Why AI maturity matters
In Italy, the gap between those who talk about artificial intelligence and those who implement it successfully widens every quarter. According to data from the AI Observatory of the Politecnico di Milano, in 2025 the Italian AI market exceeded 900 million euros, with growth of 58% compared with the previous year. And yet only 18% of Italian SMEs have launched structured AI projects. The remaining 82% swing between curiosity, confusion and isolated attempts that rarely produce measurable results.
The problem is not a lack of technology. The tools exist, they are accessible and in many cases they don’t require prohibitive investment. The problem is a lack of awareness: most companies don’t know where they stand on their path towards AI, don’t know their own strengths and weaknesses, and as a result can’t set sensible priorities.
Assessing the AI maturity of your organisation is not an academic exercise. It is the first operational step to turn AI from a buzzword into a competitive lever. Without this initial snapshot, any investment risks being misdirected: tools are bought without the data to feed them, pilot projects are launched without the skills to manage them, immediate results are sought without having built the foundations.
This guide provides a complete, practical framework for carrying out an AI maturity assessment of your company. You don’t need advanced technical skills: you need intellectual honesty, the willingness to look at your organisation with a critical eye and a structured method for doing so. Whether you are a business owner, a general manager or an innovation manager, by the end of this read you will have the tools to understand exactly where your company stands and, above all, what the next steps are.
The 5 dimensions of AI maturity
AI maturity can’t be measured with a single indicator. It is a multidimensional concept that requires assessing five fundamental areas of the organisation. Neglecting even one of them means building on fragile foundations. Let’s look at them in detail.
Data
Data is the fuel of any AI system. Without quality data, even the most sophisticated algorithm produces useless results. But "having data" is not enough: you need to assess its quality, accessibility, integration and governance.
A manufacturing SME, for example, might have years of production data stored in spreadsheets scattered across different departments, with no naming or format standard. Technically the data exists, but it is unusable for an AI system without significant cleaning and structuring work.
The key questions to ask in this dimension are: Is our data centralised or fragmented? Are there quality standards and validation processes? Is the data accessible in real time or only through manual exports? Is there a data catalogue describing what each source contains? Who is responsible for data quality, and with what mandate?
A retail company with an integrated ERP, a CRM connected to e-commerce and automated data quality processes is in a radically different position from one that works with three ERPs that don’t communicate and duplicate customer data. The difference is not technological: it is organisational and cultural.
Technology
The technology dimension assesses the existing IT infrastructure and its capacity to support AI workloads. It isn’t about having the latest GPU model or a cloud data lake: it is about understanding whether the current architecture can integrate AI solutions without having to be rebuilt from scratch.
For a typical Italian SME, this means assessing very concrete aspects. Does the ERP support APIs for integration with external services? Is the cloud infrastructure (if there is one) sized to handle machine learning workloads? Are there development and test environments separate from production? Do the critical systems have an architecture that allows AI components to be added without compromising stability?
An often underrated aspect is the maturity of integrations. A services company that uses five SaaS tools not integrated with each other will have enormous difficulty implementing an AI assistant that needs to access information spread across several platforms. Conversely, an organisation with integration middleware or well-documented APIs can experiment with AI incrementally and at low risk.
People
The human dimension is probably the most critical and the most neglected. AI doesn’t replace people: it empowers them. But to do so it takes specific skills, a mindset oriented towards change and an organisational culture that welcomes innovation rather than resisting it.
In this dimension, three levels of competence are assessed. The first is AI leadership: does management understand the potential and the limits of AI? Can it make informed decisions on where and how to invest? The second is operational competence: are there people able to manage AI projects, deal with technical suppliers, define requirements and evaluate results? The third is widespread literacy: are operational staff able to work with AI tools, interpret their results and provide feedback to improve them?
A common mistake in Italian SMEs is thinking that hiring a data scientist is enough to "do AI". Without management able to define the business objectives and an operational team able to integrate AI into daily processes, the data scientist will work in isolation and produce models that nobody uses.
Equally important is resistance to change. In a manufacturing company with workers who have been working the same way for twenty years, introducing an AI system for quality control can generate fear and hostility. Without a change management plan that accompanies people through the change, even the best technology will fail.
