenge. Businesses assess capabilities, security, integrations, licensing and potential use cases, but the condition of the underlying information estate doesn't always receive the same attention.
This is a concern highlighted in PwC's Risk Agenda for Assurance Functions 2026. In its section on AI-ready foundations, PwC identifies how initial AI projects and attempts to scale deployment can expose fragmented information, inconsistent standards, legacy infrastructure and weaknesses in data governance.
PwC also emphasises the importance of information quality, metadata, traceability and clearly defined ownership in establishing reliable foundations for AI.
These findings raise an important point. You may have the technology needed to introduce AI, but that doesn't necessarily mean the information it will use is fit for purpose.
Consider an AI assistant connected to your corporate document repositories. It may be able to locate and summarise information far more quickly than someone searching manually. But what happens if it retrieves an outdated policy, relies on a superseded contract or incorporates information that should no longer be retained?
And even if the answer is factually correct, was the underlying information appropriate for that user or purpose?
The issue isn't necessarily with the AI tool itself. It may be exposing information management problems that already existed but were less visible when employees had to search for, locate and interpret records individually.
Before asking what AI can do with your information, you need to understand what information you hold, how it is managed and whether it is suitable for the intended use.
Five information governance challenges that can undermine AI adoption.
Effective AI deployment depends on more than technical performance. The information made available to AI systems must also be appropriate, reliable and subject to the necessary controls.
There are five areas where weaknesses in information governance can create problems.
-
Do you know where all the information your AI tools might access is held?
For many businesses, information is distributed across multiple repositories, departments and applications. Historic shared drives, archived projects, collaboration platforms and older business systems may contain large volumes of information that haven't been reviewed for years.
Not all of this information will necessarily be connected to an AI system. However, when you begin making repositories available, you need to understand what they contain and whether existing governance controls are adequate.
Without that visibility, it becomes difficult to determine which information should be included, which should be restricted and which requires further review.
A useful starting point is to map the repositories relevant to your proposed AI use cases, identify the types of information they contain and establish who is responsible for them.
You don't need to catalogue every document before beginning a controlled AI pilot, but you do need sufficient understanding of the information within its scope to make informed decisions about access and use.
-
AI can make retrieving and analysing information faster, but speed is of limited value if the underlying material is unreliable.
Consider a project repository containing several versions of the same document, some approved and others still in draft. Or a collection of corporate policies where superseded versions haven't been clearly identified.
An AI assistant may retrieve information from those sources without necessarily understanding which document represents the current, authoritative position.
This can result in responses based on outdated, incomplete or conflicting information, particularly where the system lacks reliable versioning, metadata or source controls.
PwC's 2026 Risk Agenda highlights the importance of information quality, metadata standards and traceability in establishing trustworthy AI foundations. Being able to understand where information originated and how it has been managed is an important part of assessing whether it can be relied upon.
Information governance contributes by establishing ownership, classification, lifecycle management and processes for identifying obsolete or superseded records.
These controls don't guarantee accurate AI outputs. AI systems still require appropriate testing, evaluation and human oversight. However, improving the reliability of the information they use removes an important source of avoidable error.
-
Your business holds information with different levels of sensitivity and different restrictions on how it can be accessed or shared.
Employee records, financial information, commercially sensitive documents, intellectual property and confidential agreements may all require specific controls.
Introducing AI raises questions about whether those controls remain effective when information is accessed through a new interface.
For example, an employee may be authorised to use a corporate AI assistant, but that doesn't mean they should be able to retrieve information from every connected repository.
The UK's National Cyber Security Centre's Guidelines for Secure AI System Development emphasise the need to establish controls over the information AI systems can access and to apply appropriate security measures throughout the AI lifecycle.
For your business, this means understanding existing permissions, identifying restricted information and ensuring that AI integrations enforce the intended access boundaries.
It's also important to consider what happens to information once it is processed. Depending on the AI service and its configuration, you may need additional controls governing data transmission, processing, storage and subsequent use.
An AI deployment shouldn't become a route around the information security and confidentiality controls you already have in place.
-
One of the less obvious risks of AI adoption is the continued availability of information that should already have been reviewed for retention or disposal.
Most businesses accumulate historic records over time. Some must be preserved for legal, regulatory or contractual reasons, while others remain simply because no consistent process exists for reviewing and disposing of them.
If those repositories are connected to AI tools, unnecessary information may continue to be searched, retrieved and used.
This creates several concerns. Historic material may influence outputs, sensitive information may remain unnecessarily exposed and the business may find it increasingly difficult to demonstrate that its retention requirements are being followed.
The UK's Information Commissioner's Office addresses this issue in its guidance on AI and data protection. It explains that data minimisation and storage limitation remain relevant when personal data is processed through AI systems. Businesses must consider whether the personal information used is necessary for the intended purpose and whether it is being retained appropriately.
Although these requirements relate to personal data under UK law, the wider principle applies to effective information governance: information shouldn't be retained indefinitely simply because it might be useful to AI in the future.
This doesn't mean deleting historic information indiscriminately. Legal holds, regulatory requirements and legitimate business needs must always be considered.
It means establishing reliable processes for identifying what should be retained, what must be protected and what can be defensibly disposed of.
