ICYMI: Poor Deployment Can Waste 99% of AI’s Potential.
Executive Summary
BCG, Deloitte, Gartner, KPMG, McKinsey and PwC have all published surveys in 2026 on AI deployment and its impact on business.
Taken together, the surveys point to three broad groups: Experimenters testing AI, Deployers putting it into production, and Transformers generating significant business value.
The top-performing Transformer group includes 6–13% of respondents based on five of the surveys (although the definitions are inconsistent).
Generating returns from deployed AI appears to be the challenging step. Several surveys show the balance shifting from Experimenters towards Deployers, whilst Transformers remain a small minority.
Transformers pursue growth and reinvention alongside efficiency, with senior sponsorship and sustained funding, workflow redesign, measurement of results, and stronger data capabilities and technical controls. This is correlation, not causation.
Leaders should set AI priorities according to which group their firm belongs to. Experimenters need the resources and systems to move into production. Deployers need to redesign work and realise returns. Transformers should invest more in AI where returns justify it, extending and renewing what works.
The AI Deployment Surveys
McKinsey, BCG, KPMG, Gartner, PwC and Deloitte have all published surveys (see Sources for links) on business value from AI in 2026. Several build on earlier surveys, allowing some comparisons over time. Each firm defines success differently, so the headline figures are not directly comparable. Nevertheless, there are several consistent patterns that can be seen across the surveys, which I’ve tried to distil in this article.
What counts as value
The surveys measure different degrees of success, from the 6% classified as high performers by McKinsey to the 89% reporting measurable value in KPMG’s survey.
In KPMG’s survey, 89% report “measurable value”, but only 10% report “established ROI”. PwC states that 56% of CEOs report neither higher revenue nor lower costs from AI over the preceding year, whilst only 12% report both.
Five surveys put their leading groups at between 6% and 13% of respondents, although they all use different definitions:
McKinsey: about 6% are “high performers”, with 5% or more of their operating profit (EBIT) coming from AI, and reporting “significant” value.
BCG: 7.5% are “future-built”, scoring highest on its AI maturity rubric. These firms report significant business value delivered.
KPMG: 10% report “established ROI”.
PwC: 12% of CEOs report both increased revenue and reduced costs from AI.
Gartner: 13% report significant or transformational enterprise value.
The range reflects how the question was phrased, who responded and survey timing. As these are different tests, the figures shouldn’t be read as showing a common level of success or ROI. For example, even when a firm (or group of firms) reports both higher revenue and lower costs, that doesn’t tell us whether the benefits exceeded the full cost of AI. Equally, a firm reporting only one of those benefits may still have a substantial return.
Deloitte reports a higher figure: 25% say AI has had a transformative effect. Its sample is restricted to organisations already using AI daily, so I’ve discussed it separately from the headline 6–13% range.
Three groups are emerging
Taken together, most of these surveys point to three broad groups. The pattern comes through most clearly in the McKinsey, BCG, KPMG and Deloitte surveys. Each of the consultants uses their own terminology, but I’ll call the groups Transformers, Deployers and Experimenters as I think those titles best capture what the firms are actually doing with AI. BCG gives the clearest picture of their relative size, so I’ll draw on their numbers as we discuss the groups:
Transformers. These are firms using AI to change how the business works and generating significant value as a result. AI is becoming embedded in complete workflows. For example, a customer enquiry might be resolved automatically, with people handling exceptions. Roles, decisions and processes are changing alongside the systems. Management can point to business results and approaches that work are being scaled. BCG’s highest maturity tier, which provides the clearest illustration of this group, represents 7.5% of its sample.
Deployers. These firms are putting AI into production, but have yet to turn much of that activity into financial results. Developers use coding assistants, service teams have tools to answer enquiries and staff can search company knowledge. Some processes have been redesigned and individual teams can demonstrate improvements. But the gains are uneven, and the wider organisation operates much as before. Time saved by teams does not necessarily become lower costs or additional revenue for the business. They represent 41% of BCG’s sample.
Experimenters. These firms are still largely testing AI. Employees are using chatbots to draft documents and summarise information, whilst specialist teams run pilots, compare suppliers and explore possible uses. Success stories may still depend on a few enthusiastic colleagues. The firm is working through how to connect AI to its systems, control its use and support it beyond the pilot teams. There is activity and learning, but little deployment at scale. At 51.5%, this is the largest group in BCG’s sample.
Several surveys suggest that Experimenters are giving way to Deployers, whilst Transformers remain a small group. Using BCG’s numbers again, Deployers rose from 35% to 41% between 2025 and 2026, whilst Transformers increased more slowly from 5% to 7.5%. McKinsey’s equivalent group held steady at 6%.
The returns are also concentrated in relatively few firms. In a separate PwC study of 1,217 organisations, the top 20% capture 74% of AI-driven returns, measured through revenue and efficiency gains adjusted for sector differences.
This pattern suggests that turning deployment into material returns remains the challenging step. Getting AI into production is progress, but that does not make a Deployer a Transformer.
Some of the foundational capabilities are table stakes and don’t differ much between segments. For example, basic data structuring and access controls are in place across all maturity groups, according to BCG, although Transformers demonstrate some more advanced technology capabilities. Ambition to improve efficiency is also widely shared, with McKinsey finding this ambition among around 80% of both high performers and other respondents.
What Transformers do differently
Five themes recur across the surveys, covering both what Transformers aim to achieve with AI and how they manage it. There are two caveats to bear in mind. Firstly, these profiles combine different surveys, and I’ve taken some artistic licence in describing what firms in each group are doing. Secondly, some groups are, in part, defined by the practices or outcomes they show, so part of the correlation is ‘built in’.
