AI is a Change Agent, Not a Tool

Who This Paper Is For

This paper is written for CEOs, boards, senior leadership teams, and business and technology leaders in organizations that recognize AI’s potential but are struggling to move from scattered experimentation to repeatable business value.

It is intended for organizations that have moved beyond asking whether they should adopt AI and are now asking how to build the strategy, capabilities, governance, human skills, and operating model needed to compete in an AI-enabled economy.

The framework applies across organizations of different sizes and stages of maturity. The scale of the governance structure and investment will vary, but the underlying disciplines remain the same: clear strategic ownership, grassroots fluency, deliberate people change management, trusted enterprise context, and accountability for business outcomes.

Why I Wrote This

I have spent much of my career leading and governing enterprise transformations across multiple companies and through several board roles. As the business leader accountable for the outcome—not simply the technology—I oversaw three major ERP implementations using JD Edwards, SAP, and Microsoft Dynamics 365, each a multi-year, multimillion-dollar undertaking. I also led CRM, content-management, and service-ticketing implementations, including Salesforce, and have observed similar transformation patterns through extensive partner and board engagements.

Those experiences repeatedly demonstrated the lesson that shapes this paper: enterprise transformation succeeds or fails on strategy, organizational fluency, and the human work of change—not on the software itself. The technology is rarely the hardest part.

AI is the largest version of that challenge I have encountered. In many respects, AI adoption is the mother of all ERP implementations: broader in scope, faster moving, and far more dependent on human judgment than any system rollout before it.

The framework in this paper draws on those cumulative operating and governance experiences. The case study presented later is one practical illustration of the framework in action, not the sole basis for it.

Lynn Smurthwaite-Murphy CEO, Plugable Technologies

From Data Overload to Decision Advantage

Organizations have moved from being starved for information to drowning in fragmented data spread across systems, documents, reports, teams, and external sources. The information exists, but reaching a trusted conclusion can still require analysts to reconcile definitions, combine disconnected sources, and repeatedly rebuild the answer as new questions emerge.

AI creates the possibility of a different model: the ability to chat with trusted data and move more quickly from observation to understanding and action.

But the most valuable context about an organization is rarely available to a general-purpose AI model. It lives across internal systems, private databases, documents, local devices, customer interactions, and the institutional knowledge of employees. The opportunity is therefore not simply to move more data into the cloud. It is to connect the intelligence available in the cloud with trusted enterprise data wherever it resides—or bring AI securely to the data when that data cannot or should not move.

Instead of receiving a static answer, leaders can begin a conversation:

• What happened?
• Why did it happen?
• What else contributed?
• Is this part of a larger pattern?
• What should we do next?

In effect, they can apply the five whys to business performance—combining trusted information from inside and outside the organization, challenging assumptions, and moving more quickly from observation to insight and action.

This does not mean AI automatically produces truth. Credible answers still require trusted data, shared definitions, transparent sources, appropriate governance, and human judgment. But when those foundations are present, AI can dramatically reduce the distance between information, understanding, and action.

An organization that develops this capability can identify customer needs earlier, improve service, develop better products, detect risks, and make decisions while the opportunity still exists. If its competitors can do that and it cannot, the consequences will extend far beyond productivity.

We moved from being starved for information to drowning in data. The next advantage belongs to organizations that can chat with trusted data—asking not only what happened, but why, what else contributed, and what actions to consider. The goal is to turn fragmented information into credible insight, and credible insight into timely action.

Executive Summary

Artificial intelligence is reshaping how organizations compete, create value, and make decisions. Like every major technology disruption before it, from industrial automation to personal computing to the internet, AI is creating both extraordinary opportunity and significant uncertainty. New technologies rarely fit neatly into existing ways of working. They create a period of disruption and experimentation before organizations establish a new operating model. AI is no different in this respect.

Every major technology cycle has forced organizations to rethink not only the tools they use, but how work gets done, how decisions are made, and what skills leaders and employees need. Those that intentionally redesigned how they operated emerged stronger. Those that simply layered new technology onto old ways of working often struggled to realize its full value.

AI presents a similar challenge today. It democratizes capabilities once limited to specialists, allowing organizations of every size to research faster, analyze more deeply, automate routine work, and make better-informed decisions. At the same time, it raises understandable concerns around privacy, governance, ethics, accuracy, and the future of work. Employees may wonder whether AI will threaten their roles, whether they are permitted to use it, and whether its output can be trusted. Leaders face equally real concerns about security, confidentiality, accountability, and how to govern a technology evolving faster than traditional planning cycles. These concerns should not be dismissed. They are a natural part of every significant transformation.

Many organizations recognize AI's potential, yet few have progressed beyond scattered experimentation. Employees are testing new tools, leadership teams are discussing transformation, and AI capabilities keep advancing, but the way work actually gets done often remains unchanged. The technology is moving faster than organizations know how to absorb it. The challenge is not a lack of AI. It is a lack of an adoption framework.

The organizations that learn fastest, not simply those that adopt AI first, will increasingly define the next generation of competitive advantage. Microsoft's 2025 Work Trend Index describes the emergence of “Frontier Firms” that rethink how work gets done by combining human judgment with AI to expand capacity, accelerate learning, and improve decision-making. The advantage is no longer simply adopting AI. It is learning how to redesign work around it.

Organizations frequently approach AI from one of two extremes. Some treat it as another software rollout, focused on deploying tools. Others see it as an overwhelming revolution that feels too broad to know where to begin. Both stall progress, because neither addresses the organizational, operational, and human changes required to translate AI into lasting value.

This paper argues that successful AI adoption is not primarily a technology initiative. It is an organizational transformation that requires three capabilities to develop together:

• Strategic direction connects AI to business outcomes, governance, workflow priorities, and enterprise context.
• Grassroots adoption builds organizational fluency through safe experimentation, shared learning, and the identification of repeatable work that can be redesigned.
• People change management develops the human and managerial skills required to work effectively with AI: judgment, delegation, prioritization, accountability, and decision-making.

