Inside most large organizations, the last three years have followed a familiar arc. AI arrived everywhere at once. Boards and shareholders demanded AI strategies, so leadership engaged in lots of AI pilots. Pilots were launched across every function that could make a case for one. Some of them were impressive. Most of them did not turn into anything a chief financial officer could point to on a financial statement. Somewhere along the way a feeling settled in, at the top of the organization and at the bottom, that this is expensive, there is no notable return, and that someone must have gotten something wrong. That feeling is the most common condition in enterprise AI today, and it describes the market rather than any one company.
The pilots deserve a kinder reading than they usually get. The leadership teams that launched them were answering a real question from their boards and their markets, and they were answering it without a playbook, and for good reason. No playbook exists for absorbing a general-purpose technology that improves every few months, arrives inside the software an organization already owns, and can be picked up by any employee with a browser. So the pilots served their purpose in that they taught the organization what these tools can do, where they help, and where they do not, at a cost that was affordable precisely because it was a pilot. They were not a wrong turn. They were the tuition of learning and getting the workforce fluent in AI.
The numbers tell the same story from the other direction. Eight in ten people now report that AI has improved their own productivity, while the share of organizations reporting any financial impact at the enterprise level has not moved in a year. It sits at about 37 percent, and the group reporting significant impact is still around 6 percent.1 Individually, people are slowly increasing both quantity and quality of work. Institutionally, the investment is not materially showing up yet in either the top or the bottom line. That gap is not a verdict on anyone. It is what the early years of a general-purpose technology look like, and it has a shape that becomes easier to lead through once it can be seen. The human side of the gap is measurable too. Where managers actively support their teams’ use of AI, employee engagement runs at 48 percent, and where they do not, it runs at 30 percent.2 People up and down the organization are looking for someone to tell them what is going on.
What has been missing is not effort but a map. This piece is our attempt to draw one. It describes the two tracks on which AI is arriving, where organizations stand on each of them today, what it looks like to run the decentralized productivity track on purpose, the stages of the centralized, eventual transformation track, and the mechanism by which the first pays for the second. The map is drawn from what the research measures and from what we see inside enterprises. Readers should be able to find their own organization, department, or team on it, see the progress already made, and lean into what comes next.
Two Tracks Are Running at Once, and Both Are Legitimate
Decentralized track
Productivity
Tools arrive wherever work gets done
- Generative AI assistantsDraft, summarize, analyze and answer
- Agentic desktop toolsCarry multistep tasks through to completion
- AI coding toolsLet employees build their own custom solutions
Workflows unchanged. People more AI fluent.
Pays inProductivity, which shows up as profit margin
Centralized track
Transformation
When the enterprise is ready to change its business
- 1Point AI solutionsAn agentic ecosystem arrives, with humans providing oversight
- 2Workflow redesignStarting in operations, learning how best to do this for the enterprise
- 3AI business transformationNew products and services for existing and new customers, across industries
Workflows redesigned. Enterprise learns how.
Pays inTop-line growth, which compounds at the far end
Look closely at almost any enterprise and two different things are happening at the same time. They are usually described as one thing, which is why the picture feels so confusing from inside. One of them is decentralized and one is centralized, they pay off in different places on the income statement, and they call for different kinds of leadership. Naming them separately is the first step toward seeing the shape of the last three years. It is also the first step toward planning the next three.
The decentralized productivity track is the arrival of general-purpose AI tools wherever work gets done. These are the assistants that draft, summarize, analyze, and answer, and increasingly the agentic desktop tools that can carry a task through several steps on their own. They arrive in a way no enterprise technology has arrived before. Who buys them? Sometimes IT, sometimes not. Sometimes someone who runs licensing switches a tool on for every operations group at once, and sometimes a chief financial officer buys a tool for the finance organization directly, without a project, without a capital request, and without IT in the approval chain. Either way, the people who use the tools are the people who learn them. They figure out what the tools are good for in their own work. They build small solutions for themselves, often to cover their own blind spots rather than to sharpen their strengths. And they get more done, and better, without anyone redesigning how the work flows.
Nothing about the workflow changes on this track. The people inside it get more capable with these AI tools. That is why the decentralized track pays in productivity, and productivity shows up as margin. It can touch the top line, and where it does the effect is real but not material. A sales team with a little more pull-through, selling what it already sells to the customers it already has, is the decentralized track at work in revenue. The gain is worth having, but it is not transformation.