Processes
AI doesn’t improve processes that don’t exist or are poorly defined. Before automating, you need to standardise. This dimension assesses how far the company’s processes are documented, measured, repeatable and ready to be enhanced by AI.
Let’s consider a concrete case. A services company wants to use AI to automate the qualification of sales leads. But the current sales process isn’t formalised: each salesperson follows their own method, the qualification criteria are subjective and there is no CRM where the steps of the funnel are tracked. In this situation, implementing an AI lead-scoring system is not only premature but counterproductive: the system learns from inconsistent data and produces unreliable results.
The key questions for this dimension include: Are the core processes documented and standardised? Are there KPIs measured systematically for each key process? Are the processes repetitive enough to benefit from automation? Is there a culture of continuous improvement, with periodic process reviews? Are exceptions to standard processes tracked and analysed?
Strategy
The strategy dimension assesses whether and how AI is integrated into the company’s vision and planning. It isn’t about having a separate "AI plan", but about understanding whether the leadership considers AI a strategic lever for achieving business objectives.
A company with high strategic AI maturity has a clear vision of how AI will contribute to competitive advantage over the next 3-5 years, allocates a dedicated budget to AI initiatives, has defined a governance for selecting and prioritising AI projects, measures the ROI of AI initiatives with business metrics (not just technical metrics), and integrates AI into annual strategic planning on a par with other levers such as internationalisation or product development.
Conversely, a company with low strategic maturity approaches AI reactively: it responds to external stimuli (a competitor adopting AI, a supplier proposing a solution, an article read by the CEO) without a structured decision-making framework. The result is sporadic, uncoordinated initiatives that are hard to evaluate.
The 5 levels of AI maturity
For each of the five dimensions, the organisation places itself on one of five maturity levels. This scale is not a value judgement: it is a diagnostic tool that shows where to concentrate efforts. Not every company needs to reach level 5 on every dimension; the goal is to reach the level appropriate to its strategic objectives.
Level 1 — Initial
The organisation has no AI initiatives under way and awareness of the topic is minimal or absent. Data is fragmented and ungoverned, the IT infrastructure is legacy, AI skills are non-existent and there is no strategic vision linked to artificial intelligence.
Typical signs: company data lives in unstructured spreadsheets; there is no modern CRM or ERP; the IT team (if there is one) handles only routine maintenance; management has never discussed AI in a structured way; there is no budget for technological innovation.
Level 2 — Exploratory
The company has begun to take an interest in AI. Someone on the team has done research or attended events on the topic. There may be isolated experiments, often driven by enthusiastic individuals rather than by an organisational strategy. Data is starting to be centralised, but without an overall plan.
Typical signs: someone in the company uses ChatGPT for individual tasks; an AI tool has been evaluated but not implemented; there is an ERP but the data isn’t clean; management is curious but hasn’t allocated a specific budget; people read articles and attend webinars on AI.
Level 3 — Defined
The organisation has launched AI pilot projects with defined, measurable objectives. There are first people with AI skills (internal or external consultants). Data is governed at least for the areas involved in the pilot projects. Management actively supports the initiatives and has allocated a dedicated budget.
Typical signs: a pilot project is under way (for example, a customer service chatbot or a demand forecasting system); a cross-functional team is working on the project; the data for the pilot area has been cleaned and structured; there are KPIs to measure the pilot’s success; the board or general management has approved a specific budget for AI.
Level 4 — Managed
AI is integrated into several business processes and produces measurable results. There is clear governance for AI projects, with defined processes for selection, implementation and monitoring. AI skills are spread across the organisation and not concentrated in a single team. Data is managed as a strategic asset with clear policies and responsibilities.
Typical signs: AI is active in at least 2-3 processes (production, sales, customer service); there is an AI Lead or a dedicated team; the ROI of AI projects is measured and reported periodically; AI training is included in the company training plan; technology suppliers are managed with defined SLAs and KPIs.
Level 5 — Optimised
AI is a fundamental component of the company’s strategy and pervades the entire organisation. AI systems learn and improve continuously. The organisation actively innovates, experiments with new applications and contributes to industry best practice. Data is a competitive asset, with automated data pipelines and advanced governance.