These three capabilities are united by one central principle:

Context is the bridge between AI activity and AI value.

Organizations create value not simply by deploying more capable AI models, but by continuously improving the quality of the business context, trusted knowledge, governance, and human judgment that surround them. Context is what transforms AI from a productivity tool into an organizational capability.

The framework that follows provides a practical roadmap for moving from individual experimentation to repeatable organizational capability. It also turns AI-driven activity into sustainable gains in productivity, decision quality, innovation, customer experience, and competitive advantage.

The framework reflects lessons from multiple enterprise transformations and board-level observations and is illustrated through an operating case study later in the paper. In that example, early behavioural metrics evolved into workflow and business measures, while the organization materially outperformed its closest comparable market. The example does not establish AI as the sole cause of performance, but it shows how the capabilities described in this framework can translate into greater organizational speed and adaptability.

Why AI Matters to Every Organization

AI is often introduced as a technology initiative, but leaders rarely invest in technology for its own sake. They invest because it improves business outcomes. AI matters because it changes what individuals, teams, and organizations can accomplish. It reduces the cost of knowledge work, accelerates learning, improves the quality of decisions, and lets organizations redesign how work gets done.

Most importantly, AI changes the economics of expertise. Work that once depended on specialists, consultants, or analysts increasingly becomes accessible to every employee. Research becomes conversational, writing becomes collaborative, and analysis becomes interactive. Rather than waiting for information to flow through organizational layers, employees can explore, test, compare, and learn directly. AI does not eliminate expertise; it democratizes access to it. This is why AI is both an extraordinary opportunity and a competitive threat for organizations of every size.

Unlike earlier waves of automation that targeted repetitive manual tasks, AI augments knowledge work: it helps people research faster, write better first drafts, analyze more deeply, compare alternatives, generate ideas, identify patterns, and make more informed decisions. The organizations that gain the most will not simply use AI to work faster; they will use it to make better decisions, redesign workflows, and create capacity.

These outcomes explain why AI has become both an opportunity and a competitive necessity. The question is no longer whether organizations should adopt AI, but how they can do so responsibly while creating lasting value.

From Dashboards to Conversations: The Power of Chatting with Your Data

Chatting with your data turns analysis from a one-time deliverable into an iterative inquiry. A leader can begin with a business question, examine the answer, challenge its assumptions, and continue probing until the underlying drivers and possible actions become clearer.

Rather than waiting for a new report each time another question emerges, leaders can apply successive “why” questions to business performance.

What happened?

Why did our margins decline last quarter?

Why did it happen?

Which products, customers, cost changes, or channel shifts contributed most?

Is this part of a larger pattern?

Has the same change appeared in other periods, regions, products, or customer groups?

What are the implications?

Which customers or product lines have become less profitable, and why?

What could we do next?

What might happen if we changed pricing, product mix, service levels, inventory decisions, or purchasing strategy?

This does not eliminate the need for analysts. It changes their role. They spend less time producing routine reports and more time helping the organization ask better questions, validate assumptions, interpret complex findings, and support higher-value decisions.

Chatting with your data changes analysis from static to iterative. As AI is connected to trusted internal knowledge, approved data, documented workflows, and business definitions, its value expands—but so does the importance of permissions, source quality, governance, and human judgment. An organization cannot have a reliable conversation with data it does not understand or trust.

1. AI Is a Change Agent

AI is not just a technology cycle. It is a change agent. A technology cycle changes the tools people use; a change agent changes how work gets done, how decisions are made, and who has access to capability. That distinction is why AI matters beyond productivity. It is not about doing the same work faster, but about changing what is possible and for whom.

That cuts both ways. The same capability that lets a smaller organization research, draft, analyze, and prototype like a much larger one also reaches its competitors and its customers. It can erode the advantage of size, legacy expertise, or market position as readily as it can extend it.

This is why doing nothing is not neutral. If competitors are learning, producing, and deciding faster, a company that waits for certainty may fall behind even while believing it is being prudent. The pattern holds across every technology era before this one: the tools change, the cycle repeats. The real question is not whether to adopt AI, but how to use it to create capacity, improve decisions, learn faster, and redesign the work that matters most.

2. Why AI Adoption Stalls

AI adoption often stalls because organizations misread what kind of change they are facing.

Some treat AI too narrowly, as if it were another software rollout. They buy a tool, issue licenses, run a training session, and assume adoption will follow. But AI does not sit inside a predefined workflow. It can change the workflow itself.

The deeper error in the software-rollout mindset is one of scale. AI is not smaller than a major enterprise system implementation. It is larger and more encompassing.

AI adoption is the mother of all ERP implementations.

An ERP project reshapes processes and data. AI reshapes judgment, decisions, and who has access to capability. It demands the same disciplines that determine the success of any large transformation, clear strategy, genuine fluency, and deliberate change management, only at a broader scope and a faster pace. Approached as a tool, it stalls. Approached as a transformation, it compounds.

Others treat it too broadly, as a panacea. They talk about transformation, disruption, automation, and agents, but the conversation stays abstract. Employees do not know what to do on Monday morning, and leaders struggle to connect the vision to practical priorities.

Fear adds another layer. Employees worry about whether AI will replace them, whether they are allowed to use it, whether their data is safe, and whether outputs can be trusted. Leaders worry about privacy, ethics, compliance, confidentiality, brand risk, and misuse.

The result is a strange combination of overuse, underuse, and hidden use. Some experiment without guardrails, some avoid AI altogether, and some use it quietly because they are unsure whether it is approved, often producing low-quality output because they treat AI as a shortcut rather than a collaborator that requires review.

3. The Three-Pronged Adoption Model, Anchored by Context

Successful AI adoption requires strategy, grassroots adoption, and people change management. These are not sequential phases where one must finish before the next begins. They are parallel disciplines that reinforce each other.