The centralized track is AI the organization deploys when it is ready to move closer to transformation, most of it currently bought rather than built. It runs as a sequence. At the near end is the point solution, a system purchased to put AI to work on one function’s problem, such as a voice agent for tier-one customer calls or a system that matches invoices and flags the exceptions. Further along the centralized track is the redesign of a workflow around what AI makes possible. At the far end is transformation, meaning new products, new service lines, and new customers, in industries the organization already serves and in industries it does not. This track pays in top-line growth, and the far end of it is where growth compounds.
Every serious observer of enterprise AI sees this split, and each names it differently. Gartner calls the first kind everyday AI, which “enhances productivity by enabling humans to do the things they already do more efficiently.” It calls the second kind game-changing AI, which creates results “via new products and services or through new core capabilities.” By Gartner’s count, about 80 percent of organizations investing in AI are on the everyday side.3 That proportion is worth holding onto. It says that four in five organizations are on the decentralized track, and it says nothing about whether they should be anywhere else.
Deloitte, surveying more than 3,000 senior leaders, put the same finding under the heading “productivity for most, business reimagination for a few.”4 Among 271 growth leaders at U.S. companies with at least half a billion dollars in revenue, 63 percent said their organization uses AI mainly to improve efficiency or productivity. Fourteen percent said it uses AI mainly to stay ahead of competitors, and 7 percent said mainly to diversify revenue.5 Read together, those two findings put the far end of the centralized track at roughly one organization in fourteen. Researchers at MIT described the same two faces of the technology as early as 2024, broadly applicable tools for individual productivity on one side and tailored solutions for strategic objectives on the other, with the tools coming first.6 The split has been visible to anyone measuring it for two years.
The split, then, is consensus rather than our invention, described in a dozen vocabularies. What the consensus does not say, and what we believe, is that both tracks are legitimate, that the decentralized productivity track has to come first, and that most organizations are on both tracks at once and early on the centralized, eventual transformation track. Being early on the centralized track is not a shortfall. It is exactly where anyone would be at this point in the technology, and, as the rest of this piece argues, it is the right place to be. Everything that follows rests on that sequence.
Where Organizations Actually Are
It helps to see the whole field before locating any one organization in it, and the sources that measure the field this year agree on its shape. About one organization in five has scaled AI across multiple business units.7 Among chief executives, 64 percent describe their AI work as pilots and 26 percent describe it as part of a broader transformation, while nearly nine in ten say they are seeing some cost or revenue benefit in targeted areas.8 Those two findings sit comfortably together. Benefit in targeted areas is what the decentralized track and a first point solution produce, and it is what most organizations have.
A 2026 Accenture survey across 3,000 executives and 3,000 employees found 81 percent of employees stating AI tools had increased their productivity. The share of organizations reporting widespread, sustained business value from AI, though, was 23 percent, down from 32 percent earlier in the year.9 The decline looks like regression and is more likely measurement settling, as organizations count only the value that has held up. The number that remains is the real starting point. One chief executive in ten says AI primarily drives growth for the company today.10 That figure belongs to the far end of the centralized track, and it is about the size anyone should expect three years in.
Read together, those numbers describe a market that is mostly on the decentralized track, buying point solutions on the centralized one, and, by our read, planning the redesign work for 2027. An organization that recognizes itself in that description is in the company of nearly everyone. There is no cohort far ahead of it that it neglected to join. There is a sequence, and most of the market is at the same early point in it. What separates organizations now is not how far along they are but whether they are moving through the sequence on purpose.
Two Questions to Ask of Any AI Headline
Headlines will keep announcing that some company grew revenue with AI, or redesigned its workflows, or achieved transformation. Some of those stories are true and most of them are smaller than they sound. Two questions make them far easier to live with, and both come straight out of Figure 1.1 above. The first is how much, and on what base. The second is whether it happened or whether it is expected.
A few percent of revenue attributed to AI is a real result and worth applauding. It is also the marginal kind, the decentralized productivity track showing up in sales as more of the same thing sold to the same customer segments. Transformation, as this piece uses the word, means double-digit revenue growth, year over year, compounding, from products, services, and customers the organization did not have before. The base matters as much as the percentage. A 10 percent lift at a company with half a billion dollars in revenue is a rounding error at a company that books that much in a week. Surveys can mix the two without saying so, and the base question is usually the one that resizes a headline.
A large share of the AI numbers in circulation are expectations. They record what executives plan to spend, what they believe AI will do by 2028 or 2030, and what they hope to see. Expectations are useful, and they show where the market is pointed. They are not results, and a reader who treats them as results will feel behind a curve that does not exist yet. The second question separates the two in a single reading.
