AI Readiness Assessment: How to Know If Your Organisation Is Ready for AI Transformation

AI readiness measures capacity for change, not enthusiasm for tools

An AI readiness assessment should reveal whether your organisation can turn a defined AI opportunity into governed, adopted and measurable change. Use this evidence-led framework, checklist and scorecard to decide whether to reframe, repair, pilot or scale.

AI readiness is an organisation’s current ability to turn a defined AI opportunity into governed, adopted and measurable operational change. It depends on strategic clarity, decision rights, data, technology, governance, workflow readiness and people’s capacity to adopt new ways of working.

Enthusiasm is easy to mistake for preparedness. Leadership may have approved funding, employees may be testing public AI tools and a supplier may have demonstrated a convincing prototype. None of these conditions proves that the organisation can use AI safely in a live process or sustain its value.

A useful assessment therefore asks for evidence. A strategy should identify the decision or workflow that needs to improve. A governance policy should assign decision rights and operate in practice. Claims about data should be tested against the requirements of the proposed use case.

Major assessment approaches also extend beyond technology. Microsoft’s assessment covers business strategy, governance and security, data, organisational culture, infrastructure and model management. Cisco examines strategy, infrastructure, data, governance, talent and culture. The labels differ, but both treat technical capacity as one part of organisational readiness. See the Microsoft AI Readiness Assessment and Cisco AI Readiness Assessment.

AI readiness and AI maturity answer different questions

Readiness asks whether the necessary conditions exist to begin or extend a particular AI initiative. Maturity asks how consistently and effectively the organisation already develops, governs and operates AI.

An organisation may be mature in one capability and unready for another use case. It might operate forecasting models reliably but lack the content rights, evaluation methods or human-review controls required for a generative AI assistant.

The reverse is also possible. A team may be ready to run a narrow, low-risk pilot without having an enterprise-wide AI operating model.

The distinction prevents an unhelpful race towards the highest maturity level. The MITRE AI Maturity Model notes that an appropriate target depends on mission requirements and available resources.

The relevant question is not whether every capability is advanced. It is whether the capabilities required by the proposed use case are sufficient and supported by credible evidence.

Organisation-wide readiness does not guarantee that a use case is ready

Enterprise assessments identify shared strengths and weaknesses. They can inform investment planning, governance design and capability development. They do not replace a use-case review.

Every use case has its own value hypothesis, data dependencies, affected people, failure modes and accountability requirements. A customer-service assistant and a workforce-planning model may use the same cloud platform while creating different risks.

Assess readiness at two levels:

• Organisation level: Can shared capabilities support AI work repeatedly?
• Use-case level: Can this application create value within acceptable operational and governance boundaries?

Record the level being assessed at the beginning. Otherwise, participants may answer the same question with different scopes in mind and produce a score that appears precise but cannot support a decision.

A broad score can hide the constraint that stops delivery

Readiness is constrained by dependencies.

Strong infrastructure cannot compensate for data that cannot lawfully be used. A clear business case cannot compensate for the absence of an accountable owner. Training activity cannot compensate for a workflow in which employees have no time or authority to use the proposed system properly.

The LSI approach therefore does not treat a high average as proof of readiness. It combines a seven-dimension profile with evidence confidence, critical gates and comparisons between respondent groups.

An assessment is not a certification, technical audit or legal opinion. It is a structured management diagnosis. Specialist security, data protection, employment, procurement and sector-specific reviews may still be required.

Advanced AI Prompt Engineering

Large language models, such as ChatGPT, have transformed how we interact with technology. Leveraging their full capabilities, however, requires more than basic queries. This module delves into the art and science of crafting prompts that elicit...

Learn more

AI Readiness Assessment Framework

This framework uses seven dimensions. They are separated to expose gaps but should be interpreted as an interconnected system. A data problem may originate in unclear ownership, while resistance to adoption may reflect a poorly designed workflow rather than a cultural objection.

Strategic value and use-case clarity

Readiness begins with a decision or workflow worth improving, not with a technology looking for somewhere to be used.