Typical signs: AI influences strategic decisions at every level; the company develops proprietary or customised AI models; AI innovation is a continuous process with an experimentation pipeline; data is integrated in real time from all sources; the company is recognised as an AI leader in its sector.
Important note: most Italian SMEs today sit between level 1 and level 2. This isn’t a problem: it is a starting point. The companies that are most successful with AI are not those that start from the highest level, but those that clearly understand where they stand and build a realistic, sustainable path towards the next level. The goal is not to jump from level 1 to level 5, but to progress one level at a time, consolidating each gain before aiming for the next.
How to carry out a self-assessment: a practical guide
Now that you know the dimensions and the levels, you can carry out a first assessment of your organisation. Here is a structured six-step process that you can start as early as tomorrow, without needing external consultants for the initial phase.
Put together the assessment team
Don’t do the assessment alone. Involve 4-6 key people representing different areas of the organisation: general management, IT, operations, sales and, if there is one, someone with responsibility for data. Diversity of perspectives is essential to get a realistic picture. A CEO assessing alone will tend to overestimate strategic maturity and underestimate operational gaps. An IT manager will do the opposite. Comparing different voices produces the most accurate assessment.
Gather the evidence for each dimension
For each of the 5 dimensions, answer the guiding questions described above. Don’t settle for generic answers: look for concrete evidence. Instead of saying "our data is good", check: how many data sources do we have? Are they integrated? When did someone last do a systematic clean-up? Document the answers with specific examples, screenshots, numbers. This phase typically takes 2-3 hours of individual work for each team member, followed by a joint 2-3 hour session to compare and consolidate the assessments.
Assign a level to each dimension
On the basis of the evidence gathered, place your organisation on one of the 5 levels for each dimension. It is normal for the levels to be uneven: a company might be at level 3 on data (thanks to a good ERP), but at level 1 on people (no AI skills in the team). This unevenness is informative: it shows where to intervene first. If team members disagree, don’t look for a compromise but dig deeper: disagreement often reveals blind spots that nobody had considered.
Create the radar map
Display the results on a radar chart with the 5 dimensions. This visual tool is extremely powerful for communicating the situation to management and the board. An unbalanced radar (for example, high on technology but low on people and processes) tells a clear story: the company has invested in tools but hasn’t prepared the organisation to use them. A uniformly low radar says you need to start from the foundations. A uniformly medium radar says it is time to accelerate selectively on the dimensions most critical to the business.
Identify the critical gaps
Not all gaps are equally urgent. Analyse the results looking for three types of critical gap. The first are blocking dependencies: dimensions whose low level prevents progress on the others (typically, data and people). The second are bottlenecks: dimensions that slow down overall progress (often processes). The third are strategic risks: gaps that expose the company to a loss of competitiveness relative to its market. For an Italian manufacturing SME, for example, the most critical gap could be people, if all your competitors are training their staff on AI and your company isn’t.
Define the target level
For each dimension, define the level you want to reach within 12-18 months. Be realistic: moving from level 1 to level 3 in a year is ambitious but feasible on a single dimension, provided you invest adequate resources. Trying to do it on all five dimensions at once is unrealistic and leads to a dispersion of energy. Concentrate on 2-3 priority dimensions, those where the gap is most critical relative to your business objectives.
Common mistakes in assessing AI maturity
Having accompanied dozens of Italian SMEs through the assessment process, we have identified recurring mistakes that compromise the usefulness of the assessment. Recognising them in advance makes it possible to avoid them.
Confusing technology with maturity
Having an account on an AI platform doesn’t mean being mature. A company that has bought a business intelligence tool with AI features that nobody uses is not at level 3: it is at level 2 with an expensive tool gathering digital dust. Maturity is measured by actual adoption and by the results produced, not by the purchase of licences.
Overestimating to please
When the assessment is carried out internally, there is a natural tendency to give higher marks than the real ones, especially when the results will be presented to the board. It is a serious mistake: an inflated assessment produces an inadequate roadmap, which in turn leads to failures that erode trust in AI. Better to start from a conservative assessment and discover you are further ahead than expected, than the opposite.
Ignoring the cultural dimension
Many assessments focus on data and technology, neglecting people, processes and strategy. It is a systematic error linked to a technocentric bias: we think AI is a technological problem when in reality it is an organisational one. Companies that fail with AI almost never fail for technical reasons. They fail because people aren’t prepared, processes aren’t ready or the strategy isn’t clear.