None of this is raised to call anyone out. Firms and institutions publish what they measure, and they measure what they can. The reason to carry the two questions is that reading the headlines without them is a reliable way for a leadership team to conclude that everyone else is further along than they are. They are not. Almost every revenue figure in circulation is the marginal kind, earned in good faith, and almost every transformation figure is an expectation. Knowing that is a relief, and it is also the correct reading of the evidence.
The Decentralized Productivity Track, Run on Purpose
Most organizations are standing on the decentralized track, so this is where the most useful thing we can say lives. There is a version of this track in which the tools get switched on and everyone is told to explore them, and there is a version in which it is run on purpose. The second version is what turns time saved into margin. It is not complicated, and it can be described in five moves, each of which belongs to the departments rather than to a central team. The moves build on one another, and the order below is the order in which they tend to happen.
Run a fluency program across the enterprise. This is not a two-hour orientation but an actual program, with classes, practice, and time, aimed at making everyone in the organization fluent in the productivity tools they now have. The people best placed to teach it are fluent in the tools and can teach through the lens of the work, finance in finance’s language, human resources in its own, and supply chain in its own. A generic AI class teaches features. A class taught by someone who understands treasury or accounts payable or demand planning is far more impactful and influential with employees. The difference shows up within weeks in what people attempt.
Let the departments own it. The decentralized track is decentralized by nature, and the best thing an enterprise can do is respect that. Each department is responsible for getting its own people fluent, and each department, through its own leadership, sets the expectation that fluency turns into more and better work. Central teams can enable this with sanctioned tools, shared training, and shared rules, and they should. The ownership, though, sits where the work sits. A department that is told to become fluent by someone else rarely does, and a department that decides to become fluent with its leadership leading the charge usually does.
Give people permission to build. Every department is really a set of sub-departments, and every sub-department a set of teams and sub-teams, down to the individual. Finance is treasury and cash management and financial accounting and tax and compliance and reporting, and each of those is several teams and sub-teams. At every one of those levels, people should have explicit permission to create small solutions that help them do their own jobs, and especially to build tools that cover their weaknesses. One analyst builds a tool for herself, shaped around the part of the job she finds hardest. A colleague with different strengths builds something else for the same task. Neither was meant for anyone else, and both are exactly right.
Let the good ones climb. What happens next is organic, and it should be allowed to be. A tool that works for one person gets noticed by a sub-team, which adopts it. A sub-team’s tool proves useful to the sub-department, then to the department, and a few travel far enough that they become something the enterprise adopts formally. This is how a decentralized track discovers what the organization actually needs, through hundreds of small experiments with the useful ones rising on their merits, rather than through a committee guessing. Governance should meet this with an invitation rather than an audit, which means clear boundaries, sanctioned alternatives good enough that nobody needs to work around them, and legitimacy for what people build inside the lines. That is how Plaster Group’s AI Governance Methodology treats it at its second level, and it is the reason we describe this track as healthy proliferation rather than disorder.
Decentralized track
The good tools climb
Hundreds of small experiments, with the useful ones rising on their merits
EnterpriseA few are adopted formally
DepartmentFinance
Sub-departmentTreasury, tax, reporting and the rest
Sub-teamSeveral teams within each sub-department
IndividualTools that cover each person’s weak spots
Raise expectations over time. Once a department is fluent, its leadership can begin to expect more from the same workflows, in the range of 10 to 20 percent more work completed and better work delivered. That is not an aggressive number. Field studies of general-purpose AI tools measure gains of 14 to 15 percent in customer support, 26 percent in software development, and 50 percent in marketing output at the task level.11 Bain’s model of how organizations progress puts blended productivity at up to 5 percent while only the early adopters have had their moment, and as much as 25 percent across adopting teams once everyone is working faster inside existing structures.12 Ten to 20 percent, expected on purpose and department by department, sits well inside what the evidence supports, and it is the number the rest of this piece relies on.
That leaves the reason most organizations are not yet seeing increases to bottom-line margin. In most places the tools were switched on and the program stopped there. Two-thirds of employees say they get limited or no guidance on what to do with the time AI saves them, and more than half say they are not reinvesting it in more valuable work.13 Time saved without an expectation attached to it is simply absorbed. Meanwhile, 88 percent of employees with enterprise AI access also use personal tools for work, and the ones who use both are 1.7 times more likely to report significant time savings.14 That is the discovery process alive and simply unorganized, and it is good news for any leadership team willing to organize it.
