The published data says two things at once: nearly everyone is failing at AI, and the window for the mid-market is not open-ended. This page collects the evidence, the external research, our internal operating record, and the targets we have not earned yet, every number with its source and date, labeled by how sure you should be.
The premise behind everything Milton builds is not opinion. Corporate AI adoption is failing at a published, measured rate, and the published research keeps finding the same cause: the operating model never changed.
of corporate AI pilots deliver no measurable return on investment
MIT, Project NANDA (2025); reported in Harvard Business Review (Nov 2025)
of agentic AI projects projected to be canceled by the end of 2027
Gartner (2025)
65% of executives claim advanced AI understanding; 6% can demonstrate P&L impact
AlixPartners survey of 750 executives, Harvard Business Review (Dec 2024)
of corporate transformations hit challenges severe enough to derail the entire program
Harvard Business Review, survey of 846 senior leaders (Aug 2024)
of organizations have scaled agentic AI across most functions, though nearly 1 in 4 have begun deploying it
McKinsey, The State of AI (Nov 2025)
have AI embedded in the flow of work; 39% still run it as separate, standalone tools
Harvard Business Review Analytic Services survey, n=325 (Dec 2025)
The pattern behind the numbers is consistent: AI treated as an adoption project, a feature purchased, a deck commissioned, a platform built in-house, rather than a reinvention project that changes how the work is executed. The research community has converged on the same verdict: "with AI, as with previous new technologies, reinvention, not adoption, should be your goal" (Harvard Business Review, Nov 2025). Buying a tool and leaving operations unchanged is buying a treadmill for the living room and pointing at it when guests arrive. The failure rate above is what that looks like at market scale.
The largest longitudinal study of real-world AI use reads like a verdict on the feature trap: enormous activity, modest wins, and, in the researchers' own words, "marginal rather than game-changing benefits, so far."
where autonomous agentic operations debuted among the top 100 uses of AI, its first year on the list, and still mostly small-scale experiments
Harvard Business Review, AI in the Wild, 12,637 use cases (June 2026)
of the top AI use cases now involve people delegating some portion of their thinking to the model
Harvard Business Review, AI in the Wild (June 2026)
of organizations expect AI to change 30% or more of their workflows within one year
McKinsey survey (2025)
Two findings explain why the activity hasn't become advantage. First, the failure is organizational, not technical: across 1,200+ automation and transformation cases, technical issues account for only 25% of implementation challenges, organizational and managerial factors account for 75% (London School of Economics, 2024). Second, the human side is being promised and not delivered: 71% of executives say their gen-AI plans include advancing human capabilities, while only 9% report actual progress (Deloitte, Global Human Capital Trends, 2024). And the discipline is not keeping pace with the deployment: only 21% of enterprises have a mature governance model for agentic AI, even as a large majority rush to adopt it (Deloitte, State of AI in the Enterprise, 2026). Usage is not an operating model. A governed workforce is.
This is the existential part. Mid-market companies, $200M to $1B in revenue, are now squeezed from two directions at once: AI-native entrants operating on a fraction of the cost base below, and enterprise incumbents with experimental budgets above. The stakes are not an analyst saving two hours a week. Over the next decade, they are whether the mid-market enterprise survives in its current form at all.
the output an AI-native competitor can deliver with the same resources, or equal traction on one-fifth of the capital
Harvard Business Review (July–Aug 2026)
the minimum AI-native product team, against the conventional six to eight; one AI engineer now does the work of ten
Harvard Business Review (July–Aug 2026)
in annual revenue across roughly 200,000 mid-market companies, largely unserved at the autonomous-agent scale
Internal planning estimate (2026), pending registry validation
The squeeze has a structural cause. A mid-market operator is not equipped to lead an agentic implementation of this magnitude with an internal IT team, and is not a large enough engagement for a tier-one strategy firm to prioritize. So the cohort waits, and the waiting is the risk: an AI-native entrant that ships in days, implements in a quarter of the time, and cuts its legal review costs by roughly 90% (Harvard Business Review, July–Aug 2026) does not negotiate with an incumbent's timeline. The companies that sat out this transition's last equivalent are remembered a specific way, they are the companies that never built a website. The difference this time is that the technology doesn't just take the storefront. It takes the operating model.
The newest valuation research puts numbers on the manifesto. Executives expect AI to more than double firm value, then spend the budget on cost-cutting, which mathematically cannot get there.
the value premium senior executives expect an AI-leveraging firm to command within three years, a 135% increase
Harvard Business Review (June 2026)
the firm-value ceiling of a pure cost-cutting AI strategy, nowhere near the 135% the same executives expect
Harvard Business Review (June 2026)
firm value from a sustained two-point lift in organic growth, the outcome efficiency spending never buys
Harvard Business Review (June 2026)
The scaling research says the same thing from the other side: nearly 90% of companies that successfully scaled AI spent more than half their analytics budgets on adoption, workflow redesign, communication, training, not on the technology itself; only 23% of the rest committed similar resources (McKinsey, in Harvard Business Review, 2019). Better decisions first. Speed, and the valuation, is what happens after the decisions get better.
