.svg)
.jpg)
Tools have been bought. Pilots have been running. Boards have been briefed. And for most companies, the line on the P&L that AI was supposed to move has not moved.
The published research keeps describing the gap, indifferent words, the same way. Some firms have converted AI investment into measurable value. Most have not. The split is not random, and it is not technology. The two groups are running the same models, on similar data, often through the same vendors. The difference is what they are spending their effort on.
AI ROI problems are not technology problems. They're people and process problems.
Two years into the enterprise AI cycle, the small group of firms converting AI investment into measurable P&L spends the majority of their effort on people and process, not on technology. Most firms run the opposite mix. This blog explores what the inverse looks like in practice so you can create an AI adoption strategy that delivers the AI ROI you expect.
The numbers have become hard to argue with. McKinsey's 2024 State of AI found that the small group it calls AI high performers, roughly 6% of respondents, attribute more than 20% of their EBIT to AI use. By 2025, only 39% of organizations could attribute any EBIT impact to AI at all, and among those, most reported less than 5%.
BCG's read on the pacesetters has hardened into a rule. In its Leader's Guide to Transforming with AI, BCG describes a 10/20/70 split: 10% of effort on algorithms, 20% on technology and data, 70% on people and processes. That is the pacesetter pattern. The conventional enterprise pattern is closer to the inverse. Roughly 80% of AI spend lands on the technology layer. The 20% that would actually move the P&L gets done last, if it gets done at all.
That is the structural failure underneath two years of disappointing AI returns. It is not a model problem. It is not a data problem. It is a spend-mix problem upstream of both.
AI transformation pays off when capability is rebuilt before technology is deployed.
Three moves close the gap:
What follows is what those three moves look like in practice, plus how to read whether the value is actually accumulating. None of it requires writing off the technology spend that already happened. It does require redirecting the next dollar.
Across the engagements we have run, the three structural failure modes that consume the technology spend are consistent.
Failure 01. Automating the wrong thing. The work is faster, the job is done worse. Edge cases the human handled informally disappear. The consequence surfaces months later, and gets blamed on the model.
Failure 02. No verification discipline. AI output moves directly into action. Errors compound at machine speed. Trust erodes silently because there is no instrumented step between the answer and the irreversible commitment.
Failure 03. No improvement loop. Failures get fixed individually. The system that generated them stays exactly as broken. The next case starts from the same place.
Most organizations don't have a technology problem. They have an adoption problem.
The organizations generating measurable returns from AI are not deploying fundamentally different tools. They are building different capabilities around those tools. They have an operating model that connects AI investments to business outcomes through workforce adoption, governance, and continuous improvement.
Across our work with enterprise clients, we see the same pattern emerge repeatedly: successful AI transformation follows a sequence. Organizations identify the job to be done, prioritize opportunities based on business value and adoption feasibility, build the capabilities employees need to work effectively with AI, and measure progress across multiple transformation cycles rather than a single pilot.
Taken together, these practices form an AI adoption strategy that closes the gap between technology investment and business value.
The four pillars that follow provide a practical framework for moving from experimentation to enterprise-scale adoption. They are not technology choices. They are the people-and-process work that turns the technology spend into a P&L line.
Before redesigning a process, you have to name what the work is for. Most organizations have never written it down.
Two artifacts make a redesign decision defensible. The first is a job statement: what the work is actually trying to accomplish, what the real quality standard is, what variation matters. The second is a time study: what the work currently costs. Together they replace the ROI projections that today are mostly guesses.
The 20-minute interview pattern we use is five questions, in this order:
By the twentieth minute, an interview that started with "expense reports take too long" has produced a job statement(catching policy edge cases before they reach Finance), a failure-mode map, a complexity distribution, and an inventory of invisible work nobody currently owns.
Not every process is worth redesigning, and the ones that are usually are not the ones the technology team has already picked. Use cases get prioritized by what is fastest to demo, not by what is most material to the P&L. That is how organizations end up with a portfolio of pilots that all work and none of which matter.
The 2x2 we use plots business value against DVT-adoption feasibility, how realistic it is for the people who do the work to direct and verify the AI version of it. The risk-tier overlay tells you how much governance each use case actually needs.