Sanctioned access is racing to catch up with use, with the share of workers equipped with approved AI tools growing from under 40 percent to about 60 percent in a single year.4 The organization is, in effect, ratifying what its people already discovered. None of that is failure. It is the decentralized track working, one step short of being run on purpose. The step is the fluency program and the expectation that follows it, and it is the smallest step on either track for the size of what it returns.
The Centralized Transformation Track, One Stage at a Time
The centralized track is where the larger returns live, and it is also where the anxiety concentrates. It is the track on which the organization has to change how it works rather than how fast, and that is a different order of undertaking from switching on a tool. The way through it is in stages, and the order matters more than the speed. Three stages cover the ground, and each one prepares the people for the next. An organization that begins at the third without the first two usually discovers why the order exists.
Stage One: Point Solutions, and Getting Used to Agents
For most organizations the centralized track begins with a purchase. A system arrives that puts AI to work on one function’s problem, and increasingly that system is an agent, meaning software that does not simply answer a question but takes an action, makes a call, and hands the rest to a person downstream. An agent now makes decisions that a person used to make. Getting used to that is real work, and it is work for the people around the agent more than for the technology. What does that work look like? It looks a great deal like teaching.
The agent needs context to decide well, and most of the context it needs lives not in a database but in documents, emails, contracts, and tickets, the semi-structured and unstructured material that a person absorbs without noticing. Giving the agent that context is the first job, and it is a human job. The people who have been doing the work are the only ones who know where that material lives and what it means. The outcomes are now probabilistic where they used to be deterministic, so the same input will not always produce the same output, and the organization has to learn what an acceptable range looks like rather than expecting a single right answer. Some fixing will happen in production, because a production environment cannot be faithfully rebuilt in a sandbox, and the expectation should be set early that this is normal rather than alarming.
The agent will more than likely make some poor decisions soon after it is deployed. It will tend to lean toward the answer it senses the person wants, a documented property of language models that researchers call sycophancy, which is one more reason a human stays in the loop.15 The right response to all of this is the response anyone would give an employee new to a job. It is clear context, candid feedback, patience, and an expectation of improvement. Agents learn and get better, and the people who work with them get better at working with them at the same time. Both kinds of learning are happening in stage one, and both are needed for subsequent stages two and three below.
This stage is about people getting comfortable, not about scale, and the market bears that out. Only about 7 percent of companies run fully autonomous agents in production today.16 Agents dropped onto old, human-paced workflows produce task savings rather than step change.17 Both findings point the same way. Stage one is where an organization learns what it feels like to have decisions made by something that is not a person, and that lesson has to be learned before the next stage can be designed well. Scale comes later, and it comes more easily to the organizations that let this stage take the time it takes.
Stage Two: Redesign the Work, and Start in Operations
When the organization is comfortable with agents in the workstream and confident enough to go further, the next stage is redesigning a workflow around what AI makes possible. This is different in kind from a point solution. A point solution slots into a process. A redesigned workflow changes the process, and with it the roles, the handoffs, and the measures. The people who own the process are the ones who change it, and that is the first sign that the centralized track has moved past purchasing.
Start in operations. A miss in operations is a cost line, meaning something took longer, something had to be reworked, or a quarter was more expensive than planned. A miss in revenue is something shareholders and boards do not forgive, and it is not where an organization should learn. So the first redesigned workflows should live where confidence can be built at a bearable price, in the operational processes that run the business rather than the ones that book its revenue. If the redesign works, the organization has a proven pattern and a team that has done it once. If it does not, the organization has a lesson and a cost rather than a crisis.
The rule inside this stage is the one Plaster Group’s AI Business Transformation (AIBT) Methodology teaches, which is to redesign the work before choosing the technology. Decide how the work should be done, with people and AI each doing what they do best, and then select the systems that serve that design. The reverse order, buying the system and then bending the process to fit it, is the pattern behind the large share of agent projects expected to be canceled over the next two years.18 It is the pattern of automating a process that should have been redesigned first. Redesign is business work, done by the people who own the process.
Stage Three: Transformation, and What the Word Actually Means
The third stage is the one the word transformation is for, and it is worth defining precisely, since it is the most misused word in enterprise AI. Transformation is not doing the same things more efficiently. It is not selling the existing products to a few more segments in the existing industry, however welcome that is. Transformation is taking the organization’s historical data, the record of every customer, transaction, decision, and outcome it has accumulated over decades, and using current AI capability to turn that record into new products and service lines, for the industries the organization already serves and, more importantly, for industries it does not. It is reaching customer segments the organization has never served, with insight that was locked in data no one could use at that scale before. That is where growth compounds, as each new offering feeds the record that makes the next one possible.
