Most enterprises have adopted AI. Almost none have matured their AI capabilities enough to capture real value. The difference between the two is a five-stage journey built on a foundation most organizations haven’t even laid: continuous impact measurement.
What is AI maturity? AI maturity is an organization's ability to effectively deploy, measure, and scale artificial intelligence across business functions, progressing from initial experimentation through to enterprise-wide optimization with measurable business impact.
The adoption numbers look impressive. McKinsey’s 2025 State of AI report found that 88% of organizations are using AI in at least one business function, up from 78% the previous year. Gartner predicts that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from fewer than 5% in 2025. Investment is surging; KPMG’s Q4 2025 AI Pulse Survey reports that enterprises project deploying $124 million on AI annually, with 92% planning to increase AI budgets over the next three years.
But here’s the number that should alarm every executive: according to McKinsey’s report, only 1% of organizations consider their AI strategies mature. Not 10%. Not 5%. One percent.
McKinsey found that, while adoption is broad, it is not deep. Roughly two-thirds of respondents say their organizations have not yet begun scaling AI across the enterprise. Only 6% qualify as “high performers” capturing disproportionate value. The other 94% are using AI, but not transforming with it.
Gartner’s AI maturity survey tells the same story from a different angle. High-maturity organizations have dedicated AI leaders (91%), run financial analysis on AI initiatives (63%), and build trust that drives adoption. 57% of their business units trust AI solutions and are ready to use them, versus just 14% in low-maturity organizations.
The pattern is clear: AI adoption is table stakes. AI maturity is the differentiator. And most organizations don’t have a framework for understanding where they stand, let alone a roadmap for moving forward.
Figure 1. McKinsey reports that most organizations have used traditional AI
in at least one business function for years, with Gen AI catching up fast.
(Source: McKinsey & Co.)
Maturity models for AI aren’t new. Gartner publishes a five-level AI Maturity Model covering seven pillars: strategy, product portfolio, governance, engineering, data, operating models, and people and culture. McKinsey’s research identifies twelve scaling practices across six dimensions. Deloitte, Forrester, and dozens of consulting firms have their own versions.
These frameworks are useful, and often sophisticated. But they share three blind spots that limit their value for enterprise leaders navigating 2026 and beyond.
First, they’re organizational-level assessments that miss granular variation. An enterprise doesn’t have a single maturity level. Your engineering team might be at Stage 4 while your finance team is at Stage 1. Your London office might be ahead of your New York office. Your product management function might be deploying agentic workflows while your HR team is still solely using ChatGPT as a search engine. A single organizational score obscures the pockets of excellence and the pockets of stagnation that leaders actually need to see.
Effective AI maturity assessment measures by team, by department, by function, and by location, surfacing the variance, not just the average.
Second, most maturity models are input-focused, not outcome-focused. They assess whether you have a strategy, a governance framework, a data pipeline, and a dedicated AI leader, and measure how many employees have logins to LLMs and other AI-powered tools. These are necessary inputs, but they don’t tell you whether AI is actually doing anything. An organization can score perfectly on strategy and governance readiness while delivering zero business impact. The model should measure what’s happening, not what’s possible or what’s planned.
Third, they don’t account for agentic AI. Most existing maturity models were designed for the era of AI assistants, which are tools that help humans work faster. The frontier has moved. The defining question of AI maturity in 2026 is no longer “are your people using AI tools?” It’s “is AI doing independent work?” Maturity models that don’t distinguish between AI as an assistant and AI as an autonomous agent are measuring against yesterday’s standard.
Before walking through the five stages, there’s a critical architectural point that separates this model from every other maturity framework: impact measurement is not a stage you reach at the end. It’s the foundation that runs underneath every stage.
Most maturity models position measurement as something you do after you’ve deployed AI; a retrospective analysis of whether things worked. This is backward. The most important measurement you can take is the status quo ante: what things are like before implementation starts. You can’t know how far you’ve come if you don’t know where you started.
You can’t know if your visibility efforts at Stage 1 are working without measuring. You can’t know if adoption is meaningful at Stage 2 without measuring. You can’t know if your proficiency programs are moving the needle at Stage 3 without measuring. You can’t evaluate workflow intelligence or agentic deployment without measurement running continuously underneath everything.