Define the user, current problem, intended outcome and baseline performance. Explain why AI is more appropriate than a simpler process or software change.

The use case should have a testable value hypothesis, measures of success and explicit stop conditions. Otherwise, a pilot may appear successful because the team measures technical activity while overlooking cost, disruption or harm.

The MSc Digital Innovation and Entrepreneurship curriculum examines how AI opportunities can be shaped around real user needs, value propositions and commercially credible business cases.

Leadership, decision rights and funding

Executive sponsorship matters, but sponsorship alone is not ownership.

A ready initiative has someone accountable for the business outcome, named owners for delivery and risk decisions, and a funding path extending beyond a demonstration.

Decision rights should cover who can approve a pilot, accept residual risk, change the workflow, pause deployment and authorise expansion.

Data and knowledge foundations

Assess the information required by the use case rather than the organisation’s total volume of data.

Relevant considerations include availability, quality, representativeness, lineage, access controls, retention rules and permission to use the information for its intended purpose. For knowledge-based systems, examine document ownership, currency and conflicting sources.

The question is not whether the organisation has a data warehouse or a large document collection. It is whether the required information can be accessed, understood and maintained at the quality the application needs.

Where this is the central constraint, LSI’s GenAI data infrastructure readiness guide provides a more focused next step.

Technology and integration capacity

Technology readiness concerns the path from a controlled test into the systems where work happens.

Review architecture, identity and access management, integration interfaces, testing environments, observability, resilience and supplier dependencies.

A small internal pilot does not require the operating environment of an enterprise-wide service. It does, however, need a credible route to deployment if the evidence supports proceeding. Otherwise, it tests a model in isolation while leaving the difficult integration questions unanswered.

Governance, security and accountability

Governance should connect principles to decisions.

Identify applicable obligations, likely harms, affected groups, review points and the person authorised to accept or reject risk. Define how decisions and system changes will be recorded.

The NIST AI Risk Management Framework is voluntary and use-case agnostic. It can help structure risk-management work without assuming one sector or application.

ISO/IEC 42001 uses a management-system approach to establishing, implementing, maintaining and improving how an organisation manages AI. Neither source removes the need to interpret legal and sector-specific obligations for the proposed deployment.

Useful evidence may include risk classifications, impact assessments, threat modelling, approval records, incident routes and mechanisms for challenging or correcting outcomes. LSI’s article on accountability in AI adoption examines organisational ownership in greater depth.

Operating model and workflow readiness

AI creates value when it changes a real workflow.

Map the current process and the proposed one, including hand-offs, exceptions, human judgement and downstream consequences. Determine what happens when the system is uncertain, unavailable or wrong.

Operational readiness also includes support after launch. Someone must manage changes, respond to incidents, monitor performance and decide when reconfiguration, retraining or withdrawal is necessary.

A prototype without these responsibilities demonstrates technical feasibility, not operational readiness.

Skills, culture and adoption capacity

Assess the capabilities required by each group rather than looking only for technical specialists.

Leaders need sufficient understanding to make investment and risk decisions. Domain experts must be able to shape requirements and evaluate outputs. Affected employees need role-specific guidance, opportunities to practise and a credible explanation of how their work will change.

Culture is best assessed through behaviour. Examine whether people raise concerns, test assumptions, share lessons and stop weak initiatives. Survey sentiment can be useful, but it should be compared with participation and observed practice.

When leadership capability is the primary gap, see LSI’s guide to AI training for leadership teams.

AI Readiness Assessment Framework Examples

Frameworks serve different decisions. An organisation should not select one simply because it produces a convenient score.

The Digital Cooperation Organization describes AI-REAL as a national assessment containing five pillars, 17 dimensions and 39 indicators. National, enterprise, workforce and use-case assessments may share themes, but their evidence and recommendations are not interchangeable. See the DCO announcement.

What company AI transformations reveal about readiness

Company cases cannot prove that one readiness framework works universally. They can show what organisational readiness looks like once AI moves beyond experimentation.