The research, read with the two questions from earlier, points in this direction and documents almost no one who has arrived. Stanford’s Digital Economy Lab studied 51 successful AI deployments and found that most were measured as cost savings, and that the highest returns came from the companies that pointed AI at revenue. Its summary line is that AI deployed for efficiency saves money, and AI deployed into the product changes the competitive position.19 A peer-reviewed study in the Journal of Financial Economics found that AI-investing firms grew faster in sales, employment, and market value, that the growth came primarily through product innovation, and that AI investment was not associated with productivity gains as conventionally measured.20 Both studies say the same thing in different registers. The durable return from AI is on the product side of the business, and it has been visible in the data since before generative AI arrived.
Expectations point the same way. Chief executives who systematically build on their own proprietary data expect 13 percent more of their 2030 revenue to come from products and services they do not sell today.10 That is an expectation, and the second question applies to it, but it is an expectation held by the people who control the investment, which makes it a reasonable guide to where the money is going. The direction is clear. The arrivals are few, as this is the hardest thing on either track and almost every organization is, sensibly, still on the stages before it. No verified case of double-digit, compounded revenue growth from new products in new industries has yet appeared in the research this piece draws on, and that absence describes the calendar rather than the opportunity.
And here is what happens when a leadership team sees this stage clearly for the first time. The transformation effort is enormous. A new service model changes what operations has to fulfill, and a new product line changes what finance has to account for and what the sales organization has to sell. The ripple runs through every function in the wave. The first arithmetic, done carefully, looks like hiring 10 to 20 percent more people in every domain going through the transformation, and it looks cost prohibitive. That fear is reasonable, and it is also the setup for the most important thing in this piece.
The Decentralized Productivity Track Pays for the Centralized Transformation Track
Funding the centralized track
The first track pays for the second
The productivity gains from AI fluency can be taken two ways
The 10 to 20 percent an executive feared having to hire is the 10 to 20 percent already freed. Nobody is cut. The dividend is paid in people, not dollars.
The decentralized track has the opportunity to pay for the centralized track, and that sentence is what turns Figure 1.1 into a plan. Recall what the decentralized track, run on purpose, produces. It produces departments that are fluent, working the same workflows, and delivering 10 to 20 percent more quantity and quality than they did before. That gain can be taken two ways. It can be banked as cost savings, by doing the same work with fewer people, or it can be treated as capacity, by keeping the people and freeing 10 to 20 percent of their time. The choice between those two readings is the most consequential decision on either track.
The second reading is the one that makes transformation affordable, and it is the one we recommend without hesitation. So where do the people for stage three come from? They come from the departments the decentralized track made capable. Transformation is staffed by pulling people out of departments to redesign the work, build the new offering, and carry the ripple through the functions it touches. The 10 to 20 percent an executive feared having to hire is the 10 to 20 percent the decentralized track already freed. Department heads would be smart to identify the right people for the transformation work, the ones who leaned in early to become AI-fluent, built innovative tools and small solutions of their own, and got comfortable working alongside agents. These employees are the ones who optimized the existing workflows with AI productivity tools, so they are the ones to redesign them. The dividend is paid in people, not dollars. Nobody is cut to fund this, and the plan is the opposite of cutting.
The market’s own advice points the same way, even where it does not say why. IBM’s Institute for Business Value, surveying more than 2,000 executives, frames the same two phases, productivity within existing business models now and reimagined industries later. Its recommendation is to reinvest productivity savings rather than bank them as profit, which 70 percent of the executives surveyed said they plan to do.21 Gartner predicts that by 2027, three-quarters of the organizations that capture AI productivity gains as cost savings will be eclipsed by competitors that reinvest those gains.22 Both are expectations, and both are held by the people who will act on them. We would only add what the gains should be reinvested in, which is the people, and the work of the centralized track.
Where an Organization Stands Is the Right Place to Start
The whole picture fits in a paragraph. AI is arriving on two tracks, a decentralized track that pays in productivity and bottom-line margin, and a centralized track that pays in transformation and top-line growth. Nearly every organization is on the first and early on the second, which is where the technology’s calendar puts everyone. Run on purpose, the decentralized track makes departments fluent and creates the opportunity to free 10 to 20 percent of their capacity inside the same workflows. That capacity, kept rather than cut, staffs the centralized track through point solutions, redesigned operations, and finally the new products and industries where growth compounds. Every place on that map is a real place. A department that just switched the tools on, a department running fluency on purpose, a team learning to work with its first agent, an operations group redesigning a workflow, and a leadership team sizing the top-line play are all on it. Each is making progress, and each has a next step that is smaller than it looks from where it stands.