The measurement framework: tracking impact across five dimensions (effectiveness, quality, time, revenue, and cost); is the operating system of AI maturity. It’s what makes every stage actionable rather than aspirational. Without it, you’re navigating blind at every stage, not just the last one.
This means that, even at Stage 1, organizations should be establishing measurement infrastructure. Not the full five-dimension suite, which evolves as maturity increases, but the baseline telemetry that lets you know whether you’re making progress. At each subsequent stage, the measurement layer deepens and expands, adding new metrics and new dimensions as the organization’s AI capabilities become more sophisticated.
The five dimensions of impact measurement, as detailed in the Measuring AI Impact guide, are:
Each stage of maturity emphasizes different dimensions. Stage 1 focuses primarily on cost (what are we spending?) and basic effectiveness (is anyone using this?). Stage 3 adds quality and time metrics as proficiency programs take effect. Stage 5 requires the full suite, including revenue attribution and financial translation.
The key principle: measurement is not something you earn the right to do at the end. It’s something you must do from the beginning, and deepen as you progress.
Larridin’s AI Maturity Model is designed to be measured at every organizational level: team, department, function, and location, so leaders see the real distribution of maturity across their enterprise, not a single blended score. And it accounts for the full spectrum of AI capability, from basic tool usage through truly agentic deployment.
Organizations can’t skip stages, but they can accelerate through them. And critically, different parts of the same organization will be at different stages simultaneously.
The question this stage answers: “Do we even know what’s happening?”
This is the foundation. Before you can measure adoption, improve proficiency, or deploy AI agents, you need to know what AI tools exist in your organization, who’s using them, and whether basic controls are in place.
Most organizations think they’re past this stage. They’re not.
Larridin’s research across 350 finance and IT leaders found that 83% report shadow AI adoption growing faster than IT can track, and 84% discover more AI tools than expected during audits. When an organization can’t even inventory its AI landscape, everything else: measurement, optimization, governance; is built on sand.
Stage 1 maturity means:
Most enterprises are at Stage 1, or have partly but not fully completed it. The most common mistake is assuming that deploying Copilot licenses or approving a ChatGPT Enterprise subscription, with accompanying vendor dashboards, means you have visibility. It doesn’t. Deployment is not visibility. Until you can see the full AI landscape, including the tools employees brought in themselves, you’re operating blind.
What measurement looks like at Stage 1: Basic cost tracking (what are we spending on AI tools?), tool inventory completeness metrics, shadow AI discovery rates, governance coverage percentage. The measurement layer is thin here, but it exists, and it’s what tells you whether Stage 1 is actually complete.
The diagnostic question: If someone asked you right now to list every AI tool in use across your organization, could you do it with confidence? If the answer is no, you’re still working on Stage 1.
The question this stage answers: “Who’s using AI, how often, and where are the gaps?”
Stage 2 moves from “what tools exist?” to “what’s happening with them?” This is where organizations begin to understand adoption patterns at a granular level — and where they discover that having tools deployed is very different from having tools used.
The data makes the case. McKinsey’s 2025 State of AI report found that, while 88% of organizations use AI, only about one-third have begun scaling AI programs organization-wide. Most are still in experimentation or pilot mode. The gap between “we have AI tools” and “our people are using AI tools meaningfully” is enormous.
Stage 2 maturity means:
The critical insight at Stage 2 is that adoption data, when properly segmented, immediately reveals where to focus. If your sales team has 20% adoption while marketing has 80%, that’s a departmental signal. If your London office has double the adoption of New York, that’s a geographic signal. If senior leaders have lower adoption than individual contributors, that’s a cultural signal. The measurement foundation, running underneath, turns adoption data into actionable intelligence.
What measurement looks like at Stage 2: Adoption rates by segment, usage frequency distributions, tool portfolio analysis, time series trends showing adoption trajectory. The measurement layer deepens; you’re now tracking not just what exists (Stage 1), but what’s being used, and how.
The diagnostic question: Can you produce a dashboard showing AI adoption rates by team, department, and location, with trend lines, right now? If not, you’re still working on Stage 2.
The question this stage answers: “Are people using AI well, and are we actively making them better?”
Stage 3 is where the model shifts from observation to intervention. Stages 1 and 2 are about seeing what’s happening. Stage 3 is about changing what’s happening; specifically, driving the proficiency of your workforce in using AI effectively.
