MF
Field notes

What the Market Is Saying About Enterprise AI Adoption and Value

The Adoption Gap · Oct 15, 2026 · 7 min

Adoption is nearly universal and value is not. The companies closing the gap are redesigning work and governing agents, not buying more tools.

The short version

  • Use is broad, financial proof is thin. McKinsey's August survey (opens in a new tab) found 44% of organizations scaling AI enterprise-wide and 80% of respondents reporting better personal productivity. Only 37% attribute any EBIT to AI, and about 6% qualify as high performers. Both numbers are flat year over year.
  • The answer depends on how you ask. BCG (opens in a new tab) counts nearly half of companies (7.5% future-built plus 41% scaling) as capturing meaningful value. KPMG (opens in a new tab) finds nearly 6 in 10 leaders report measurable value, but only 37% point to stronger financial performance. Productivity is easy to claim. P&L impact is hard to prove.
  • The constraint has moved from technology to the operating model. How work is designed, how data is governed and who is accountable now decide results.
  • Agents are the next test. 62% of KPMG respondents are building or deploying agents. BCG finds 42% of companies expect to give agents real decision authority by 2030, yet only 5% have the full set of controls today.
  • The winners redesign work. Nearly three-quarters of McKinsey's high performers fundamentally redesigned workflows, versus about one-quarter of everyone else.
Adoption is broad. Proof is thin.
  • 80%say AI improved their own productivity
  • 44%are scaling AI across the enterprise
  • 37%attribute any EBIT to AI
  • 6%are high performers, with 5% or more of EBIT from AI
The drop from 80% to 37% is the value gap. The last two figures are flat year over year.Source: McKinsey, The state of AI in 2026 (Aug 25, 2026; 1,719 respondents in 97 countries).

What is driving adoption

Four forces show up across the recent research.

  • Money is flowing, and it is leaving IT. BCG reports AI spend rising from about 1.7% of revenue in late 2025 to 3.3% now, with more than 80% sitting outside the enterprise IT budget. McKinsey finds 60% of respondents expect to raise AI investment next year.
  • Access has widened fast. KPMG reports significant workforce adoption at 44% of organizations, up from 23% last quarter and 10% a year ago. Deloitte's 2026 report (opens in a new tab) describes sanctioned AI access moving from under 40% to around 60% of workers.
  • Individuals feel the gain first. McKinsey's 80% productivity figure fits a Financial Times survey of 6,000 digital workers, reported in McKinsey's September newsletter: about 11 hours saved a week, but only 13% seeing better organizational performance.
  • Agents and coding tools change the build-or-buy math. McKinsey finds 40% of large organizations scaling agents, up from 27%, while smaller firms sit flat at 22%. Nearly a third have decided against buying software they could build in-house with coding agents. Gartner's forecast, published in 2025, is that 40% of enterprise apps will embed task-specific agents by the end of 2026.

The barriers, ranked by the weight of evidence

Six barriers. Few are technical.
  1. 01~25%Workflow bolt-onof non-high-performers redesigned workflows
  2. 025%Governance laghave full agent controls today
  3. 0380%Data readinessof banking leaders cite it as the top challenge
  4. 0447% vs 31%Misalignment and unmeasured valueof mid-level managers and ICs versus executives report negative effects
  5. 05TrustTrust and strainUX research: test tools against real tasks, not demos
  6. 0620%Costof organizations say they are cost-constrained
Sources: McKinsey (Aug 25, 2026) for 1, 4 and 6. BCG (Sep 30, 2026) for 2. KPMG (Sep 24, 2026) for 3. Nielsen Norman Group (Aug to Oct 2026) for 5.
  1. Workflows are bolted on, not redesigned. Publicis Sapient (opens in a new tab) found AI supports 53% of enterprise work, yet only 18% of organizations have fully integrated it, and 22% name the way their organization runs as the primary barrier. Deloitte puts 37% of companies at surface-level use with little structural change.
  2. Governance trails the agents. BCG's 42% versus 5% gap is the clearest signal. A Retool survey of 307 technology leaders, as reported by First Line Software, found only 8% rate their governance as strong and 22% had a production incident from an AI-built internal tool. Gartner's 2025 survey of IT application leaders found only 13% had appropriate governance structures.
  3. Data and system readiness. In KPMG's banking cut, 80% name data readiness and access as the top challenge to agent deployment, up from 63% a quarter earlier.
  4. Leaders are misaligned and value goes unmeasured. Gartner found only 14% strongly agree that IT, the business and executives align on which problems AI should solve. Those that do were over three times as likely to find significant value from generative AI. McKinsey's high performers are twice as likely to have defined processes for measuring impact. Forrester predicted that fewer than one-third of firms could tie AI value to profit and loss, and that finance scrutiny would delay about 25% of planned spend into 2027.
  5. Trust and strain among the people doing the work. McKinsey finds 47% of mid-level managers and individual contributors report a negative effect from AI, against 31% of executives and senior managers. WRITER's 2026 survey reports 54% of C-suite leaders saying adoption is tearing their company apart, and 75% saying their AI strategy is mostly for show. Both figures come from a vendor survey, so weigh them accordingly.
  6. Running costs. About 20% of McKinsey respondents say operating costs, including tokens, have constrained their AI use.

