In One Sentence
AI tools make individual workers faster almost everywhere they're deployed, but that speed only becomes company earnings when the surrounding workflow, headcount, and handoffs are redesigned around the new capability — and McKinsey's 2026 survey found only 6% of organizations have actually done that redesign.
Brief
The chain starts at rollout, where a company licenses or builds an AI tool and makes it available to employees. By 2026, this stage is essentially solved at scale: McKinsey's August 2026 global survey of 1,719 executives across 97 nations found that 89% of organizations regularly use AI in at least one business function, and 44% report enterprise-wide scaling, up from 38% a year earlier. Rollout is no longer the constraint; it has become close to universal.
The second stage — individual productivity gain — is also largely working. The same McKinsey survey found 80% of individual workers report AI has improved their personal productivity, and separate GitHub data cited in industry commentary puts coding-assistant productivity gains as high as 55% for some developer tasks. A National Bureau of Economic Research working paper covering roughly 6,000 executives across the US, UK, Germany, and Australia corroborates that adoption is real but usage is shallow: two-thirds of executives use AI regularly, but average only about 1.5 hours per week, and employees average a similar 1.8 hours weekly. This is the stage where AI genuinely delivers — a faster first draft, a faster code commit, a faster document summary — but the gain is captured entirely inside one person's task time.
The chain breaks at the third stage: workflow and organizational redesign. This is the load-bearing bottleneck. McKinsey's 2026 report found that AI high performers — the roughly 6% of organizations that attribute more than 5% of EBIT to AI — are distinguished not by better models or bigger budgets but because they fundamentally redesign workflows that are enabled by AI rather than insert AI into existing ones, with about three-quarters of high performers reporting this kind of redesign compared with roughly a quarter of everyone else. MIT's Project NANDA GenAI Divide report, based on 300 analyzed deployments, 52 executive interviews and a 153-leader survey, found the same fracture from a different angle: 95% of pilots delivered no measurable P&L impact, while just 5% of integrated systems created significant value — and the report noted that generic tools succeed for individuals precisely because of their flexibility, while they stall in enterprise use because they don't learn from or adapt to workflows. The reason redesign is hard is structural, not technical: a faster individual output only becomes a faster organizational output if the steps before and after that person's task — approvals, handoffs, staffing levels, quality checks — are re-architected to actually use the freed-up time or headcount. Most organizations skip this step because it requires re-engineering job design and process ownership, not just procuring a license.
The fourth stage — measurable P&L or EBIT impact — is where the stall becomes visible in the numbers. McKinsey found that only 37% of respondents said AI had contributed positively to their organization's EBIT, a figure it described as broadly unchanged from 2025 despite adoption climbing over the same period; other 2026 accounts of the same McKinsey dataset describe 63% reporting no measurable enterprise earnings impact at all, with a further roughly 31% reporting some impact below the high-performer threshold. The NBER working paper adds a starker, independently sourced data point: more than 90% of the surveyed executives reported no effect of AI use on employment over the prior three years, and 89% reported no impact on labor productivity at the firm level — even though those same executives, looking forward, expected productivity to rise roughly 1.4% and output to rise roughly 0.8% over the next three years. That gap between backward-looking measured impact and forward-looking expected impact is itself evidence that the stall is diagnosed but not yet resolved industry-wide.
A second, quieter failure mode compounds the redesign bottleneck: ungoverned usage. Because much AI adoption runs through personal accounts and browser-based tools that were never wired into a formal workflow, the productivity gains that do occur often aren't structured in a way that a redesign process could even capture and reallocate — the gains exist, but they're invisible to the systems that would need to convert them into a P&L line.
Components (5)
AI tool layer (LLM chat assistants, coding agents, workflow platforms)
Provides the raw capability — text generation, code completion, document analysis — that gets deployed to individual workers.
Individual worker adoption
Employees use the tool inside their existing job design, capturing personal time savings without changing what happens before or after their task.
Workflow redesign function (or its absence)
The organizational effort — reallocating steps, headcount, and handoffs — that determines whether individual speed gains propagate into a measurably different process.
Measurement and attribution infrastructure
The baseline data and accounting linkage needed to trace a workflow change through to a P&L or EBIT line; without it, gains can occur but remain unattributable.
Shadow/ungoverned AI usage
Employee use of personal AI accounts outside IT-sanctioned tools, which generates real task-level productivity but sits outside any system that could formally reallocate the freed capacity.
How It Works (6 steps)
1Enterprise licenses or builds AI tool
A company procures a generative AI product (chat assistant, coding agent, or workflow-automation platform) or builds one internally, and makes it available across one or more business functions.
IT/procurementBusiness unit leadersAI vendors
Why this step: This is the entry point of the chain — without deployment nothing downstream can occur, and by 2026 this step is essentially universal.
2Individual employees use the tool inside existing jobs
Workers plug the AI tool into tasks as currently designed — a analyst drafts faster, a developer commits code faster — without any change to the process around them.
Individual employeesTeam managers
Why this step: This is where the personal productivity gain is generated and where most organizations' AI story currently stops.
3Gains are captured as personal time savings, not reallocated
The freed-up time or effort from step 2 accrues to the individual (they finish sooner, or produce more per hour) but the surrounding process — approvals, staffing levels, downstream handoffs — is untouched, so the aggregate throughput of the team or function doesn't structurally change.