1. Growth and reinvention
Transformers use AI to pursue growth and reinvention as well as efficiency. McKinsey and PwC both identify this broader ambition. They seek new sources of revenue through innovation, new offerings and changes to the business model, alongside improvements to existing work.
2. Senior sponsorship and sustained funding
Transformers typically give a senior owner responsibility for a funded, multiyear programme. McKinsey’s Transformers are twice as likely to say senior leaders are visibly committed. BCG finds that 61% of its top performers fund one multiyear programme, compared with 17% of Experimenters, and their C-suite is 20 percentage points more likely to hold the AI budget.
Sustained funding also requires knowing when to kill pilots that aren’t working. Gartner finds that organisations with rules for stopping AI pilots are 2.3 times as likely to be Transformers. Those with rules for abandoning low-value initiatives already in production are 2.6 times as likely to be top performers.
3. Workflow redesign
Workflow redesign means rebuilding a few functions end to end, rather than injecting AI tools into existing workflows. In McKinsey’s survey, nearly three-quarters of Transformers redesigned workflows, versus a quarter of other respondents. BCG finds that 62% of Transformers’ initiatives reshape or reinvent processes, versus 44% of Experimenters.
Importantly, work and roles need to be considered together. As agents take on more tasks, firms need to decide what people will do, which decisions agents can make and where human approval is required. That has a knock-on effect on job descriptions, skills and measures of performance. Someone must remain accountable for the outcome, including the work done by agents.
Gartner finds that firms with explicit plans to redesign roles for a workforce of people and AI agents are 2.4 times as likely to be Transformers.
4. Measurement of results
Transformers are more likely to track the money. BCG finds that 95% of them use clear KPIs or track value directly in the profit and loss, versus 53% of Experimenters. McKinsey’s Transformers are twice as likely to have defined processes for measuring impact. BCG also finds that firms tracking value directly in the P&L report three times the AI value of firms with no formal measurement.
But even Transformers have more to do. Only 48% of KPMG’s top group consistently weigh value against cost. BCG finds that nearly half of its Transformers track AI returns directly in the profit and loss, whilst most of the rest use proxy measures.
5. Data and technical controls
Transformers are more likely to have formal systems for managing AI across teams. These can control the data that AI can access and determine when human approval is required. KPMG finds systems of this kind at 86% of Transformers, versus 31% of Experimenters.
BCG finds basic data structuring and access controls already in place at 58–72% of firms across all of its groups. Advanced data capabilities, including consistent data definitions across systems, clear data ownership and live access to company data for AI agents, are more often found in Transformers.
What this means for the Deployment Chain
In my July article, I used various survey proxies to illustrate how up to 99% of AI’s potential value could be lost across the Deployment Chain.
I advised firms to cost the whole chain and give one owner responsibility. I stand by that advice, but I would now put the emphasis in different places for each group, as follows:
Experimenters: Reallocate and Rewire. Firms that are still experimenting should focus on the commitment and capability needed to move AI into production. This includes ensuring that sustained funding and talent are available for the most promising use cases, and that the right systems and data foundations are funded and built. The intended business outcomes must be clear from the outset.
Deployers: Reimagine, Reshape and Realise. These firms already have AI in production. They should focus on redesigning work and roles around deployed AI, addressing any remaining technical constraints and converting operational improvements into lower costs or profitable growth. The priority is to get more value from what is already deployed. That can be achieved by finding out where demonstrated gains are getting stuck. If AI speeds up one part of the process but the overall result doesn’t improve, there may be a problem with handoffs, approvals or bottlenecks. Additionally, check that time saved is being converted into lower costs or profitable growth, through changes to budgets, staffing plans or work allocation.
Transformers: Reallocate and Repeat. These firms have demonstrated that they can generate real business value. I would now reallocate more funds to AI, scaling approaches that have worked and revisiting them as AI’s capabilities and economics change. That should include further investment in new products, services and business models, whilst continuing to test whether the returns justify the investment.
Monday Morning Actions for Executives
Establish whether your firm is an Experimenter, Deployer or Transformer, based on production use and realised business value.
Check that your agenda matches your segment: production capability for Experimenters, realised returns for Deployers, and further investment in scaling for Transformers.
Pick one workflow and trace the delivery of the benefit all the way to the bottom line, estimating leakage along the Deployment Chain.
Shift AI funding to a multiyear programme, not a string of pilots, with go/no-go checkpoints defined.
Questions for the Board
Which group best describes our AI maturity, what evidence supports that assessment, and is our AI plan aligned?
What returns will we deliver from our AI investments, and how do they compare with other uses of capital?
How do we know that the gains we report come from AI, have actually been realised and exceed the full cost?
What needs to change in roles, budgets or organisational boundaries to realise the benefits, and are we prepared to support those changes?
Sources and notes
These surveys use different samples and definitions, rely largely on executives’ assessments and show associations, not causes. All were published in 2026, although some fieldwork and reported benefits date from earlier years.
McKinsey: The state of AI in 2026: On the road to ROI, August 2026.
BCG: The Formula for Agentic AI Value, Applied AI Index, September 2026.
Gartner: CIO survey and Symposium findings, October 2026. Comparisons use its 11% vanguard group.
PwC: 29th Global CEO Survey, January 2026.
PwC: How leading companies generate ROI from AI, May 2026.
Deloitte: The State of AI in the Enterprise, January 2026.
Considered Machines: Poor Deployment Can Waste 99% of AI’s Potential, July 2026.
Workflow redesign showing up in the Transformer group makes sense. Bolting AI onto the old workflow mostly gets you a faster version of the same process, which might be why so many teams stall as Deployers.