These are not market projections, they are operating records from the framework running in production, documented as they happened. The full engagement detail lives in the case studies.
of continuous operating history before the first Milton customer, 43 named agents alongside 24 humans
Internal operating record (2024–26)
changelog lines and an 892-page institutional wiki, failures, edge cases, and protocols included
Maintained daily in production
raw-materials inventory cost reduction at the flagship franchise client, against their own baseline
Documented engagement outcome
Truth over narrative cuts both ways: some numbers on this site are targets, not history. They are listed here so no one mistakes one for the other.
The M2 lighthouse design target is a 30–60% function-level improvement against a documented baseline, with a defined share of the engagement fee held against it, if the target is missed, it is named as missed and the credit policy applies. The M4 managed-operations renewal expectation above 95% is design intent for the recurring tier, not yet a measured cohort. And every engagement outcome is committed to publication, anonymized where required, within 90 days of completion, which means this page gets harder to write every quarter, on purpose.
What we cite, where it was published, and what we take from it. If a claim appears on this site without a row here, hold us to it.
| Source | Publication | Year | What we take from it | Tier |
|---|---|---|---|---|
| MIT, Project NANDA | State of AI in Business; reported in Harvard Business Review | 2025 | 95% of corporate AI pilots deliver no measurable return | External |
| Gartner | Agentic AI projections | 2025 | 40%+ of agentic AI projects projected canceled by end of 2027 | External |
| AlixPartners | Digital Disruption Survey of 750 executives; Harvard Business Review | 2024 | 65% of executives claim advanced AI understanding; 18% cutting-edge operational understanding; 6% can demonstrate P&L impact | External |
| Harvard Business Review | Transformation research, 846 senior leaders, 840 employees | 2024 | 96% of transformations hit derailment-level challenges | External |
| McKinsey | The State of AI: Agents, Innovation, and Transformation | 2025 | Nearly 1 in 4 organizations deploying agentic AI; fewer than 10% scaled across most functions | External |
| Harvard Business Review Analytic Services | Workforce-AI collaboration survey, n=325 | 2025 | 12% have AI embedded in the flow of work; 39% standalone tools | External |
| Harvard Business Review | AI in the Wild, 12,637 documented use cases | 2026 | Autonomous agentic operations debut at #6 of 100; 1 in 4 top uses delegate thinking; benefits "marginal rather than game-changing, so far" | External |
| McKinsey | Workflow-change survey | 2025 | 47% expect AI to change 30%+ of workflows within a year | External |
| London School of Economics | Study of 1,200+ automation & transformation cases | 2024 | Technical issues are 25% of implementation challenges; organizational factors are 75% | External |
| Deloitte | Global Human Capital Trends, 14,000 executives, 95 countries | 2024 | 71% plan to advance human capabilities with gen AI; 9% report progress | External |
| Deloitte | State of AI in the Enterprise | 2026 | 85% of enterprises plan to customize agents to fit their business | External |
| Deloitte | State of AI in the Enterprise | 2026 | Only 21% of enterprises have a mature governance model for agentic AI | External |
| Harvard Business Review | How Agentic AI Supercharges Startups and Threatens Incumbents | 2026 | AI-native competitors: 5x output or one-fifth capital; 2-person product teams; ~90% legal-review cost reduction | External |
| Harvard Business Review | Companies Are Using AI for Efficiency. They Should Use It to Grow. | 2026 | Executives expect a 2.35x AI value premium; cost-cutting strategies cap near +10%; a 2-point organic-growth lift adds ~50% firm value | External |
| McKinsey | Building the AI-Powered Organization, Harvard Business Review | 2019 | ~90% of successful AI scalers spent over half their budgets on adoption, not technology; only 8% of firms practice what scaling requires | External |
| IDC | Worldwide AI Spending Guide | 2024 | Enterprise AI market of $68–90B by 2028 | External |
| McKinsey | The economic potential of generative AI | 2023 | Trillions in value-creation potential across business functions | External |
| Grand View Research | AI Governance Market report | 2023 | AI-governance market of $1.4B by 2030 (26.5% CAGR) | External |
| Grand View Research | Artificial Intelligence Market report | 2023 | Global AI market of $1.81T by 2030 | External |
| U.S. Census Bureau | SUSB Annual Data Tables | 2021 | 1.3M+ U.S. professional & business services establishments | External |
| Milton, Inc. | Internal planning model | 2026 | ~200,000 mid-market companies worldwide, $10T+ in annual revenue; ~12,000 in North America, pending registry validation | Internal |
| Milton, Inc. | Internal operating record | 2024–26 | 18 months · 43 agents : 24 humans · 892-page wiki · 50,000-line changelog · documented case outcomes | Internal |
| Milton, Inc. | Commercial design, M-ladder | 2026 | 30–60% M2 design target with credit policy · >95% M4 renewal intent | Target |
Mid-market cohort sizes (~200,000 worldwide, ~12,000 North American firms at $200M–$1B revenue) are internal planning estimates pending validation against commercial registries, labeled Internal above, per the validation discipline.
If a number on this site does not carry a source and a date, it does not belong here.
Tell us the function that is eating your team's time. We'll talk through what is going on, whether Milton is a fit, and what a first engagement would look like. A senior person, one real conversation, no demo and no pitch deck.