Tactics companies are deploying

  • Redesign the workflow end to end, then add the AI. This is the clearest separator in McKinsey's data (about 75% of high performers versus 25% of others). BCG says the same of its leaders: they reshape workflows end to end in high-value areas and measure impact directly in the P&L.
Winners redesign the work. Everyone else adds a tool.
  • ~75%of high performers redesigned workflows
  • ~25%of all other organizations did
The clearest separator between value and no value is not the model. It is the workflow.Source: McKinsey (Aug 25, 2026). High performers are about 6% of respondents.
  • Put financial discipline around AI spend. KPMG reports 74% of organizations now include cost reviews in AI approvals (up from 61%), 70% use monitoring dashboards and 43% have usage or token budgets.
  • Draw autonomy boundaries before agents ship. 49% of KPMG respondents have defined high-risk use cases where agents may not decide on their own. In KPMG's banking cut, 75% do not let agents touch sensitive data without human oversight, up from 51%. BCG lists oversight, rollback gates, security, audit and cost guardrails as the controls agents need.
Agents are arriving faster than controls.
  • 62%are building or deploying agents (53% a year ago)
  • 42%expect to give agents decision authority by 2030
  • 5%have full controls in place today
Leaders pair agents with cost reviews, monitoring, token budgets and no-autonomy lists for high-risk uses.Sources: KPMG AI Quarterly Pulse (Sep 24, 2026; 314 respondents) for 62%. BCG (Sep 30, 2026; 1,330 respondents) for 42% and 5%.
  • Fund the people side alongside the technology. BCG finds seven in ten future-built and scaling companies already retraining staff, and expects dedicated AI roles to grow from 7% to 22% of the workforce by 2030.
  • Reopen build versus buy, with discipline. Almost half of McKinsey's high performers have skipped a software purchase to build in-house. They also report cost constraints on coding agents about three times as often as others, which is what heavier use looks like.
  • Co-design with employees. A 2025 HBR study found 76% of executives believe employees are enthusiastic about AI while 31% of individual contributors agree. Organizations that involve employees in AI decisions were seven times more likely to succeed. That is earlier research, but it matches KPMG's jump in workforce adoption where governance became visible.

What the UX experts add

Nielsen Norman Group's recent AI writing treats adoption as an evidence and design problem, not a rollout problem.

  • Trust is a design problem. NN/g's State of UX 2026 (opens in a new tab) calls this the year of AI fatigue and warns that bolting AI onto everything turns it into noise.
  • Test tools against tasks, not hype. In August, NN/g argued that pressure to adopt is not evidence a tool helps, and that one AI output is an example, not an evaluation. Use representative inputs and repeated runs.
  • Fast building outruns judgment. An October piece warns that cheap building lets agentic systems grow piecemeal into fragile messes their owners no longer understand. An August piece says teams now build faster than UX can evaluate.
  • Context is the new interface. NN/g's September and October research on agent power users shows they curate context but rarely edit their context files and often cannot find them. A governed, findable context library is an adoption lever, not an IT detail.

The NN/g AI archive (opens in a new tab) carries these pieces.

Advice for leaders about to start

These seven moves are my synthesis of the evidence above, not a finding of any single study.

  1. Pick two or three workflows tied to a P&L line. Redesign them end to end before you buy anything.
  2. Set the measure before the pilot. Agree the baseline, the owner and the finance sign-off up front, since finance will ask.
  3. Write the autonomy policy first. Define what agents may do, what data they may touch, how you roll back, and who can switch them off.
  4. Fix data access for those workflows only. Enterprise-wide data cleanup is how programs stall.
  5. Budget for adoption. Retraining, champions and co-design belong in the plan. Treat mid-level managers as the group under the most strain.
  6. Evaluate with evidence. Use representative inputs and repeated runs, not a single impressive demo.
  7. Meter cost from day one. Token budgets and cost reviews are now ordinary practice.
Seven moves before you start.
  1. 1Pick workflows, not tools. Start from the work you want changed.
  2. 2Redesign before you deploy. Change the process, then add the AI.
  3. 3Name an owner and a value metric. Measure outcomes, not usage.
  4. 4Fix data for one use case first. Do not boil the ocean.
  5. 5Set agent guardrails early. Define the no-autonomy list and monitor.
  6. 6Budget and review cost. Use cost reviews and token budgets.
  7. 7Bring managers and teams in. Test tools against real tasks with the people who use them.
Synthesis of McKinsey (Aug 25, 2026), BCG (Sep 30, 2026), KPMG (Sep 24, 2026) and Nielsen Norman Group (Aug to Oct 2026). The ordering and wording are the author's.

Start narrow, redesign deep, and govern before you scale.

Sources

Building something where AI meets people? Let's talk.

I'm always glad to talk about speaking and teaching, or mentoring people breaking into AI. Hablemos.