AI readiness is the degree to which an organisation’s identity controls, managed devices, data access rules and security policies can support AI safely before any AI is switched on. It measures whether the foundations AI will run on are funded, owned and working, not how many licences or pilots exist.
The concept goes by several names: AI foundations, AI preparedness, AI-ready infrastructure, the AI readiness baseline and, in governance circles, AI management system readiness. All describe the same test: if an AI tool or agent were given access tomorrow, would the organisation know what it could reach, and could it stop it?
The standards boards already rely on make this a governance matter rather than an IT preference. ISO/IEC 42001:2023 requires an organisation to determine and provide the resources its AI management system needs, and its Annex A controls ask for those resources, including data, tooling, computing and people, to be documented for each AI system. ISO/IEC 27001:2022 includes controls for access control, identity management and user endpoint devices. The NIST Cybersecurity Framework 2.0 places identity management, authentication and access control within its Protect function.
Only about 13% of organisations, which Cisco calls “Pacesetters”, have consistently outperformed their peers on AI value for three years running, and 64% of organisations struggle to centralise their data (Source: Cisco, AI Readiness Index, 2025).
This analysis builds on a recent conversation with Lior Bela, Business Director for Microsoft Intune, about what organisations and their service providers must put in place before AI adoption.
Why does AI readiness start before AI?
AI readiness starts before AI because AI inherits whatever access, data and controls already exist. An AI assistant that can search the organisation’s files will find everything a user can technically reach, including material that was never meant to be shared. The AI does not create these weaknesses; it exposes them at speed. That is also why the pace of adoption is a board decision; see how boards should set the pace on AI.
Bela’s argument is that AI depends on three things being in place first: connected workloads, managed endpoints, and consistent identity and access controls. He likens buying advanced AI without them to buying a new car before anyone has paved the road. The conversation also covers how legacy tooling creates a ceiling on AI adoption.
What is an AI readiness framework? The five foundations to fund first
The foundations are the controls that decide who and what can reach the organisation’s data, and under which rules. They rarely feature in AI business cases, yet they determine whether AI can be deployed at all.
| Foundation | The executive question it answers | What goes wrong without it |
|---|---|---|
| Identity and access | Who, and which AI agents, can reach which data and systems? | AI surfaces information to people who should never see it |
| Device and endpoint management | Are the devices AI runs on known, patched and governed? | Sensitive outputs land on unmanaged or personal devices |
| Data location and classification | Do we know where our sensitive data is and how it is labelled? | AI draws on ungoverned copies and outdated sources |
| Security policy enforcement | Are the rules applied consistently, whatever the device or location? | Controls hold in the office and fail everywhere else |
| A sanctioned AI option | Do employees have an approved tool that is easier than the alternatives? | Shadow AI moves data into environments the organisation does not control |
The last row matters more than it appears. As the discussion put it, when organisations fail to offer sanctioned AI tools, “shadow AI fills the gap”. The data dimension is explored further in Privacy and Security Are One Risk.
The AIPEX view: fund readiness as a gate, not a line item
At AIPEX we see organisations approve AI spending and foundation spending in separate budgets, owned by separate executives, on separate timetables. The AI budget moves quickly; the foundation budget waits because it looks like maintenance. The result is AI deployed onto ground that cannot bear it.
Our recommendation is a reframing: treat foundation readiness as a formal gate that every AI investment must pass, not as a separate IT line item competing for funds. In practice that means three executive decisions:
- One readiness standard. The executive team agrees a short, written baseline for identity, devices, data and policy that any AI use case must meet before it reaches real users.
- Joint funding. Each AI business case carries the cost of closing its own foundation gaps, so the true cost of AI is visible and the foundation work is not starved.
- Agents as workers. Every AI agent is given defined access, environment, data reach and authority, as a new employee would be, a point Bela also makes. We set out the ownership model in Who Owns the AI Agent When It Breaks?
The identity-first thinking behind zero trust applies directly, and, as we argue in Zero Trust Is a Leadership Decision, it cannot be delegated to the technology function alone.
How should executives assess AI readiness, and where does ISO/IEC 42001 fit?
Executives should assess AI readiness on evidence, not assurance: for each proposed AI use, can the organisation show who has access, what data is in reach and who can stop it?
ISO/IEC 42001:2023 provides a useful structure for that assessment even where certification is not the goal. Its requirements for AI risk assessment, defined roles and documented resources force the questions most AI business cases skip. A readiness review against it usually shows the gaps lie in the underlying controls, not the AI policy.
Bela describes a maturity path in three tiers: baseline alignment, where policies are in place; assisted automation, where AI recommends and people verify; and autonomous operation, where trust has been earned through repeated validation. For boards, the point is that autonomy is earned, and no organisation should move up a tier while the baseline is incomplete. Organisations that skip the baseline tend to end up in AI pilot purgatory, with pilots that work in isolation and cannot be released safely.
What should the board ask before approving the next AI project?
- If this AI tool went live tomorrow, what data could it reach, and is that appropriate?
- What proportion of the devices our people use are managed and governed today?
- Do we know where our sensitive data is, and is it classified consistently?
- Which AI tools are employees already using without approval, and what sanctioned alternative have we given them?
- Does this business case include the cost of closing its foundation gaps?
These five questions are a starting point. Our briefing Is Your Business Ready for AI? includes the full AI readiness checklist for boards, with eight scored questions, the six warning signs and the six readiness domains. The AI Readiness Assessment scores where your organisation stands.
If these questions have no clear owner, that is the first gap to close. A Fractional Chief AI Officer can hold AI investment and foundation readiness under one accountable executive, so neither is funded without the other. The Executive Strategy Day assesses your readiness baseline against each planned AI use case and sets out which foundations must be funded first.
Frequently asked questions
What does AI readiness actually mean for an executive team?
AI readiness is whether the identity controls, managed devices, data access rules and security policies that AI will run on are already funded, owned and working. It is measured by evidence of who and what can reach the organisation's data, not by licences bought or pilots launched. For a full explanation, see AI Readiness Starts Before AI: The Foundations Executives Must Fund First →
Can ISO/IEC 42001 be used to assess AI readiness without seeking certification?
Yes. ISO/IEC 42001:2023 requires AI risk assessment, defined roles and documented resources for each AI system, which makes it a practical structure for a readiness review even where certification is not the goal. Such reviews usually show that the gaps lie in underlying controls rather than AI policy. For a full explanation, see AI Readiness Starts Before AI: The Foundations Executives Must Fund First →
Why should AI agents be governed like employees?
AI agents act continuously across systems, so they need defined access, a defined environment, defined data reach and defined authority, just as a new employee would. Someone must also own the agent and be able to suspend it. For a full explanation, see Who Owns the AI Agent When It Breaks? Delegated Authority for Boards →