The proficiency gap is the biggest untold story in enterprise AI. OpenAI’s 2025 State of Enterprise AI report revealed a 6x engagement gap between AI power users and typical employees. Meta reported a 30% average improvement in output from AI tools, but an 80% improvement among power users. EY’s 2025 Work Reimagined Survey found that 88% of employees use AI daily, but only 5% use it in advanced ways. Having AI and being good at AI are fundamentally different things.
Stage 3 maturity means:
The measurement layer at Stage 3 adds the quality dimension in force. As employees become more proficient, you should see the 37% AI Tax declining: less rework, fewer corrections, and higher-quality, AI-assisted output. If proficiency scores are rising, but rework rates aren’t falling, your proficiency measurement is miscalibrated. Quality metrics are the truth check on proficiency claims.
What measurement looks like at Stage 3: Proficiency score distributions by segment, proficiency trajectory trends, rework and edit rates (the quality dimension), correlation between proficiency scores and business outcomes, enablement program effectiveness metrics. The measurement layer is now substantial, tracking not just activity (Stage 2) but skill development and its impact on output quality.
The diagnostic question: Can you identify your top 10% of AI power users right now? That is, not by self-report, but by actual behavioral data? Can you show that their business outcomes are measurably better than the bottom 50%? If not, you’re still working on Stage 3.
The question this stage answers: “Where should we actually deploy AI, and how does work really get done?”
This is where most organizations stall, and it’s the stage that separates enterprises that get transformational value from AI from those that get incremental improvement. Stage 4 requires something that sounds simple, but is profoundly difficult at enterprise scale: understanding how work actually happens.
There are a handful of use cases where AI deployment is obvious. Software engineering (code generation, PR review). Customer service (chatbots, resolution automation). Content creation (drafting, editing). These are the low-hanging fruit, and most organizations have picked them already.
But enterprise work is far more nuanced than these obvious cases. The CIO who’s deployed Copilot to 10,000 employees and automated customer service with a chatbot has captured maybe 10% of the available value. The other 90% lives in the thousands of workflows: across sales, finance, legal, operations, marketing, HR, and product management; where the path to AI deployment is unclear.
Based on conversations with nearly 50 CIOs, the consistent finding is the same: the biggest barrier to AI value isn’t the technology. It’s that leaders don’t know where to deploy it. Beyond the obvious use cases, organizations struggle to identify which workflows can be transformed, which tasks should be automated, and where AI can create the most impact.
The typical approach is to ask: “Where can we deploy AI?” This is the wrong question. It starts from the current state and tries to bolt AI onto existing processes. It produces incremental improvements at best.
The right question is: “Why are humans doing this work?”
Start from AI-first. The default assumption should be that every workflow is a candidate for AI execution. Then work backward: where does human judgment, creativity, empathy, or domain expertise add value that AI cannot replicate? The humans should be doing that work. Everything else is a deployment target.
This mental model, which is AI-first by default, with human involvement by exception, inverts the typical approach. Instead of looking for places to add AI, you’re looking for reasons to keep humans in the loop. It’s a dramatically more powerful way to identify high-value AI opportunities, because it surfaces candidates for improvement that the traditional approach would never consider.
But executing this approach requires deep visibility into how work actually flows through your organization. Not how it’s supposed to flow according to the org chart and the process documentation, but how it actually happens day to day.
Stage 4 maturity means:
The process mining market is growing at 45.5% CAGR, projected to reach $3.4 billion in 2026 and $15.1 billion by 2029. Companies like Celonis have built multi-billion-dollar businesses around understanding how work flows through enterprises. But most organizations still have limited visibility into their own workflows, especially the knowledge work that happens across emails, documents, meetings, and AI tools rather than in structured ERP systems.
This is where the earlier stages become prerequisites. Stage 2 and 3 data, including adoption patterns, proficiency levels, and which tools people reach for in which contexts, provides a unique window into how work actually gets done. When you can see which AI tools people use in which workflows, how they move between tools, and where they spend their time, you have raw material for workflow intelligence that traditional process mining can’t capture.
What measurement looks like at Stage 4: Workflow completion times (the time dimension in force), process efficiency metrics, AI deployment opportunity scores, and readiness assessments for agentic deployment. The revenue dimension begins to emerge as you can start correlating AI-augmented workflows with business outcomes.