Strategy provides context. Grassroots adoption builds context fluency. People change management applies judgment to context.

None of the three works on its own. Strategy without grassroots adoption is a top-down plan no one has the fluency to execute. Grassroots adoption without strategy is just scattered experimentation. And without people-change management, both simply produce more output, with no guarantee that judgment, quality, or decisions actually improve. The goal is to make the three disciplines reinforce one another and turn scattered AI activity into governed organizational capability.

4. Context Is the Bridge Between AI Activity and AI Value

AI creates productive chaos: it can generate more ideas, drafts, summaries, analysis, and options than organizations are used to absorbing. The way to control that chaos is not to slow everything down or wait for perfect certainty. It is to improve context.

Context is what turns AI from generic output into useful work. It tells the system what matters, what data to use, what constraints apply, what decision is being made, what audience is being served, and what quality standard must be met. This principle is widely recognized across the AI industry: frontier developers consistently emphasize that model performance depends not only on the model itself, but on the quality of the instructions, examples, business context, and trusted knowledge provided to it. Better models reduce effort; better context improves decisions. Think of this as onboarding a new hire. The same principle applies when an organization begins to chat with its data: the conversation is only as credible as the context, definitions, sources, and judgment behind the answers.

Sources: Anthropic Prompt Engineering Guide; OpenAI Prompting Best Practices.

There is a useful precedent for this in enterprise systems. Every well-run ERP implementation begins with a blueprint, whether it is called a business blueprint or a solution blueprint review, a detailed description of the organization's processes, master data, structures, and requirements. It is the agreed-upon picture of how the business will operate, and nothing gets built until it is right.

AI needs the same discipline, but the scope is larger.

ERP blueprints redesigned processes around a new system. AI requires us to redesign the relationship between people, information, judgment, and technology.

The deterministic logic is still there, the rules and data, but the blueprint now has to capture the qualitative layer ERP never had to write down: the judgment, standards, and institutional knowledge experienced people carry in their heads, entered in plain language rather than code. That is what the four context layers above are, the blueprint for AI, extended to include how good people think.

That shift is more consequential than it first appears. With an ERP, the business was expected to own the transformation, but ownership was structurally compromised by the technology itself. The system was so deeply technical that intent had to pass through a translation layer: the business explained what it wanted, IT rendered it into the system, and the results were translated back. Meaning was lost in both directions, and the business ended up accountable for an outcome it could not directly steer. AI removes that translation layer. Because you direct it in plain language, the business can express the outcome, the context, and the intent itself, without an interpreter in between. For the first time in a transformation of this scale, the business can genuinely drive its own destiny. The very accessibility that makes AI feel overwhelming is what finally puts the business in the driver's seat.

This is why context belongs at the center of AI adoption. Better context produces better output. Better output builds trust. Trust encourages adoption. Adoption reveals repeated work patterns. Those patterns become redesigned workflows, and redesigned workflows become organizational capability.

The difference between weak and useful AI output is often not the tool; it is the clarity of the context. Better context does not simply mean giving AI more information. It means giving AI the right purpose, audience, constraints, source material, decision frame, and definition of quality.

5. Prong One: Strategy

Strategy is how leadership decides what context AI should have access to, and why. This includes trusted data, business definitions, workflows, permissions, policies, customer knowledge, financial information, operating history, and strategic priorities. Without enterprise context, AI remains generic; with trusted context, it becomes business-aware.

This is why data readiness is a strategic requirement, not just a technical one. The organization is deciding what AI is allowed to know, what it should ignore, what it must protect, and what standards it must follow. A strong AI strategy answers six questions.

1. What outcomes are we trying to improve?

This is directly tied to the organization’s strategic goals. AI can support faster execution, better decisions, better customer experience, lower cost to serve, revenue growth, improved retention, faster innovation, better knowledge management, and stronger employee capability. An organization need not choose only one, but it must know which outcomes drive investment, governance, prioritization, and measurement. The alternative is spreading resources too thin to quickly gain traction.

2. What data and context are required and in what order of priority?

Enterprise value grows when AI can safely use trusted internal context: customer history, product knowledge, support data, financial information, policies, workflows, and institutional knowledge. Because building that foundation requires effort and investment, leadership should prioritize the data and knowledge needed for the highest-value use cases first.

3. What workflows should change?

The goal is not more tools; it is better work. AI should eventually be integrated into recurring processes: research, support, marketing, finance, product, sales, and management workflows.

4. What guardrails are required?

Organizations need clear rules for privacy, confidentiality, security, intellectual property, source verification, human review, and accountability. Good governance should not simply block AI use. It should make safe use easier.

5. How will value be measured?

AI activity is not the same as AI value. Better measures include time saved, quality improved, cycle time reduced, decisions improved, workflows standardized, customer response improved, employee capacity created, risk reduced, and enterprise outcomes improved.

Measures should also mature with adoption. Early on, the useful signals are behavioural (who is experimenting, testing, and sharing) because leadership first needs to know whether fluency is taking hold. As adoption matures, the measures shift to workflow improvement and business value: which uses repeat, which tasks can be redesigned, and where AI is creating measurable capacity or better decisions. In short, measure what you need to learn at each stage of maturity.

This view is increasingly shared beyond the framework. In its work on the economics of AI agents, McKinsey argues that the executive task is not to minimize token cost but to govern machine work: to allocate AI capability like capital and to judge agents by business outcomes rather than technical metrics.

6. Who owns the Transformation?

Strategy also has to answer a question many organizations leave dangerously vague: who owns this? In the enterprise transformations that most resemble AI adoption, ownership follows a consistent pattern, and it is worth copying.

The CEO sets direction. Because AI adoption is tied directly to business strategy, it begins with the chief executive, who frames why it matters, what it is expected to change, and how it connects to where the organization is going. This cannot be delegated at the outset without signaling that AI is a technology project rather than a business priority.