Doing the assessment only once
AI maturity is not static: it evolves (for better or for worse) over time. An assessment done in January can be obsolete by June if in the meantime the IT manager leaves, a pilot project is abandoned or the market changes radically. The assessment should be repeated at least every 6 months to monitor progress and recalibrate priorities.
Not involving middle management
An assessment carried out only by top management or only by IT produces a partial view. It is the department heads, team leaders and operational coordinators who really know the state of the processes, the quality of the data and the team’s openness to change. Their involvement not only improves the quality of the assessment, but also creates valuable allies for the implementation phase.
From assessment to action: how to build the roadmap
An assessment without an action plan is an academic exercise. The real usefulness of the assessment emerges when it translates into an operational roadmap with concrete actions, realistic timescales and clear responsibilities. Here is how to proceed.
Prioritise with the Impact-Feasibility framework
For each gap identified, assess two dimensions: the impact on the business (how much value closing it generates) and feasibility (how realistic it is to do with the resources available). Focus first on the high-impact, high-feasibility actions: they are the quick wins that generate visible results and build the organisational trust needed to tackle more ambitious projects. For example, for a retail company at level 2, a high-impact, high-feasibility action could be cleaning and integrating customer data between the CRM and e-commerce. It doesn’t require huge investment, but it lays the foundations for any future AI initiative in marketing and sales.
Define quarterly sprints
An AI roadmap shouldn’t be a monolithic three-year plan. Adopt an agile approach with quarterly sprints: each quarter has 2-3 specific, measurable objectives, with assigned resources and responsibilities. At the end of each sprint, review the results, update the assessment and recalibrate the priorities for the following quarter. This adaptive approach is particularly effective for Italian SMEs, which often operate in volatile markets where the ability to react quickly to change is a competitive advantage. Mixed teams, made up of technical and business people, ensure that each sprint produces concrete value and not just technical progress for its own sake.
Start from the foundations
If your assessment reveals that you are at level 1-2 on the Data and People dimensions, it makes no sense to invest in advanced AI tools. The first actions on the roadmap should focus on: centralising and cleaning the existing data; defining standards and responsibilities for data quality; basic AI training for management (not technical courses, but an understanding of the potential and the limits); identifying 1-2 candidate processes for a first pilot project; and building a preliminary business case for the pilot, with clear success metrics.
Build an AI-first organisational mindset
The roadmap mustn’t be limited to technical and project actions. It must include a path of cultural evolution that leads the organisation to develop an AI-first mindset. This means: creating regular moments of sharing where the team discusses the opportunities and challenges linked to AI; celebrating small successes to build enthusiasm and reduce fear of change; being transparent about what AI can and can’t do in order to manage expectations; involving operational staff from the earliest stages to create ownership and reduce resistance.
Measure progress
Define from the start how you will measure progress. Don’t limit yourself to technical metrics (models deployed, data integrated) but include business metrics: reduction in process times, increase in customer satisfaction, reduction in errors, impact on revenue. Business metrics are the ones that keep management’s support and justify subsequent investments. A manufacturing company implementing an AI predictive maintenance system should measure not only the accuracy of the model, but the number of machine stoppages avoided, the savings on maintenance costs and the increase in OEE (Overall Equipment Effectiveness).
The first step is always the most important
Assessing AI maturity is not a finishing line: it is a starting point. But it is a fundamental starting point, because without knowing where you are, any direction looks equivalent. And in the world of AI, where the possibilities are infinite but resources are finite, choosing the right direction makes the difference between investing with a return and burning budget without results.
Italian SMEs have an often underrated advantage: flexibility. Compared with large corporations, they can move faster, experiment with less bureaucracy and adopt an incremental approach that produces tangible results in a short time. But this advantage only comes into play if the path is guided by a clear understanding of your starting point.
Whatever your current level, the message is the same: start. An honest assessment carried out with method is worth more than a thousand slides on digitalisation. Take the framework described in this guide, gather your team and spend half a day understanding where you really are. You may discover that you are further ahead than you thought on some dimensions and further behind on others. Either way, you will have the information you need to act with awareness.