The diagnostic question: If a new CIO started tomorrow and asked, “Show me the top 20 workflows where AI would create the most value, and why,” could your organization produce that analysis: based on data, not opinions? If not, you’re still working on Stage 4.
The question this stage answers: “Is AI actually doing the work, independently?”
Stage 5 is the frontier, and it’s where the definition of AI maturity changes fundamentally. The first four stages are about humans using AI as a tool, understanding their workflows, and improving them. Stage 5 is about AI operating as an independent worker.
The distinction matters enormously. There’s a clear boundary between AI as assistant and AI as agent, and the boundary isn’t about sophistication of prompts or quality of output. It’s about two things: duration of autonomous operation and whether actions are being taken.
If a user asks Claude to draft an email and then reviews and sends it, that’s AI assistance. The human is in the loop at every step. The AI produced information; the human took action.
If an AI agent independently monitors customer support tickets, identifies escalation patterns, reroutes tickets to specialized agents, drafts and sends follow-up communications, updates the CRM, and runs for six hours without human intervention, that’s agentic AI. The AI isn’t just producing information. It’s taking actions. It’s executing a workflow end to end. The human oversees at decision points, not at every step. Humans improve workflows that machines execute.
A useful heuristic: if the agent runs independently for 2, 4, 6, 8 hours, then you’ve crossed the boundary from assistant to worker. And if the agent is taking real actions during that time. It's not just researching and drafting pieces humans review and send, but executing the communications loop. You’re in agentic territory.
Stage 5 maturity means:
The data on agentic adoption is early, but accelerating. McKinsey’s 2025 State of AI report found that 62% of organizations are at least experimenting with AI agents, and 23% report scaling agents somewhere in their enterprise; however, most are only doing so in one or two functions. Gartner’s prediction that 40% of enterprise applications will feature task-specific AI agents by the end of 2026 suggests rapid acceleration is coming.
The organizations that reach Stage 5 first will have a significant and potentially durable competitive advantage. An enterprise with autonomous AI agents handling routine workflows frees human talent for judgment-intensive, strategic, and creative work; the work where humans actually add irreplaceable value. That’s not an incremental improvement; it’s a structural advantage.
What measurement looks like at Stage 5: The full five-dimension measurement framework is in operation. Agent autonomy metrics (duration, action counts, exception rates). Workflow efficiency before and after agentic deployment. The financial translation layer is fully active: Capacity Reallocation Value, Cost of Delay savings, risk mitigation value. The measurement system doesn’t just report what happened; it directs where to invest next.
The diagnostic question: Do you have any AI agents that run independently for more than two hours, taking real actions in production workflows without human involvement at every step? If not, you haven’t entered Stage 5.
The most important design principle of this maturity model is that it’s not measured at the organization level alone. An enterprise-level maturity score is useful for board reporting, but it’s too blunt for decision-making.
Real maturity assessment happens at four levels simultaneously:
By team. A product management team of 12 people might be at Stage 4 (workflow intelligence) while the legal team down the hall is at Stage 1 (still discovering what AI tools people are using). Team-level maturity data tells managers where to focus enablement, training, and tool deployment.
By department. Engineering as a whole might be at Stage 3 (proficiency development) while Sales is at Stage 1. Department-level data tells VPs and SVPs where their organizations stand relative to the rest of the enterprise, and relative to industry benchmarks.
By function. “Marketing” isn’t monolithic. Content marketing might be at Stage 5 (deploying AI agents to handle production workflows) while brand strategy is at Stage 1. Function-level granularity reveals the real distribution of maturity within departments that often look uniform from the outside.
By location. Global enterprises often find significant geographic variation. An innovation hub in Tel Aviv might be at Stage 4 while a regional office in a different market is at Stage 1. Location-level data surfaces cultural, regulatory, and infrastructure factors that affect maturity, and helps leaders tailor their AI strategy to local realities.
The variance across these dimensions is itself a valuable metric. An organization where every team is uniformly at Stage 2 has a very different challenge than one where some teams are at Stage 5 and others at Stage 1. The first needs a broad push forward. The second needs to understand what’s working in the advanced teams and accelerate knowledge transfer.