A C-suite sponsor owns the program. The CEO then assigns a senior executive to own AI adoption end to end: the budget, the roadmap, the priorities, and the accountability for outcomes. This person keeps the effort moving, resolves cross-functional conflicts, and reports on progress. Without a single accountable owner, adoption fragments into disconnected departmental experiments. This is a standing responsibility, not a launch-day one. The sponsor convenes the ongoing forum where priorities are reset and resources are realigned as the work evolves, the cadence described in Section 11.

The reason this role cannot be divided is that its real job is to hold three things together that otherwise drift apart: the strategy the CEO set, the budget that funds it, and the business value it is meant to produce. When those responsibilities are split across finance, IT, and the business, each may be managed well on its own while the connection between them is owned by no one. That gap, not the technology, is where large programs quietly fail. A single accountable sponsor exists to keep strategy, spend, and value tied to one another as the work evolves.

A cross-functional team does the work. Under the sponsor sits a team that deliberately spans functions, because AI adoption is not one department's project. It includes IT, which enables platforms, security, and integration; the business units where the work is actually changing, which own the use cases and outcomes; and HR, which leads the people side of change, from capability building to how roles evolve. Each seat maps to a prong of the framework: the sponsor carries strategy, the business units carry grassroots adoption, and HR carries people change management, with IT enabling all three.

The lesson from large system implementations is blunt. The ones that succeed are owned by the business and sponsored from the top. The ones that stall are handed to IT and treated as software.

How authority is distributed across the enterprise once this structure is in place is taken up in Section 16.

The discipline matters more than the size of the structure. Organizations should scale the governance model to their size, complexity, risk, and stage of maturity while preserving clear ownership, cross-functional input, regular prioritization, and accountability for outcomes.

Strategy doesn't constrain experimentation. It's what enables scattered experiments to add up to something that compounds.

6. Prong Two: Grassroots Adoption

Grassroots adoption is where people build fluency. It is context training. Through practice, employees discover that vague prompts create vague outputs, that examples improve quality, and that AI performs better when the user defines the task, audience, constraints, source material, and desired format.

Organizations should not wait until every system is connected, every data policy is complete, and every use case is approved before employees begin learning. If they do, people never develop the practical judgment required to use AI well. At the same time, random experimentation cannot be the strategy. The best path is the middle path: start with low-risk, practical work while building the larger strategy in parallel.

Grassroots adoption can begin with safe use cases that do not require sensitive internal data:

• Public research and learning unfamiliar topics
• Meeting preparation and preparing questions before a decision
• Summarizing articles or reports
• Drafting emails or memos, and organizing messy notes
• Brainstorming options and creating first-pass outlines
• Turning spoken thoughts into structure

These early uses matter because they help people learn what AI is good at and where it fails: where it is useful, where it is weak, where it needs verification, and what kind of context improves the result. This is why grassroots adoption is not just “playing with tools.” The goal is not to make everyone an AI expert; it is to make people fluent enough to ask better questions, delegate better first-pass work, and recognize where AI can create capacity.

The first goal is not perfection. The first goal is fluency.

Beyond building fluency, grassroots adoption is a strategic discovery mechanism. When clusters of individuals independently apply AI to comparable tasks (synthesizing client histories, drafting standard communications, structuring meeting preparation), those recurring patterns signal high-value opportunities for redesign. The goal is not to collect prompts for their own sake, but to notice repeated work patterns that reveal where better data access, workflow redesign, and formal governance may create value.

7. Prong Three: People Change Management

The human side of AI adoption is often underestimated. Organizations may assume that if employees receive a tool and some training, behaviour will change. But AI asks people to work differently. It changes how they start tasks, ask questions, decide what matters, evaluate quality, and delegate work. This is not just technical; it is managerial.

Even employees who do not manage people may now manage digital helpers, AI workflows, or agents. That requires skills that look very much like management:

• Defining the outcome and providing context
• Setting constraints and explaining what good looks like
• Reviewing output and giving feedback
• Deciding what to accept, revise, reject, or verify
• Keeping accountability for the final result

Change Management Is More Than Training

People do not resist AI only because they lack technical skills. They resist when they don’t understand what the change means for their role, what they are permitted to do, how their performance will be evaluated, or whether the efficiency they create will ultimately be used against them.

Leaders do not need to have every answer, but they do need to communicate honestly and consistently. Employees need to understand why the organization is adopting AI, which uses are encouraged, where the boundaries are, what remains their responsibility, and how expectations are likely to evolve. Silence creates its own operating model: hidden use, uneven adoption, unnecessary fear, and inconsistent quality.

Managers are the translation layer between enterprise strategy and everyday work. A training program may introduce AI, but managers determine whether it becomes part of normal work. They create the space to experiment, model responsible use, share examples, reinforce good judgment, and help employees rethink established workflows. They also need to make it safe for people to discuss what did not work. Failed experiments are not wasted effort when they help the organization understand where AI is unreliable, where more context is needed, or where a workflow is not ready to change.

People also need clarity about the capacity AI creates. If employees are asked to save time but every gain is immediately converted into more workload, or interpreted only as an opportunity to reduce headcount, they will have little incentive to surface what they learn. Organizations should be explicit about how additional capacity will be used, whether to improve customer service, strengthen decision-making, accelerate innovation, reduce low-value work, or allow employees to focus on activities requiring human judgment.

There is a public example of an organization that made this choice deliberately. When Ingka Group, the largest Ikea franchisee, rolled out its customer-service bot Billie, the bot came to handle a large share of routine inquiries, work that had previously required its call-center staff. The conventional response would have been to cut headcount in proportion. Ikea did the opposite: it retrained roughly 8,500 call-center workers as remote interior-design advisors, redirecting them from queries the bot could absorb toward consultative work it could not. That reskilled channel became a new business line rather than a cost saving, generating on the order of €1.3 billion in FY2022 (about 3.3 percent of the group's revenue), with a stated goal of reaching 10 percent by 2028. Ikea has been careful not to promise the roles are permanent, noting that any future cuts would more likely stem from broader economic conditions than from AI. But the pattern is instructive: the capacity the technology created was treated as a demand signal and reinvested in people, and the people, in turn, produced growth.