Individual employeesImmediate managers
Why this step: Without a deliberate redesign step, individual speed has no mechanism to convert into organizational throughput or cost reduction.
4A minority of organizations redesign the workflow
High-performing organizations restructure the process itself — reassigning roles, removing steps AI has made redundant, changing team composition — rather than simply inserting AI into the workflow that existed before.
Executive sponsorsOperations/process ownersChange-management teams
Why this step: This is the step that converts individual capability into organizational capability, and it is the step most enterprises skip.
5Redesigned process changes cost or revenue structure
Once a workflow is rebuilt around AI, the organization can measure changed headcount needs, cycle times, or output volume against a pre-redesign baseline.
Finance/FP&AOperations leadership
Why this step: A measurable financial effect requires a before/after comparison at the process level, not the individual level.
6Financial impact is attributed and reported
The organization's finance function traces the redesigned process's cost or revenue change through to a P&L line and reports it as an EBIT contribution.
CFO/FinanceExecutive leadership
Why this step: Without this attribution step, even a real financial effect may go unrecognized as AI-driven, understating true impact — this is the final gate the McKinsey and NBER surveys are measuring against.
Where It Breaks (4)
Tool deployed without process redesign ('bolt-on AI')
Consequence: Individual speed increases but organizational cost structure, headcount, and cycle time remain unchanged, so no P&L effect is ever generated.
Safeguard: Deliberate workflow-redesign programs tied to leadership commitment and defined success metrics — present in the minority of organizations McKinsey classifies as high performers.
Shadow/ungoverned AI usage outside sanctioned tools
Consequence: Real productivity gains occur but happen through personal accounts and unmanaged tools invisible to IT and finance, making them impossible to formally measure, govern, or reallocate into a redesigned process.
Safeguard: Enterprise AI governance programs and sanctioned tool rollouts designed to be attractive enough to displace personal-account usage.
Absence of pre-deployment baselines
Consequence: Without a documented 'before' state for a process, any productivity change AI produces cannot be measured against it, so the pilot is recorded as having 'no measurable impact' even if real change occurred.
Safeguard: Baseline instrumentation and outcome-metric definition prior to pilot launch.
Misallocated investment toward customer-facing tools while back-office automation offers the larger measured return
Consequence: Budget and attention concentrate where the AI usage is most visible rather than where the workflow redesign opportunity is largest, delaying the stage-three transition for the functions with the biggest potential financial impact.
Safeguard: Portfolio review that weights investment by evidenced ROI category rather than by function visibility.
Why It's Built This Way
Enterprise AI deployment is structured around the assumption that giving individuals a faster tool is sufficient, because that step is the cheapest and fastest to execute; the tradeoff is that skipping the harder, slower step of redesigning the surrounding workflow means the investment optimizes for adoption metrics (seats activated, usage hours) rather than for the P&L outcome the investment was ultimately meant to produce.
The claims behind this analysis, each with its verification status — including what is contested, unverified, or could not be established.
What each grade meansIndividual productivity gains don't automatically aggregate into organizational throughput — A worker finishing a task faster only changes company output if the process around that worker is redesigned to use the freed capacity — otherwise the time saved simply becomes slack.
○ REPORTED
Workflow redesign is the strongest documented predictor of EBIT impact — McKinsey's 2026 survey found that high-performing organizations are distinguished primarily by whether they fundamentally redesigned workflows around AI rather than inserted AI into existing ones.
○ REPORTED
Measurement lag between perceived and measured productivity — Executives report much larger expected future productivity effects than the effects they can currently measure at the firm level, consistent with a delay between when gains occur and when accounting systems can attribute them.
○ REPORTED
McKinsey's August 2026 global survey of 1,719 executives across 97 nations found 89% of organizations regularly use AI in at least one business function, and 44% report enterprise-wide scaling, up from 38% a year earlier.
Establishes that the rollout stage of the chain is no longer the bottleneck — adoption is near-universal, so the stall must be located downstream of deployment.
The same McKinsey 2026 survey found 80% of individual workers report AI has improved their personal productivity, while only 37% of respondents said AI had contributed positively to their organization's EBIT, a figure described as broadly unchanged from 2025.
Directly quantifies the gap between individual productivity gain and enterprise financial impact that this analysis is tracing, and shows the gap did not close year over year.
McKinsey's 2026 report found that roughly three-quarters of AI 'high performers' (organizations attributing more than 5% of EBIT to AI) have fundamentally redesigned workflows around AI, compared with about one-quarter of other respondents.
Identifies workflow redesign specifically as the stage that separates organizations that convert AI usage into earnings from those that don't.
MIT's Project NANDA GenAI Divide report, based on analysis of 300 AI deployments, 52 executive interviews, and a 153-leader survey, found that 95% of generative AI pilots delivered no measurable P&L impact, while only 5% of integrated systems created significant value.
Provides an independent, methodologically distinct data source corroborating the same stall pattern McKinsey documents, strengthening confidence the pattern is real rather than a single survey's artifact.
A National Bureau of Economic Research working paper covering roughly 6,000 executives across the US, UK, Germany, and Australia found more than 90% reported no effect of AI on employment over the prior three years and 89% reported no impact on labor productivity, even though average AI usage was only about 1.5 hours per week among executives.
Shows that shallow, low-intensity usage patterns at the executive level compound the redesign problem, and provides an independently sourced academic data point distinct from both McKinsey and MIT.