That sequence buys something for everyone in the organization. For the board, it turns AI spend from an uninformed expectation into a sequence with a visible next step. For the executive team, it answers the hiring arithmetic before it is asked. For a department leader, it makes fluency something the department owns and the expectation that follows it something the department can meet. For the individual, it creates the right motivation to learn and lean into this new AI world. Wherever you find your own organization on this map, the ground under it is firmer than the last three years have made it feel. No organization skips the decentralized track. The ones that grow next will be the ones that ran it on purpose, kept the people it made capable, and pointed them at the business they do not have yet.
Sources
- 1.McKinsey & Company, “The state of AI in 2026: On the road to ROI,” August 25, 2026. Eight in ten respondents report that AI has improved their individual productivity, while the share reporting enterprise-level financial impact held at about 37 percent and the share reporting significant impact at about 6 percent. 1,719 participants in 97 nations, fielded May 4 to June 8, 2026, with 36 percent from organizations with more than $1 billion in revenue. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- 2.Gallup, “Employee Engagement Remains Flat as AI Adoption Accelerates,” July 2026. Engagement is 48 percent among employees whose managers actively support their use of AI and 30 percent where managers do not. 43,262 U.S. employees, fielded February and May 2026. https://www.gallup.com/workplace/712433/employee-engagement-remains-flat-adoption-accelerates.aspx
- 3.Gartner, “Get AI Ready: Action Plan for IT Leaders,” the AI Opportunity Radar. Defines everyday AI as enhancing productivity by enabling humans to do the things they already do more efficiently, and game-changing AI as creating results via new products and services or through new core capabilities, and reports that about 80 percent of organizations investing in AI are pursuing everyday AI. https://www.gartner.com/en/information-technology/topics/ai-readiness
- 4.Deloitte, “State of AI in the Enterprise 2026: The Untapped Edge,” January 21, 2026. Characterizes the market as “productivity for most, business reimagination for a few,” and reports that the share of workers with sanctioned access to AI tools grew from under 40 percent to about 60 percent in a year. 3,235 director-level to C-suite respondents in 24 countries, fielded August to September 2025. https://www.deloitte.com/content/dam/assets-shared/docs/about/2025/state-of-ai-2026-global.pdf
- 5.EY-Parthenon, “2026 EY-Parthenon Growth Survey,” April 2026. Sixty-three percent of respondents said their organization uses AI mainly to improve efficiency or productivity, 14 percent to stay ahead of competitors, and 7 percent to diversify revenue. 271 corporate growth leaders at U.S. companies with at least $500 million in revenue, fielded February 19 to March 12, 2026. https://www.ey.com/en_us/insights/strategy/growth-survey
- 6.MIT Center for Information Systems Research, “Managing the Two Faces of Generative AI,” September 19, 2024. Distinguishes broadly applicable tools for individual productivity from tailored solutions built for strategic objectives, and observes that organizations adopt the tools first. https://cisr.mit.edu/publication/2024_0901_GenAI_VanderMeulenWixom
- 7.Gartner, “Gartner Survey Finds Only 22% of Organizations Have Successfully Scaled AI Across Multiple Business Units,” September 1, 2026. Twenty-two percent of organizations report having scaled AI across multiple business units. 1,303 respondents at organizations with at least $50 million in revenue, fielded January to April 2026. https://www.gartner.com/en/newsroom/press-releases/gartner-survey-finds-only-22-percent-of-organizations-have-successfully-scaled-ai-across-multiple-business-units
- 8.Boston Consulting Group, “CEOs Are Starting to See Value from AI. Now Comes Execution,” July 2026. Sixty-four percent of CEOs describe their AI work as pilots and 26 percent as part of a broader transformation, while nearly nine in ten report some cost or revenue benefit in targeted areas. 152 CEOs. https://web-assets.bcg.com/35/b4/84dd34074e659e1fe61484c95238/ceos-are-starting-to-see-value-from-ai-now-comes-execution.pdf
- 9.Accenture, “Pulse of Change,” April to June 2026 wave. Eighty-one percent of employees say AI tools have increased their productivity, while 23 percent of organizations report widespread, sustained business value from AI, down from 32 percent earlier in the year. 3,000 C-suite executives and 3,000 employees. https://www.accenture.com/us-en/insights/pulse-of-change
- 10.IBM Institute for Business Value, “2026 CEO Study: Rewiring the C-suite,” May 2026. Ten percent of CEOs say AI primarily drives growth for their company today, and CEOs who systematically build on proprietary data expect 13 percent more of their 2030 revenue to come from products and services not offered today. 2,000 CEOs, fielded February to April 2026. https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/ceo