For each of the five stages, rate your organization (or team, or department) on a scale of 1-5:
1 = Not started | 2 = Early progress | 3 = Partially complete | 4 = Mostly complete | 5 = Fully achieved
Scoring:
This is the self-reported version. Self-assessment is a useful starting point, but it’s inherently limited; — people tend to overestimate their readiness and underestimate their blind spots. Behavioral data from actual AI usage patterns tells a more accurate story than any survey.
The most common and most expensive mistake. An organization that deploys AI agents (Stage 5) without understanding how work flows through the organization (Stage 4) will automate the wrong things. An organization that tries to build workflow intelligence (Stage 4) without driving proficiency (Stage 3) doesn’t have the human capital to identify what’s possible. Each stage produces the knowledge and capability needed for the next stage.
If you wait until Stage 5 to build measurement capability, you’ve wasted every stage before it. You have no baselines. You have no trend data. You can’t prove that your adoption programs, proficiency investments, or workflow analysis produced results. Measurement should run from Day One; it’s the foundation, not the finish line.
An organization-level maturity score of “Stage 2” means nothing if it masks a distribution where engineering is at Stage 4 and half the company is at Stage 1. Measuring only at the enterprise level prevents leaders from seeing where the real opportunities and blockers are. Always measure at team, department, function, and location levels.
Buying Copilot licenses for 10,000 employees doesn’t make you a Stage 2 organization. Deploying a customer service chatbot doesn’t make you Stage 5. Maturity is about what’s happening with the tools, not whether the tools exist.
The traditional approach to AI deployment starts with existing processes and asks where AI can help. This produces incremental improvement. The maturity-accelerating approach starts from AI-first and asks where human involvement is truly necessary. This produces transformational change. Every workflow is a candidate for AI execution until you can articulate why a human must be involved.
Stage 5 doesn’t work without Stage 3. Agentic AI requires people who can design, deploy, orchestrate, and oversee autonomous systems. Those people need to be at Level 4 or 5 on the proficiency spectrum: Power Users and AI-Native Orchestrators. If your workforce is stuck at Level 1 or 2, you don’t have the human capital to make agentic deployment successful. Proficiency development isn’t optional; it’s a structural prerequisite.
AI maturity isn’t a fixed destination. The landscape evolves with new tools, new capabilities, and new possibilities. An organization that was at Stage 4 six months ago might effectively be at Stage 3 if it hasn’t kept pace with new agentic capabilities. Maturity assessment should be continuous, not annual.
Whatever stage you’re at, the path forward follows the same structure: assess, prioritize, execute, measure.
At the end of 90 days, you should be able to present something no existing maturity model enables:
“Here’s where every team in our organization stands on the AI maturity spectrum, measured by behavioral data, not surveys. Our engineering teams are at Stage 3, with rising proficiency scores and declining rework rates. Sales is at Stage 2, with strong adoption, but showing proficiency gaps that we’re addressing through targeted enablement. Our Singapore office is ahead of our London office by one full stage, and we’re investigating why. Three workflows in customer operations are ready for agentic deployment, with projected Capacity Reallocation Value of $X. Our measurement infrastructure is tracking all five impact dimensions continuously, and here’s what the trend lines show.”
That’s a narrative built on data, segmented by reality, and actionable at every level. It’s a fundamentally different conversation than “our AI maturity score is 3.2 out of 5.”
Larridin measures AI proficiency across nine dimensions, recalibrated every 30 days, giving enterprises a real-time view of how effectively their workforce uses AI—and exactly where to invest to move the needle.
Learn how Larridin measures AI proficiency
The five stages of AI maturity are: (1) Awareness; recognizing AI's potential; (2) Experimentation; piloting AI tools in isolated use cases; (3) Integration; embedding AI into core workflows; (4) Optimization; measuring and improving AI's business impact; and (5) Transformation; AI fundamentally reshaping business models and competitive advantage.
Measure AI maturity across four dimensions: adoption breadth (what percentage of employees use AI tools), fluency depth (how effectively they use them), workflow integration (whether AI is embedded in core processes), and impact measurement (whether you can quantify business outcomes). Tools like Larridin provide this visibility across all four dimensions.
Most initiatives stall at the experimentation stage because organizations lack measurement infrastructure. Without data on what's working, which teams are adopting, and where impact is occurring, leadership can't make informed decisions about scaling. The result: perpetual piloting with no path to enterprise-wide value.
To systematically identify where AI fits into your existing processes, explore our AI Workflow Mapping methodology.