DBS built readiness before generative AI arrived. The bank reports that it began industrialising AI in 2014. By 2025, it had more than 430 use cases supported by over 2,000 models and attributed approximately SGD 1 billion in economic value to its data analytics and AI initiatives.

More important for an assessment is what sat behind those figures. DBS identifies data architecture, responsible AI, talent and organisational agility as foundations for expansion. Its reporting also describes governance guardrails, reusable components and operating-model transformations.

The lesson is not that another organisation should reproduce its number of models. It is that access to a powerful model is not equivalent to possessing the organisational system required to use AI repeatedly. The outcomes remain company-reported rather than independent causal evidence. See the DBS 2025 CIO Fstatement and its response to shareholder questions.

IKEA Retail combined automation with role redesign. Ingka Group reported that its AI-supported chatbot, Billie, resolved approximately 47% of the enquiries it received between 2021 and 2023. This covered 3.2 million interactions and generated nearly EUR 13 million in reported savings.

Ingka also reported reskilling 8,500 customer-support employees in remote interior design, digital retail sales, relationship building and complex problem-solving.

The readiness evidence is therefore not only the chatbot’s resolution rate. It includes a redesigned division of labour and investment in the capabilities required for the work retained by people. These results are company-reported and do not constitute an independent evaluation of service quality. See Ingka Group’s account of Billie and remote selling.

Klarna shows why readiness must be reassessed. The company reported extensive adoption of its AI customer-service assistant and approximately USD 39 million in cost savings during 2024. It also estimated that the system performed work equivalent to hundreds of full-time agents.

Klarna subsequently acknowledged that it had overemphasised cost reduction and began correcting its approach towards service quality, product improvement and a continued role for people in complex interactions.

This does not mean the implementation failed. It demonstrates that technical adoption and efficiency can coexist with weaknesses in the outcome being optimised. An assessment limited to conversation volume or resolution speed might classify the initiative as fully ready. An assessment incorporating customer outcomes, exception handling and human escalation would produce a more useful diagnosis.

Klarna’s figures are company-reported in its regulatory filing. Its later strategic correction was covered by Reuters.

Across the cases, readiness becomes visible when outcomes connect to real workflows, employees are prepared for changed roles, governance works alongside delivery and operational evidence can change the original strategy.

MSc AI for Business Transformation

Artificial intelligence is revolutionising the way businesses operate, creating new opportunities and challenges; This programme is designed to equip you with the expertise to navigate this evolving landscape with confidence and precision. You will...

Learn more

Run the assessment as a cross-functional investigation

The quality of an assessment depends on who participates, what evidence they examine and how disagreements are handled.

Do not let the most senior participant determine the initial group score.

Independent scoring exposes differences that a consensus workshop may conceal. Leaders may believe data is accessible while delivery teams know permissions remain unresolved. Governance colleagues may assume human review exists while operational staff know that workload makes it impractical.

These disagreements are findings, not inconveniences to be averaged away.

Use this checklist to test the evidence behind each score

Answer every question with a link to evidence, a named owner and the date the evidence was last verified.

A yes or no response is not sufficient. Record whether the evidence applies across the organisation, to one business unit or only to a controlled pilot.

The checklist is a diagnostic prompt rather than a universal control set. Add sector-specific requirements and questions concerning the people affected by the system.

London School of Innovation

LSI is a UK higher education institution, offering master's degrees, executive and professional courses in AI, business, technology, and entrepreneurship.

Our focus is forging AI-native leaders.

Learn more

The LSI scorecard turns seven dimension scores into four decisions

The LSI scorecard is an editorial decision framework created for this guide. It has not been validated as a predictive instrument.

Its purpose is to improve management decisions by making evidence, constraints and decision conditions explicit.

Score current evidence from absent to scalable

Score each dimension from 0 to 4. Record the rationale and supporting evidence next to the number.

Do not award a higher score because a policy, platform or recruitment programme is planned.