The four quadrants give a defensible sequencing decision. Start here is high value, high feasibility; first pilots, move fast. Invest before you deploy is high value, low feasibility; build the verification and direction capability before you ship, and accept that the payoff lands in the next cycle. Deprioritize is low value, high feasibility; technically interesting, strategically irrelevant, and where most AI showcase pilots quietly live. Don't touch is low value, low feasibility; no business case, no adoption path, leave them.
Risk tiers cut across the matrix. Tier 1 is autonomous AI action with periodic verification. Tier 2 is AI-assisted decision with verification required before action. Tier 3 is human decision with AI support, with explicit authorization regardless of model confidence. The tier dictates how much verification overhead the use case actually costs, which is why it changes the feasibility read.
Why this matters for the budget: sequence is strategy. Two firms with identical AI portfolios and identical platforms will produce very different P&L outcomes depending on which use case they ship first, second, and tenth.
This is the part that consumes the 70%, and it is the part most programs underbudget.
Organizations need a practical AI operating model built around three human roles — Director, Verifier, and Transformer (DVT) — working in a continuous cycle with AI execution to scale intelligence across the organization. It defines how humans direct AI systems, verify outputs, and continuously improve the intelligence powering the organization.
DVT is not three steps in a process. It is three competencies in every job and employee. Each one closes one of the structural failures named above. Here's how it works:
These three competencies are what employees in a pacesetter firm acquire. They are not roles; they are capabilities. The accountant who uses an AI agent to draft a memo is being a Director when she writes the prompt, and a Verifier when she reads the draft. If she reports the recurring error pattern back so the prompt and the workflow improve, she is being a Transformer.
Microsoft has named the role: agent boss. BCG has named the discipline: people and process. We have named the operating mechanic: Direct, Verify, Transform. They are pointing at the same shift.
Why this matters for the budget: the technology investment will not produce value unless the people doing the work can direct it, verify it, and improve it. That capability does not come from a license. It comes from deliberate practice, instrumented by governance, supported by the kind of training most AI rollouts treat as an afterthought.
The most common reason an AI program gets killed in cycle one is that the leadership team measured the wrong thing. Cycle one is not where P&L impact shows up. It is where capability and governance maturity get built.
In cycle one (Establish), human intervention is high. Verification overhead is high. Modest value is fine. The thing being measured is whether DVT capability and governance discipline are landing in the affected functions. If you measure cycle one on P&L, you cancel the program before it has had a chance to compound.
In cycle two (Accelerate), the improvement loop pays off. Verification overhead drops on routine cases because the Transformer step has hardened them. Value accelerates visibly. This is the first cycle where the financial case starts to look credible.
In cycle three and beyond (Compound), DVT is instinctive. Governance trust extends to higher-stakes jobs. The competitive asset gets built here, not in cycle one. By the third cycle, the gap between a pacesetter and a stuck firm is no longer about which AI vendor either one chose. It is about which one built the capability around it.
Why this matters for the budget: the return on AI transformation is not a milestone. It is a trajectory. Cycle-one boards that ask for cycle-three numbers will get neither.
If you came out of Cycle 3 with one number, this is the one to leave with: pacesetters allocate roughly 70% of AI effort to people and process. If your AI budget today is closer to 80% technology and 20% everything else, you are running the laggard pattern. The fix is not to spend more. It is to redirect the next dollar.
What that looks like on the ground:
The technology you already bought is not the problem. It is in the wrong place in the budget.
The gap between AI investment and AI ROI is rarely closed through technology alone. It is closed through a series of practical decisions that align people, process, and adoption around the work that matters most.
Here are three actions leaders can take this week to begin closing the gap between AI ambition and business outcomes.
The next wave of AI transformation will not be won by organizations with access to better models. Increasingly, everyone has access to the same technology.
The differentiator is adoption.
The organizations creating measurable AI ROI are building the people, processes, and operating models required to turn AI capability into business outcomes. They understand that AI transformation is not a technology initiative with a change-management workstream attached to it. It is an organizational transformation enabled by technology.
The 70/30 reality is becoming increasingly clear: technology creates potential, but people and process create value.
The companies that close that gap will move beyond pilots, accelerate enterprise AI adoption, and build lasting competitive advantage.
If you're evaluating where your organization sits on the AI adoption curve, Launch can help. Our team works with enterprise leaders to identify high-value AI opportunities, design scalable operating models, and build the Direct, Verify, Transform capabilities that turn AI investments into measurable business outcomes.
Connect with a Navigator to assess your highest-impact opportunities and create a practical path from experimentation to enterprise-scale value.