In the transformations I led, the hardest part was not explaining the 'why' once. It was keeping it alive. People need to be reminded, repeatedly, why the organization is doing something this disruptive, because the effort is hard and the North Star fades under day-to-day pressure. I found this had to be reinforced at every stage of every rollout, paired with visible encouragement and shared wins, however small. A single announcement at the start isn't enough; people forget it. The communication that changes behaviour is the kind that repeats.

Change management therefore includes more than teaching people how to use AI. It includes:

• communicating the why repeatedly, not once, because the North Star fades under day-to-day pressure;
• making it safe to run experiments that fail, since a failed experiment still shows where AI is unreliable or a workflow isn't ready;
• equipping managers to turn a training program into normal work, because that translation is where adoption succeeds or stalls;
• redesigning roles, expectations, and performance measures so people aren't judged against a way of working that AI has changed;
• protecting time to rebuild workflows, since adoption can't happen on top of an already full job; and
• being explicit about how freed-up capacity will be used, so people aren't afraid the efficiency they create will be used against them.

The objective is not simply to make employees comfortable with a new tool. It is to help them develop the confidence, judgment, and new working habits required to operate effectively in a changing environment.

8. Delegation Without Abdication

A good manager does not delegate by saying “Do this” and walking away. They explain the objective, the standard, the constraints, the audience, and what good looks like. Working with AI requires the same behaviour. Delegation quality depends on context quality: the quality still comes from the person, but now it often comes through how well they direct and improve the work. A useful working heuristic is that AI can often create the first 70 percent: the outline, draft, summary, comparison, first analysis, or first set of questions. That first pass eliminates blank-page friction and gives the human something to react to, refine, and improve.

Our CTO, a software engineer by trade, has a way of putting this that I find useful: AI is like a brilliant but naïve developer. Capable of remarkable work, fast and widely read, yet lacking the judgment that comes from knowing the business, the stakes, and the context it is working in. The capability is real, but so is the need for direction and review. Left unsupervised, a naïve expert produces confident work that may be wrong in ways only someone with judgment will catch.

But delegation is not abdication. AI output is only as good as the context it receives and the judgment applied to the result. If the prompt is vague, the output will be vague. If the source material is incomplete, the output may miss nuance. If the reviewer does not check facts, tone, ethics, or fit, the work may move faster but not better.

This framing reduces fear. It does not say AI replaces human judgment; it says AI reduces the time people spend getting to the point where judgment can begin. People still own accuracy, context, quality, ethics, prioritization, relationship nuance, brand voice, tone, timing, and final decisions.

9. Clarity Creates Velocity

AI reduces the cost of producing output, which means organizations can quickly create more material than they can absorb: more documents, summaries, ideas, options, analysis, and artifacts. Without context, that output becomes noise. Clarity is the human control system for AI. The same discipline that improves an AI instruction also improves organizational execution. When a team needs to decide, implement, or escalate, leaders should make five things explicit:

• What is the fastest safe path?
• Who owns the decision?
• What would block us?
• What level of precision is required?
• Are we exploring, deciding, implementing, or escalating?

AI makes this discipline more important, not less. The more output AI can create, the more humans must provide prioritization, context, and judgment. Organizations that develop that capability will move faster without losing quality, trust, or value.

People build fluency while the organization builds the foundation to scale it.

10. The Adoption Flywheel

AI adoption does not come from one training session. It comes from repeated use, visible wins, peer learning, and leadership reinforcement. This loop is what drives the organization up the maturity path described later.

The flywheel matters because AI adoption is a behaviour change. The goal is not simply to make tools available; it is to normalize a new way of working.

Permission without capacity creates intention, not adoption. Changing how work gets done requires time. People cannot meaningfully redesign their workflows while operating at full capacity, moving from meeting to meeting, or absorbing AI on top of an already overloaded job.

This is one of the practical barriers to adoption. It is often easier to keep doing the work the old way, because the old way is known even when it is inefficient. Building a new workflow takes experimentation: testing tools, documenting steps, creating prompts, rethinking handoffs, and learning what can be delegated to AI, to automation, or to a person.

People need time to reconstruct the work.

Organizations therefore need to create safe space for reconstruction. If leaders want employees to change how work gets done, they must give them time to do that work. AI adoption cannot be treated as extra homework added on top of existing responsibilities. This matters even more in resource-constrained environments: without dedicated space to redesign work, an organization may never capture the productivity gains it hopes AI will create.

Protected capacity does not have to begin as a large new program. It can come from temporarily pausing lower-value work, reducing meeting load, selecting one workflow at a time, assigning short experimentation sprints, or allowing teams to reinvest a portion of the time AI has already saved. The goal is not to create a separate AI workload. It is to deliberately exchange some existing activity for the work of building a better way to operate.

For a resource-constrained organization, focus matters more than scale. One protected workflow experiment with a clear owner and business outcome is more valuable than asking every team to pursue transformation at once.

11. From Experiments to Repeatable Workflows

Grassroots experimentation is the beginning, not the destination. At first, AI use is personal and informal: one employee drafts a memo, another summarizes a report, another prepares for a meeting. These uses are valuable but local. The next step is pattern recognition. Leaders should ask:

• Which AI uses are repeating across roles?
• Which tasks are high-frequency and low-risk?
• Where is AI improving speed or quality, and where are people still uncertain?
• Which repeated patterns are strong candidates for workflow redesign or integration?
• Which workflows are ready to scale?

The key is not to scale every experiment, but to identify the patterns that repeat, add context, build guardrails, and turn them into workflows. Reusable prompts are not the destination. They are evidence. When similar prompts or AI uses appear across people and functions, they signal a recurring task pattern that should be evaluated, prioritized, and redesigned into workflow pilots, governed processes, or integrated capabilities. The goal is not to institutionalize prompts, but to institutionalize better ways of working.