Reducing unnecessary information can help reduce exposure, control costs and limit the volume of irrelevant material available to AI. Where information is removed from source repositories, you must also consider how that change is reflected in AI search indexes, cached content and other downstream copies.
-
When AI produces a summary, recommendation or answer, how confidently can you establish which information it used?
For routine tasks, this may be relatively straightforward. For decisions involving customers, financial information, regulatory obligations or commercially sensitive matters, understanding the source of information can become much more important.
If a response is based on incomplete or outdated records, can you identify the source and correct the underlying problem? If information was used inappropriately, can you establish who was responsible for managing it and which controls should have applied?
PwC's research highlights the importance of traceability, metadata, versioning and defined ownership in supporting confidence in AI outcomes.
Information governance contributes by establishing consistent information management practices, clarifying responsibility and maintaining records of governance decisions.
However, source information governance alone cannot provide full AI explainability. AI systems also need appropriate logging, monitoring, evaluation and oversight to establish how information has been processed and used.
The two disciplines need to work together.
Information governance and AI governance: Why do you need both?
As AI adoption increases, businesses are understandably investing more attention in AI governance. This includes establishing policies for acceptable use, evaluating models, managing security, assessing regulatory requirements and monitoring AI-generated outputs.
These are important responsibilities, but they don't replace the need to govern the information made available to AI.
AI governance addresses how AI systems are selected, developed, deployed, used and monitored. Information governance addresses the information those systems can access and how it is managed throughout its lifecycle.
The distinction matters because even well-designed AI controls can be undermined by poorly managed source information.
You might have an approved AI platform, clear user policies and appropriate technical safeguards. But if the connected information estate contains outdated records, unclear ownership, excessive permissions or information retained unnecessarily, important risks remain unresolved.
Conversely, good information governance doesn't automatically make an AI system safe or reliable. Model behaviour, output accuracy, third-party processing and human oversight still require dedicated controls.
The strongest approach connects the two, ensuring that information governance forms part of AI planning and implementation rather than being considered after deployment.
Where should you start?
Six questions to ask before deploying AI.
You don't need to resolve every information governance issue across the business before introducing AI. However, you should understand the information risks associated with each proposed use case and establish appropriate controls before expanding access.
A practical assessment should begin with six questions.
-
Identify the repositories, information categories and data sources relevant to the proposed AI deployment. Establish whether any contain sensitive, historic or poorly classified material that requires additional review.
-
Determine which departments are accountable for the information and who can authorise its use. Involve the appropriate business owners, Information Governance, IT, Security, Legal, Risk and Privacy teams.
-
Consider whether the information is current, whether authoritative versions can be identified and whether appropriate metadata or classification controls exist. Don't assume that making more information available will automatically improve AI performance.
-
Review how the AI solution handles existing permissions, confidential information and restricted repositories. Test whether users can retrieve information beyond their authorised access and confirm that the relevant controls operate as intended.
-
Review applicable retention requirements, legal holds and business needs. Identify information that is no longer required and establish processes for defensible disposition, including consideration of information held in downstream AI indexes or copies.
-
Establish how information access, governance decisions and relevant AI activity will be monitored. Define responsibility for reviewing exceptions, investigating problems and maintaining controls as the deployment expands.
These questions provide a starting point for assessing information readiness without turning every AI initiative into a large-scale information management project.
Start with controlled information, then expand.
One of the risks of AI adoption is assuming that greater access to information will always produce better results.
In practice, providing access to large, poorly understood repositories can introduce unnecessary complexity, increase exposure and make it harder to establish whether the system is working as intended.
A more effective approach is to begin with a defined use case, select appropriate information sources and establish the necessary governance controls before expanding.
For example, you might start with a carefully managed repository of approved corporate policies or operational documents. Once you understand how the AI system accesses and uses that information, you can assess its performance, review governance requirements and identify what is needed to extend the deployment.
This approach reflects PwC's recommendation to assess foundational data capabilities and establish appropriate ownership, monitoring and controls before scaling AI use cases.
It also creates an opportunity to address information management weaknesses as part of the AI programme, rather than treating them as unrelated problems.
Over time, better information visibility, retention controls, classification and ownership can support additional AI use cases while delivering wider benefits for risk management, compliance and operational efficiency.
Don't let information governance become an afterthought in your AI strategy.
AI has significant potential to change how your business works with information. But the benefits depend partly on the quality, relevance and control of the information it can access.
If you don't understand what you hold, who owns it, whether it is accurate or how long it should be retained, those weaknesses don't disappear when AI is introduced. They may become more difficult to identify and manage.
That's why information governance needs to be part of the AI conversation from the beginning.
By establishing visibility, ownership, retention, defensible disposition and appropriate access controls, you can reduce unnecessary information risk and create a more reliable foundation for AI adoption.
The objective isn't to delay innovation until every information management problem has been solved. It's to make informed decisions about what information AI should use, establish the controls required and expand adoption with greater confidence.
Before you give AI access to your information, make sure you understand what you're giving it access to - and whether it should be there at all.
Put information governance into action.
iCompli helps you understand and manage information across your existing systems, apply consistent retention and disposition policies, coordinate governance workflows and maintain evidence of decisions.
Reduce information risk, control unnecessary costs and turn governance requirements in to practical action.