Add an evidence-confidence rating to every dimension:

A score of 3 with low confidence should trigger verification rather than confidence. Keep the seven scores visible as a profile. If stakeholders request an average, show it only as a summary and never use it to override a weak critical dimension.

Critical gates cap the overall result

Before recommending a pilot or expansion, test five gates:

1. The use case and measurable value hypothesis are defined.
2. The required information can be used lawfully and at sufficient quality.
3. A named owner is accountable for the operational outcome.
4. Governance and security boundaries are proportionate to the use case.
5. There is a viable path for human oversight, correction and stopping the system.

If a gate is not met, the decision cannot be scale.

If the missing condition would make even a controlled trial unsafe, unlawful or operationally misleading, the decision should be repair or reframe.

Cross-functional disagreement is a readiness finding

Compare the scores from each respondent group before discussing consensus.

A difference of two or more points deserves investigation. A substantial difference in evidence confidence also matters, even when the numeric score is the same.

Ask which evidence each group used, whether they assessed the same scope and whose work would absorb the risk. Do not automatically average the responses.

Agreement produced by hierarchy is not the same as shared operational understanding.

Four decisions: reframe, repair, pilot or scale

Apply the decisions in order.

Do not choose pilot merely because the overall score appears acceptable. Do not choose scale because the technology performed well in isolation.

An unresolved difference of two or more points on a critical dimension prevents a scale decision until the evidence and scope have been reconciled.

Fix the blocker that matters to the first use case

A long transformation roadmap can dilute responsibility. Start with the constraint that most directly limits the first worthwhile use case.

Avoid solving the easiest gap simply because progress will be visible.

If data rights are the constraint, an awareness course will not make the use case ready. If no one owns the workflow outcome, another technical demonstration may increase activity without resolving accountability.

Once a pilot has credible evidence, the challenge becomes operationalisation. LSI’s guide to moving AI from pilot to production examines that transition.

The cost of AI implementation also becomes clearer after the assessment has exposed integration, governance and adoption work that a software price omits.

Choose an assessment tool by what it can actually prove

AI readiness assessment tools are useful when they match the decision and make their limitations visible.

Before selecting one, ask what evidence it accepts, whether it distinguishes organisation-level and use-case readiness, how it treats critical gaps and whether recommendations can be traced to responses.

Use a combination when the decision carries material cost or risk. A survey can identify a perception gap, a workshop can investigate it and evidence review can determine whether the claimed capability exists.

GenAI readiness requires additional evidence

A gen AI readiness assessment should retain the seven organisational dimensions while examining risks created or intensified by generative systems.

The NIST Generative AI Profile is a cross-sector companion to the AI RMF and can inform this additional work.

Add evidence questions covering:

• the provenance, rights and permitted use of grounding and user-supplied content;
• evaluation of factual reliability, harmful output and user needs;
• prompt injection, sensitive-data disclosure and unsafe system connections;
• changes to third-party models, service terms and data handling;
• boundaries of human review where an output can trigger an action;
• logging, incident investigation and feedback processes.

For agentic systems, assess the permissions attached to every action, not only the quality of generated text.

A system that can retrieve a document presents a different operational risk from one that can modify a record, contact a customer or initiate a transaction.

Readiness becomes credible when reassessment changes investment decisions

An AI readiness assessment expires as the organisation and use case change.

Reassess on a planned cycle and after material events such as a new data source, supplier change, workflow redesign, security incident or expansion to a different group of users.

Track whether the assessment changes decisions. Useful signals include unsuitable ideas being stopped earlier, evidence gaps receiving named owners, pilots having explicit exit criteria and scale funding being released only after operational proof.

If every assessment concludes that the organisation is ready, the method is probably confirming ambition rather than testing it.

The goal is not a perfect readiness score. It is a more defensible sequence of decisions about where AI can create value, what must be repaired first and what evidence is required before greater exposure.

These connections are examined in LSI’s MSc AI for Business Transformation through AI strategy and implementation, governance, change management and requirements engineering.