Every AI conversation in the mid-market eventually arrives at the same word: faster. Faster drafts, faster tickets, faster close. The instinct is understandable and the instinct is the problem, because the numbers on either side of it refuse to reconcile. Executives expect an AI-capable firm to be worth 2.35 times a comparable one within three years, a 135% premium, while the efficiency programs they actually fund top out, under generous assumptions, at roughly a 10% lift in firm value (Harvard Business Review, June 2026). Faster cannot carry that load. It was never going to.
The case for a different target starts with an uncomfortable observation about what is actually broken inside most organizations. The technology works. The pilots still fail, 95% of them deliver no measurable return (MIT, 2025). A failure rate that high, against tooling that good, means the constraint is somewhere other than the tooling.
The goal was never speed. Speed is what happens after the decisions get better.
Most organizations have a truth problem, not a technology problem
Sit in the Monday operating review of a typical $400M company and watch what the executives actually argue about. Rarely the decision itself, usually the facts beneath it. Sales has one number for pipeline, finance has another, operations has a third, and twenty minutes evaporate establishing what is true before anyone can discuss what to do. The information exists. It is simply fragmented across people, systems, and workflows, and the fragments disagree.
That is the truth problem, and it is the real constraint on performance in the mid-market. Decisions get made late because assembling the facts takes days; decisions get made wrong because the assembly was partial; decisions get avoided because nobody trusts the assembly at all. No drafting assistant touches any of that. A workforce of named agents with standing access to the underlying systems does, because its first job is not producing work product faster, it is making the organization's truth available on demand.
of corporate AI pilots deliver no measurable return, against models that keep improving every quarter, which means the constraint is the target, not the technology
MIT, Project NANDA (2025)
Better decisions first. Speed is what happens afterward.
Effectiveness has a precise meaning here: the organization chooses better, not just produces faster. The distinction sounds philosophical until it shows up on an engagement clock. In one Milton deployment, finance questions that previously took days to answer, pulling, reconciling, chasing, now come back in minutes (documented engagement outcome). The minutes are not the point. The point is that the executive asked a second question, and a third, and made a different call than the one the stale version of the data would have produced.
The same pattern holds at higher stakes. Intelligence reports analyzing hundreds of data points, work that once consumed a week of an analyst's calendar, now complete in under an hour (documented engagement outcome). One client converted that capability directly into revenue: a $2M international account secured because the intelligence gathering moved at deal speed instead of analyst speed (documented engagement outcome, client anonymized). Notice the causal order in every case. The decision got better because the truth got complete; the speed was a byproduct. Efficiency programs run the arrow backward and wonder why the value never appears.
Effectiveness is measured in decisions and growth, not hours
An efficiency program reports hours saved, and hours saved is a metric with a known pathology: it gets reabsorbed. Saved time becomes more meetings, more polish, more inbox, and twelve months later the board asks where the outcomes are and receives a usage dashboard. An effectiveness program reports different units: decision latency, from question asked to question answered against complete data; decision quality, measured by how often the post-mortem finds the facts were wrong; and ultimately organic growth, where a sustained two-point lift adds roughly 50% to firm value (Harvard Business Review, June 2026).
Those units force different design choices. An agentic workforce built for effectiveness gets deployed where decisions concentrate, pricing, supply, accounts, capital, rather than where keystrokes concentrate. Its agents are named, scoped, and paired with the senior people whose judgment they feed, because the output that matters is the operator's call, not the agent's document. Milton's M2 stage is designed against a 30–60% function-level improvement target with a credit policy if the target is missed, a design target rather than a guarantee, but priced like a commitment because effectiveness, unlike adoption, is measurable (internal operating record).
The manifesto, stated plainly
AI applied to efficiency makes the existing organization marginally cheaper, and the arithmetic caps the prize near 10% of firm value (Harvard Business Review, June 2026). AI applied to effectiveness changes what the organization knows at the moment it decides, and decisions are the only mechanism through which any company has ever grown. One of these is a procurement exercise. The other is an operating model.
So the position is this. The goal is not the same work faster; it is better decisions, with speed as the visible symptom of an organization that finally agrees on what is true. The 135% expectation executives carry is not naïve, it is an accurate intuition about what complete, current truth would be worth to a firm that acts on it. The blindspot is funding that intuition with a typing-speed budget. Point the fleet at the truth problem, measure it in decisions and growth, and the efficiency takes care of itself on the way through.