The Organizational AI Maturity Path

Organizations rarely jump directly to enterprise AI. Maturity develops through a series of increasingly valuable stages. This is distinct from the adoption flywheel in Section 10. The flywheel describes how adoption behaviour compounds — the loop of permission, practice, and reinforcement that gets people using AI. The maturity path describes what the organization becomes capable of as that behavior takes hold: the ladder from personal productivity to an AI-enabled operating model. One is a loop that turns; the other is a ladder that climbs.

One discipline carries over directly from large system implementations, with a twist. After an ERP go-live, organizations move into a managed cadence, waterfall or agile, where adjustments are queued, priorities are framed, and resources are assigned to keep improving the system. AI needs the same cadence, but with an important difference: there is no go-live. Adoption is continuous, so the discipline that a go-live normally forces has to be built in deliberately.

In practice, the most effective mechanism is a regular steering committee of the key stakeholders. This is where priorities are set and resources are aligned, and where the organization makes the genuinely hard calls, not only what to prioritize but what to say no to. Those decisions are often the difference between progress and drift. Without that standing rhythm, adoption plateaus once the initial push fades. With it, momentum compounds. The organizations that sustain progress treat AI adoption as an ongoing operating cadence, not a project with an end date.

13. Better Context and Data Create Better AI

Individual productivity can begin immediately without any connected internal data. The turning point comes when AI gains trusted enterprise context and moves from a personal productivity tool to a decision-support partner.

AI Must Meet Enterprise Data Where It Lives

The ability to chat with your data depends on what information AI can securely access. A general-purpose cloud model may be a remarkably capable researcher, writer, and analyst, but without access to an organization’s own information, it remains a generalist with little understanding of how that particular business operates.

Much of the context that creates competitive value is distributed across private clouds, on-premises applications, operational systems, local and edge devices, internal documents, and employee knowledge—environments that general-purpose public AI cannot automatically access.

Gartner describes AI inference as increasingly distributed across cloud, data-centre, and edge environments and advises organizations that they may need to “bring AI to where the data is.” It also predicts that by 2030, more than 60% of enterprises will conduct intensive AI model activity in one cloud while using data located in another.

For many organizations, the enterprise AI opportunity will therefore be hybrid. Organizations need to combine the capabilities and external knowledge available through cloud AI with the private data, business definitions, workflows, and institutional knowledge that make the answers relevant. Depending on security, cost, latency, privacy, and regulatory requirements, that may mean connecting AI securely to the data, moving selected data into a governed environment, or bringing AI processing to where the data already resides.

The objective is not to put every piece of data in one place. It is to make the right information available to the right AI capability, under the right permissions, for the right business purpose.

Much of a company’s most valuable intelligence is hidden in the data public AI cannot see.

Making that intelligence usable requires a strong strategic foundation:

• Data quality and governance
• Clear business definitions and permissioned access
• Workflow documentation and trusted knowledge repositories
• Source transparency and human review standards
• Accountability for decisions and outputs

Access alone is not enough. Enterprise information must be current, permissioned, consistently defined, and connected to business meaning. Otherwise, AI will amplify the same process confusion, fragmented ownership, and inconsistent definitions already present. Context is not a technical afterthought; it is the control system that makes enterprise data usable.

14. Build to Learn, Write to Decide

AI also changes the sequence of work. Traditionally, many teams moved from idea to document to debate to approval to build. That process still has value, especially when risks are high or when writing is needed to clarify thought. But AI increasingly lets teams create a first-pass artifact much earlier: a document, prototype, mockup, workflow map, customer brief, dashboard concept, research summary, campaign outline, or operating procedure.

Writing still matters, because it forces clarity. But when first-pass artifacts can be created quickly, teams can learn from something real before writing the final plan. AI dramatically reduces the cost of experimentation: rather than debating ideas in theory, organizations can rapidly create drafts, prototypes, analyses, and scenarios, learn from real outcomes, and make better-informed decisions. The role of planning shifts from predicting the future to accelerating learning.

The updated sequence is: frame the problem, identify the assumption, create a first-pass artifact, use it the way a customer or stakeholder would, find the gaps, improve the thinking, write the better plan, and decide what to scale.

Build to learn. Write to decide. Do not just use AI to think faster—use it to learn faster.

15. Prioritizing What to Scale

As organizations become more fluent, they often discover more possible use cases than they can responsibly scale. At that point the constraint shifts from “Can we build it?” to “What should we build first?”

The best early scaled workflows are often not the most glamorous. They sit at the intersection of business value, repeatability, data readiness, context quality, and adoption readiness.

Be cautious about framing AI only around cost reduction at the start. Cost efficiency matters, but if AI is introduced primarily as a cost-cutting tool, employees may become defensive or fearful. Strong adoption often starts with effectiveness: better service, faster decisions, reduced friction, improved quality, and more capacity. Cost benefits can follow, but they should not be the only message. The goal is to build both value and organizational muscle. Reducing your workforce and then expecting the remaining staff to generate change is unrealistic.

16. Leadership Behaviours That Sustain Adoption

1. Define the outcomes

Leaders must clarify what the organization is trying to accomplish: faster work, better decisions, better customer experience, lower cost to serve, growth, innovation. Without this, AI activity becomes disconnected.

2. Create the conditions for fluency

Employees need permission, tools, examples, time, and safe rules of the road. They need to know what is acceptable, what is risky, and where to go for guidance. Leaders must model experimentation and reinforce visible wins, and they must carve out time for experimentation, workflow rebuilding, prompt development, peer learning, and process improvement. AI adoption requires both permission and an investment of time before it produces a return of time; temporary capacity allocation is part of the change process.

3. Build the foundation to scale

Leadership must ensure the organization invests in data readiness, governance, workflow redesign, and accountability. This is the work that turns individual productivity into enterprise capability.