- 11.Stanford Institute for Human-Centered Artificial Intelligence, “Artificial Intelligence Index Report 2026,” Chapter 4: Economy, April 2026. Compiles field studies measuring task-level productivity gains from generative AI of 14 to 15 percent in customer support, 26 percent in software development, and 50 percent in marketing output. https://hai.stanford.edu/assets/files/ai_index_report_2026_chapter_4_economy.pdf
- 12.Bain & Company, “Mobilizing the Organization in the Agent Economy,” June 17, 2026. Describes a four-stage progression in which blended productivity gains run up to 5 percent while only early adopters are active and reach as much as 25 percent across adopting teams once everyone is working faster inside existing structures. https://www.bain.com/insights/mobilizing-the-organization-in-the-agent-economy/
- 13.Boston Consulting Group, “AI at Work: Why Strategy Matters More Than Tools,” June 2, 2026. Sixty-six percent of employees report limited or no guidance on what to do with the time AI saves them, and more than half say they do not reinvest it in more valuable work. Close to 12,000 frontline employees, managers, and leaders in more than a dozen markets. https://www.bcg.com/publications/2026/ai-at-work-why-strategy-matters-more-than-tools
- 14.Gartner, “Gartner Predicts by 2027, 50% of Enterprises Without a People-Centric AI Strategy Will Lose Their Top AI Talent,” May 13, 2026. Eighty-eight percent of employees with access to enterprise AI tools also use personal AI tools for work, and those who use both are 1.7 times more likely to report significant time savings. 12,004 employees and managers in 40 countries, fielded first quarter 2026. https://www.gartner.com/en/newsroom/press-releases/2026-05-13-gartner-predicts-by-2027-50-percent-of-enterprises-without-a-people-centric-ai-strategy-will-lose-their-top-ai-talent
- 15.Sharma, M., Tong, M., Korbak, T., et al., “Towards Understanding Sycophancy in Language Models,” arXiv, October 2023, revised May 2025. Documents the tendency of language models trained with human feedback to favor responses that match a user’s stated views over accurate ones. https://arxiv.org/abs/2310.13548
- 16.Bain & Company, “How Do Companies Create Value with AI?,” June 15, 2026. About 7 percent of companies run fully autonomous agents in production. https://www.bain.com/insights/how-do-companies-create-value-with-ai/
- 17.Forrester, “The State of Agentic AI in 2026: Companies Are Chasing, Few Are Catching,” 2026. Agents added to human-paced legacy workflows produce task-level savings rather than step-change value. https://www.forrester.com/blogs/the-state-of-agentic-ai-in-2026-companies-are-chasing-few-are-catching/
- 18.Deloitte, “Tech Trends 2026,” December 10, 2025. Cites Gartner’s June 2025 prediction that over 40 percent of agentic AI projects will be canceled by the end of 2027, in support of redesigning processes before automating them. https://www.deloitte.com/us/en/insights/topics/technology-management/tech-trends.html
- 19.Stanford Digital Economy Lab, “The Enterprise AI Playbook: Lessons from 51 Successful Deployments,” April 2026. Most deployments studied were measured as cost savings, the highest returns came from pointing AI at revenue, and the authors conclude that AI deployed for efficiency saves money while AI deployed into the product changes the competitive position. 51 case studies across 41 organizations, interviewed August 2025 to February 2026. https://digitaleconomy.stanford.edu/app/uploads/2026/03/EnterpriseAIPlaybook_PereiraGraylinBrynjolfsson.pdf
- 20.Babina, T., Fedyk, A., He, A., and Hodson, J., “Artificial intelligence, firm growth, and product innovation,” Journal of Financial Economics, 2024. Firms investing in AI grew faster in sales, employment, and market value, primarily through product innovation, with no measured association between AI investment and conventional productivity gains. https://alexxihe.github.io/jfe.pdf
- 21.IBM Institute for Business Value, “The Enterprise in 2030: Engineered for Perpetual Innovation,” January 2026. Describes a first phase of productivity within existing business models and a second phase of reimagined industries, recommends reinvesting productivity savings rather than banking them as profit, and reports that 70 percent of executives plan to reinvest productivity gains in growth. More than 2,000 senior executives across 33 geographies and 23 industries, fielded third and fourth quarters of 2025. https://newsroom.ibm.com/2026-01-19-ibm-study-ai-poised-to-drive-smarter-business-growth-through-2030
- 22.Gartner, “Gartner Identifies 4 Shifts Shaping the Future of Work,” September 9, 2026. Predicts that by 2027, 75 percent of organizations that capture AI productivity gains as cost savings will be eclipsed by competitors that reinvest those gains. https://www.gartner.com/en/newsroom/press-releases/2026-09-09-gartner-identifies-four-shifts-shaping-the-future-of-work
