AI is not a tech problem.
Ninety percent of firms report no productivity impact from AI. Not because the technology isn't ready. Because they bolted it onto broken workflows and skipped the organizational homework. That's a people and operations problem. Which is what we fix.
The data doesn't match the hype.
An NBER study of six thousand executives across four markets found ~90% of firms report no impact from AI on employment or productivity over three years. Two-thirds of executives use AI — about 1.5 hours a week. A quarter don't use it at all. Meanwhile corporate AI spend crossed a quarter trillion in 2024.
Apollo's Torsten Slok put it bluntly: "AI is everywhere except in the incoming macroeconomic data." That's the paradox in one sentence. Any CEO whose AI budget is tracking above last year's number should stop and look twice at the output.
The gap between narrative and number now has its own literature. The question isn't when the gains will land. It's who captures them.
Quote Torsten Slok, Chief Economist, Apollo Global Management.
Same paradox. Forty years later.
The 1970s and 1980s saw enormous IT investment with almost no productivity gain — until the mid-1990s. The dominant explanation: "lags due to learning and adjustment." IT benefits take two to five years. They only appear when firms reorganize around the technology.
The firms that won the post-1995 boom were the ones that used the lag period to re-engineer how work got done. Not the ones with the most PCs. Not the ones with the biggest IT budgets. The ones that did the organizational homework.
The returns came from the complementary investment — process redesign, new job descriptions, retraining, different performance metrics, different decision rights. Walmart didn't win the 90s because they had better computers. They won because they rewired the retail operating model around what the computers made possible.
The same lesson applies cleanly to AI. The gains are coming. They'll accrue to the firms that use this lag period to reorganize, not the ones still trying to solve the problem with more licenses.
Thesis Organizational investment, not more hardware.
Read Brynjolfsson · Lags-due-to-learning.
What's actually breaking.
The research tells the macro story. The ground tells the operational one. Across hundreds of threads from people inside the rollouts, a cleaner picture emerges — companies are bolting AI onto broken workflows. AI amplifies what's already there. If what's there is broken, you get faster broken.
The real bottleneck is upstream.
Organizations spend months speeding up the doing when the bottleneck was always the deciding. Scoping, prioritization, review, sign-off. AI accelerates the execution layer and leaves the decision layer more exposed. The queue doesn't shorten. It moves upstream.
Seniors get 4x leverage. Juniors lose the learning path.
AI automates the tasks juniors used to learn from: research, first drafts, synthesis. The senior with judgment becomes dramatically more productive. The junior who would become that senior in five years loses the apprenticeship. A three-to-five-year knowledge-work time-bomb. IBM's CHRO has flagged it publicly. Few firms have a plan.
Flagged byIBM CHRO (Fortune).
Faster output. Heavier review.
AI-assisted output is voluminous and uneven. More drafts, more decks, more emails — all generated in the time it used to take to produce one. The reviewer carries more cognitive load, not less. The "productivity gain" quietly transfers to the producer and burdens the person signing off.
Employees save time. The business doesn't see it.
A Stanford study tracked what workers did with AI-saved time. Honest answer: TV and friends. Why would anyone hand efficiency gains back for free without a mechanism? Without an explicit capture mechanism, the gains evaporate into slack. The P&L stays flat.
TakeawayBuild capture mechanisms or capture nothing.
Past three tools, productivity goes backwards.
BCG found that once workers exceed three AI tools, productivity drops from context-switching. The instinct — buy more tools, pilot more vendors, layer more copilots — is self-defeating. Pick two or three per role. Go deep.
CapThree per role. Case required for the fourth.
Five moves that actually stick.
The five moves we push on every engagement. Not glamorous. Not framework-shaped. The organizational homework the research says wins.
Find the real bottleneck first.
For most organizations, the blocker isn't speed. It's scoping, prioritization, review, or billing. We build a bottleneck diagnostic into every engagement so we don't apply AI to the wrong stage. A faster execution layer on top of a broken decision layer is worse than what you had before.
Two or three tools. Go deep.
The BCG brain-fry finding is one of the most practical insights in the literature. Resist the instinct to push ten tools at a team. Pick two or three high-leverage ones. Go deep on workflow integration. The shelf with fifteen tools is a shelf with zero adoption.
Pair every rollout with a redesigned workflow.
Central lesson from the Solow resolution: IT delivered returns only when paired with complementary organizational investment. Translated to AI — no training session ends with "here's the tool." It ends with "here's the redesigned workflow, here's what stops being done, here's how the handoff changes."
When AI frees 20% of capacity, where does it go?
Without an explicit capture mechanism, saved time evaporates. Rationally. The question: when AI frees up a fifth of someone's week, where does it go? Higher-value work? New revenue? A shared pool? Pick one and commit.
Build the apprenticeship back, deliberately.
AI automates the tasks juniors learned from. Unless you rebuild the development path on purpose, you're cooking the future senior bench. Structured problem-solving reps without AI. Required rationale-writing. Supervised AI-output review. A leadership-pipeline risk, not nostalgia.
It's a people and operations problem.
That's what we do.
The homework. While everyone else buys licenses.
The Solow paradox took twenty years to resolve. The firms that won in the 90s used the lag to rewire themselves. The rest were busy buying hardware. That's the pitch.
Everything above is operational work dressed up as a tech rollout. Decision rights. Process redesign. Pipeline development. Compensation structures that capture gains. Review rhythms. Tool discipline. These are the things we've been doing for twenty years, before anyone called it AI strategy.
Our AI Fluency engagement is the operator version of the playbook. A baseline. A leader cohort. Champions in every team. A review process that ships builds in days. Three re-measurements over twelve months. Not a training course — an organizational rewire, with AI as the reason we're doing it now.
If you're a CEO watching the $250B bet not show up on your own P&L yet, we help you be one of the firms where it eventually does.
Read Next AI Fluency Engagement · 7 min.
Do the homework.
First conversation is free. Ninety minutes with your leadership team. We give you an honest read on where the bottleneck is, which move matters most, and whether we're the right hands. No deck. No deliverable. No obligation.