4. Provide context

Leaders provide context when they clarify priorities, define outcomes, explain constraints, name trade-offs, and decide what matters most. In many ways, leadership itself becomes an exercise in context engineering: great leaders increasingly differentiate themselves not by having all the answers, but by providing the clarity, priorities, constraints, and business context that let both people and AI make better decisions. The better leaders are at providing context, the better people will be at using AI.

5. Communicate the People Impact

Leaders must communicate how AI is expected to affect roles, workflows, accountability, and the use of newly created capacity. They do not need to have every answer, but they should be candid about what is known, what remains uncertain, and what employees will continue to own. Consistent communication from leaders and managers prevents uncertainty from becoming resistance.

AI accountability should be federated. The enterprise owns standards, governance, security, and foundation. Business units own use cases and outcomes. The people closest to the workflow help define how AI is used, because they understand the work, the edge cases, and the customer impact. Technology teams establish secure platforms, standards, and governance, while business leaders remain accountable for where AI is applied, which workflows are redesigned, and how value is created. Accountability is federated while leadership and coordination remain centrally owned. The single-threaded executive owner does not make every decision or control every use case; they ensure that the distributed parts operate as one enterprise system.

Centralized enablement. Distributed execution.

18. Case Study: One Organization’s Journey from Experimentation to Organizational Capability

The framework in this paper reflects lessons drawn from multiple enterprise transformations, operating roles, and board-level observations. The following anonymized case study illustrates how many of those principles came together in one organization’s AI adoption journey.

The organization faced significant business pressure, rapid industry change, and a need to understand AI well enough to incorporate it into both current operations and future strategy. It did not begin with a finished enterprise architecture or a perfectly defined roadmap. It began with a clear need to learn quickly while leadership developed the strategy, data foundation, governance, and organizational capabilities required to scale responsibly.

Starting with a working strategy

Leadership created an initial AI strategy as a working hypothesis rather than waiting for a perfect long-term plan. It established the business rationale, set initial expectations, and gave employees permission to begin learning within appropriate boundaries. The strategy was deliberately communicated as evolving: as the organization learned more about the technology, employee behaviour, emerging risks, and business opportunities, leadership refined the plan and explained what those changes meant for employees. That communication mattered. Employees were not simply handed new tools. They were helped to understand why AI mattered, what responsible use looked like, and how it might change the way their work was performed.

Measuring the behaviours that build fluency

The organization introduced adoption metrics early, even though its first measures were intentionally basic. Initial goals focused on observable learning behaviours: how many employees were experimenting, how many custom assistants they created, how many tests they completed, participation in training and shared learning, and examples of AI applied to real work. These were not treated as measures of business value. They were leading indicators of fluency. Early on, leadership needed to know whether people were practicing, testing, sharing, and developing judgment.

Measure what you need to learn at each stage: use early metrics to build behaviour, and mature metrics to track value

As adoption matured, the measures evolved toward sharper questions: Which experiments improved speed or quality? Which uses were repeating across functions? Which tasks could be standardized or redesigned? Where would better internal data improve the result? Which workflows were ready for governance or investment? Where was AI creating measurable capacity or improving decisions? Which models and tools worked best for which workflows?.

Grassroots adoption as a discovery mechanism

Employees were given multiple approved AI options and encouraged to apply them to practical work; training, demonstrations, shared channels, and coworker examples reduced the friction of getting started. Early uses spanned research, drafting, meeting preparation, content development, analysis, knowledge retrieval, and customer support. As those experiments accumulated, recurring patterns emerged, and those patterns became evidence, showing leadership where employees were repeatedly applying AI and which tasks might deserve templates, redesign, better data access, or deeper integration. The organization was not trying to institutionalize every prompt; it was using grassroots activity to identify repeated work that might become organizational capability.

Developing human capability

The organization reinforced that AI could take a first pass, but employees remained accountable for the result. People were expected to provide context, review facts, protect confidential information, apply judgment, and retain the final call.

Managers highlighted visible wins, encouraged peer learning, and helped employees build skills in delegation, context-setting, review, prioritization, and decision-making. Leadership also recognized that meaningful adoption required time: employees needed protected capacity to test tools, reconstruct workflows, document what worked, and change established habits. The objective was not simply greater tool usage. It was to develop the skills and behaviours required for a different way of working.

Building enterprise context in parallel

While grassroots fluency developed, the organization pursued a multi-year data and context strategy designed to mature through three connected layers: a trusted data foundation establishing common metrics, definitions, and reliable sources; a conversational intelligence layer that let employees ask questions of trusted business data and surface patterns and anomalies; and an execution layer that prioritized actions, established ownership, and translated insight into follow-through. These were not isolated technical projects but a single path from trusted information, to understanding, to action. Leadership communicated this evolving roadmap so employees could see how their individual experimentation connected to the larger strategy.

The journey was not perfectly linear, and the metrics changed over time. That was a feature of the approach, not a failure: the organization measured what it needed to learn at each stage and refined its strategy as its capabilities matured.

Observed business outcomes

The most meaningful evidence was not increased AI usage. During 2025, the first full year of the organization's more structured AI adoption journey, the company performed approximately eight to nine percentage points above the estimated growth rate of its closest comparable market.

That momentum has continued into the current year. Based on year-to-date performance, the company is currently tracking approximately 12 to 13 percentage points above the external market growth benchmark.

The external comparison is based on a range of independent third-party estimates for the closest comparable market. The specific market category, underlying company results, and detailed calculations have been withheld to preserve confidentiality.

Many factors contribute to growth, and this comparison does not establish that AI alone caused the company's performance. Product strategy, market position, customer demand, and execution also played important roles. However, leadership observed a clear connection between the capabilities developed through AI adoption and the organization's improved execution: employees could research and synthesize information faster, teams could explore more alternatives before deciding, ideas moved from concept to test more quickly, and leaders had greater capacity to analyze issues and evaluate trade-offs.