- 23.Capgemini Research Institute, “The multi-year AI advantage: Building the enterprise of tomorrow,” January 15, 2026. More than half of organizations are committing to sustained AI investment, with many planning over a five-year horizon. 1,505 executives, director level and above, at organizations with more than $1 billion in annual revenue across 15 industries in North America, Europe, APAC, and Latin America, fielded November 2025. https://www.capgemini.com/insights/research-library/ai-perspectives-2026/
- 24.McKinsey & Company, “Rewiring for AI: From ambition to advantage,” May 7, 2026. Among a cohort of 20 companies in McKinsey’s research set that applied the rewired approach in full, it takes one to two years on average to become cash accretive, and DBS made foundational investments for three or four years before generative and agentic AI arrived. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/rewiring-for-ai-from-ambition-to-advantage
Frequently Asked Questions
What does an enterprise AI fluency program actually involve?
A real program rather than an orientation, with classes, practice, and time, aimed at making everyone fluent in the productivity tools they now have. It is taught by people who are fluent in the tools and can teach through the lens of each department’s own work, it is owned by the departments themselves with central enablement, and it is followed by an expectation, set by each department’s leadership, that fluency turns into more and better work within the same workflows.
What counts as transformation, and what is productivity?
Productivity is doing the same work better and faster, and selling more of the same things to the same or adjacent customers. It is valuable and it shows up as bottom-line margin. Transformation is using the organization’s historical data and current AI capability to create new products and service lines for the industries it serves and for industries it does not, reaching customer segments it has never served. It shows up as double-digit, compounding top-line growth, and very few organizations have reached it yet.
Why start workflow redesign in operations rather than in products or sales?
A miss in operations is a cost, and a miss in revenue is something boards and shareholders will not forgive. Redesigning an operational workflow first lets the organization build confidence, prove a pattern, agree on a methodology, and train a team at a bearable price. Once it has done that, it can take the same capability to the revenue-generating work with far less risk.
How does an organization pay for transformation without hiring more people?
By running the decentralized track on purpose. Departments that are fluent and working the same workflows deliver 10 to 20 percent more quantity and quality. If that gain is treated as capacity rather than banked as cost, it frees the people the transformation needs, and they are the right people, as they already know the work. The decentralized track funds the centralized track, and nobody is cut to do it.
How long should both tracks take to reach transformation that shows up in top-line revenue?
Plan on three to five years with a strong emphasis on getting employees AI-fluent and focused on productivity via the decentralized track. No large enterprise has completed the full arc yet, so the answer is a planning horizon rather than a measured benchmark, and the horizon the people funding this work already hold is consistent with it. Most senior executives now expect AI to contribute significantly to revenue by 2030,21 more than half of large organizations are committing to sustained AI investment with many planning over five years,23 and chief executives are being told that AI rarely breaks even inside a year.8 Among the organizations furthest along, one to two years to become cash accretive is the reported average, and in the case most often pointed to, several years of foundational work came before that.24 Point solutions arrive in weeks and a redesigned operational workflow in quarters. New products and service lines in new segments take years, because the data, the fluency, and the redesigned work have to be in place first, and that is the part worth being patient about, since the decentralized track is preparing it the whole time.
About the author
Shawn Plaster
Founder & CEO, Plaster Group
Shawn is the author of Plaster Group's five-level AI Business Transformation Methodology and its 27-article Insights series, and leads the firm's enterprise AI transformation work.
© 2026 Plaster Group, LLC. All rights reserved. This article may not be reproduced, distributed, or transmitted in any form without prior written permission from Plaster Group. Brief excerpts may be quoted for review or commentary purposes with attribution to the author and a link to the original article.
Ready to move forward?
Let's discuss how your organization can build with AI — securely, strategically, and starting from where you are today.
Start a Conversation