Industry peers and external business partners independently remarked on the organization's pace and asked what it was doing differently. Their observations, combined with sustained performance above the external benchmark, provided meaningful evidence that the organization was becoming faster, more adaptive, and more capable of converting learning into execution.

The strongest evidence of adoption was not the number of tools deployed or experiments completed. It was that the organization became visibly faster, more adaptive, and increasingly able to outperform its market.

The central lesson

The organization did not begin by building an enterprise AI platform. It began by setting direction, communicating an evolving strategy, measuring experimentation, building fluency, observing repeated work, and developing its data and context foundation in parallel. As those capabilities matured, AI became part of a broader change in how the organization learned, decided, and executed.

This supports the central argument of this paper: strategy, grassroots adoption, and people change management are most effective when they advance together. Metrics reveal whether the behaviours required for adoption are taking hold. Repeated work patterns reveal what deserves investment. Enterprise context and governance allow the most promising opportunities to mature into repeatable organizational capability.

Conclusion: From Productive Chaos to Competitive Capability

AI adoption is not about chasing tools. It is about responding to a change agent that is reshaping how organizations work, learn, decide, and compete.

AI democratizes capability and insight. It gives smaller teams access to capacity that once required more people, more time, or more outside expertise. That creates opportunity and risk. If competitors can learn, research, draft, analyze, and prototype faster, then doing nothing is not a neutral choice.

AI also creates productive chaos, generating more ideas, drafts, summaries, analysis, and options than many organizations are prepared to absorb. The answer is not to wait for perfect certainty; it is to build the capability to channel that chaos. That requires a three-pronged approach. Strategy gives AI direction. Grassroots adoption builds fluency. People change management develops the human skills required to use AI well: judgment, delegation, prioritization, context-setting, feedback, quality control, and accountability.

Across all three, context is the bridge between AI activity and AI value. Strategy provides enterprise context. Grassroots adoption builds context fluency. People change management teaches humans how to provide, evaluate, and act on context.

AI supplies the capacity; people supply the clarity, judgment, and quality that turn it into advantage.

Organizations that combine AI's capacity with human judgment, trusted context, and redesigned workflows will not simply become more productive. They will be able to chat with their data, interrogate performance more deeply, and turn fragmented information into faster, better-informed action. They will become fundamentally more adaptive and unlike the transformations that came before it, this is one the business can finally own directly, because AI speaks its language rather than requiring translation into someone else's. The organizations that succeed will not be the ones that adopt the most tools. They will be the ones that build fluency, create trust, improve context, redesign repeatable workflows, and convert AI capacity into better decisions, stronger execution, and durable competitive advantage.

Technology creates disruption. Leadership creates the operating model. AI simply expands what is possible.

Source Notes

This paper synthesizes the author’s experience leading and governing enterprise transformations across multiple organizations, her AI adoption framework, related presentation materials, and operator observations. The case study illustrates the framework through one organization’s experience. Company-specific details have been generalized to protect confidential and proprietary information.

Benchmark note: Prior-year outperformance is based on completed annual company results compared with published third-party market estimates. Current-year outperformance is based on year-to-date company performance compared with the same estimated market growth range and should be viewed as an indicator of continuing momentum rather than a completed annual result.

References

• Microsoft. “2025: The Year the Frontier Firm Is Born.” Work Trend Index Annual Report, April 23, 2025. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-fir m-is-born
• McKinsey & Company. “The State of AI: How Organizations Are Rewiring to Capture Value.” March 12, 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-how-org anizations-are-rewiring-to-capture-value
• McKinsey & Company. “Seizing the Agentic AI Advantage.” June 13, 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/seizing-the-agentic-ai-a dvantage
• Hämäläinen, Lari, Mark Patel, Sven Blumberg, Tanguy Catlin, and Wasim Lala. “Is That AI Agent Worth It? Agentic Economics and the Modern Operating Model.” McKinsey Quarterly, 2026. https://www.mckinsey.com/capabilities/quantumblack/our-insights/is-that-ai-agent-worth-it -agentic-economics-and-the-modern-operating-model
• McClain, Colleen, Brian Kennedy, Jeffrey Gottfried, Monica Anderson, and Giancarlo Pasquini. “How the U.S. Public and AI Experts View Artificial Intelligence.” Pew Research Center, April 3, 2025. https://www.pewresearch.org/internet/2025/04/03/how-the-us-public-and-ai-experts-view -artificial-intelligence/
• Anthropic. “Prompt Engineering Overview.” Claude Documentation. Accessed July 29, 2026. https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview
• OpenAI. “Best Practices for Prompt Engineering with the OpenAI API.” OpenAI Help Center. Accessed July 29, 2026. https://help.openai.com/en/articles/6654000-playground-and-prompt-engineering
• Gartner. “Gartner Identifies the Top Trends Shaping the Future of Cloud.” May 13, 2025. https://www.gartner.com/en/newsroom/press-releases/2025-05-13-gartner-identifies-top-t rends-shaping-the-future-of-cloud
• Gartner. “Gartner IT Infrastructure, Operations & Cloud Strategies Conference 2026 Mumbai: Day 1 Highlights.” June 1, 2026. https://www.gartner.com/en/newsroom/press-releases/2026-06-01-gartner-it-infrastructur e-operations-and-cloud-strategies-conference-2026-mumbai-day-1-highlights
• Inside Ikea's Big Bet on Humans in the Age of AI, Fortune, July 30, 2026. https://fortune.com/2026/07/30/ikea-ai-workforce-reskilling-jobs-billie-chatbot-global-500 /

AI use and authorship disclosure: The ideas, observations, arguments and conclusions in this paper are the author's own and were developed over several months. Generative AI tools were used during the development and editorial process as thought partners and editing tools, including to test ideas, identify repetition, improve organization and grammar. Generative AI was also used to create or refine some illustrations based on concepts developed and directed by the author.