For most of my career, the limiting factor in commercial decision-making was access to expertise.
Companies hired consultants because the analytical capacity to make a good pricing decision, structure a major bid, evaluate a complex procurement, or model a market entry was scarce inside most organizations. Expertise was the bottleneck. The work of consulting was, fundamentally, transferring expertise into specific decisions on specific timelines for specific clients.
That world is ending faster than most people are acknowledging.
Today, an executive can sit down at a screen and obtain — in a single afternoon — ten pricing recommendations for a tariff shock, five procurement strategies for a complex sourcing decision, three market-entry plans for an unfamiliar geography, dozens of competitive analyses, hundreds of pages of synthesized research, draft term sheets, scenario models, decision trees, and risk assessments. The recommendations come from AI systems that are improving rapidly and were nearly inconceivable five years ago.
The supply of recommendations has become, for practical purposes, unlimited.
Yet organizations are not, by all available evidence, making better decisions. Recent industry research consistently shows that only about five percent of organizations report substantial financial gains from AI investment. The capability is real. The decisions are not improving in proportion to it. Many executives privately describe their situation as having more analysis than ever and less clarity about what to do.
That is not a paradox. It is a structural shift.
The bottleneck has moved
The reason organizations are not converting AI-generated abundance into better outcomes is that the constraint is no longer analytical capacity. It is something else.
The ten pricing recommendations conflict with each other. They embed different hidden assumptions about willingness-to-pay, competitive response, segment elasticity, and channel structure. They optimize for different objectives — some maximize revenue, some maximize margin, some prioritize share defense. They ignore organizational realities about what the sales force can execute, what the legal team will sign, what the CFO will tolerate in a quarterly result. They handle uncertainty inconsistently. None of them, on their own, creates commitment among the people who will have to live with the outcome.
What this looks like in practice is recognizable to anyone who has tried. Numbers cited with confident precision — "an estimated 12 to 18 percent margin uplift" — with no traceable basis. Frameworks deployed as wallpaper: Porter's Five Forces applied to industries the model has shallow training data on. Confident assertions about competitive response stated as fact rather than hypothesis. Internal contradictions inside the same response — page two recommends premium positioning, page four recommends penetration pricing, with no acknowledgment of the conflict. Documentation that disappears under interrogation: asked to explain an elasticity assumption, the model produces a different, equally confident answer than the one it gave in the original recommendation. Calibrated-sounding confidence that does not match the model's actual epistemic state. The output is not wrong in any single specific way. It is unanchored in the things that would make it usable for a high-stakes decision.
The abundance of recommendations has actually increased the burden of deciding. Where an executive once had to find the expertise to generate one recommendation, they now have to choose among ten — and the choosing turns out to be harder than the generating.
In economic terms: recommendations have become a commodity. Decision quality has not.
Decision quality is the capacity of an organization to take a flood of input — analyses, recommendations, options, scenarios — and convert it into a defensible, committed course of action that the relevant people will actually execute. It requires:
- making objectives explicit, so different recommendations can be evaluated against the right criteria
- making trade-offs visible, so the organization knows what it is giving up to gain what it values
- making assumptions interrogable, so they can be tested rather than smuggled
- making uncertainty quantifiable, so it can be priced into the decision rather than papered over
- aligning the people responsible for the outcome around the path forward, so the decision actually gets executed
None of that is what AI does. AI generates analysis. Decision quality is a different category of work.
That is the scarcity that has replaced the old one.
A scaffold, not a platform
If the diagnosis is correct, the response cannot be another platform. Platforms ask the organization to adopt an embedded worldview — here is how the work should be done; here is the workflow you will conform to. Every executive who has lived through an ERP, CRM, CPQ, or procurement platform rollout carries the scar tissue. The implementation becomes the project, the use case drifts, and by the time the system is "live" the success metric has quietly shifted from did we make better decisions to did we get the system adopted.
The work of converting recommendation abundance into decision quality is not a workflow problem. It is a structured judgment problem. It needs to happen at the level of a specific decision, with a specific stakeholder group, in a specific competitive context, on a specific timeline.
The right metaphor is a scaffold rather than a platform.
A scaffold is temporary. It supports the construction of something else. It adapts to what is being built — not the other way around. Then it fades into the background, leaving behind what it helped construct.
The scaffold is judged by what it helps build, not by the scaffold itself.
A scaffold around a high-stakes commercial decision can be temporary and still produce extraordinary value. The methodology, the configured analytical environment, the AI-assisted challenge layer, the structured stakeholder engagement, the governance trail — all of it is scaffolding that helps a specific decision get made well. When the decision is made, the scaffold has done its job. What's left behind is the decision, the governance record, and the configured methodology — which can be reused, or not, depending on the organization's choice.
This is what I have been building for the last several years.
A small illustration of the thesis
I want to make one observation about the Decisiums platform itself, because it illustrates the thesis better than any abstract argument.
The Decisiums workbenches operationalize twenty-five years of B2B pricing and procurement methodology — the discrete-choice modelling, the bargaining engines, the evaluation frameworks, the governance layers. The methodology came from that quarter-century of work. The build that turned the methodology into functioning workbenches was accelerated by AI tools used as a development instrument — writing code under direction, surfacing edge cases, drafting against specifications I provided across an extended series of iterative interactions.
The methodology is the asset. AI was the accelerator.
I mention this because the asymmetry is the point. The reason AI was a productive tool in developing these workbenches is that the decision framework already existed before any AI tool was invoked. I knew what a good pricing decision looks like, what an evaluation framework should produce, what a Nash bargaining solution should weigh, what a decision tree should do and where it should refuse to over-collapse uncertainty. That framework — calibrated against years of real engagements with real clients — structured every interaction with the AI tools that helped construct the code. The framework made the AI productive. Without it, the same tools used by someone without the underlying methodology would have produced thousands of lines of code and many screens — but not a coherent commercial decision environment.
That is the same principle the workbenches now operationalize for clients. The methodology is the scaffold. The configured analytical environment is what makes AI's contribution productive rather than just abundant. The structured engagement around a specific decision is what converts AI-generated analysis into decision quality.
The build process and the client engagement model run on the same logic: AI accelerates work against a clear framework; it produces noise without one. The platform is the framework. The Sprint is how that framework gets applied to specific decisions. AI participates throughout — but always inside a structure that exists before it arrives.
That is the thesis, applied. Not theoretical. Demonstrated.
What this means
If you are an executive watching your organization accumulate AI capability without seeing the decision quality improve in proportion, the diagnosis may be simpler than it appears.
Recommendations are abundant. Decision quality is scarce. The capacity to convert one into the other is the work that has to happen, and AI cannot do it on its own — not because AI is weak, but because the work is not the kind of work AI does.
The response is not more AI. The response is structured judgment scaffolded by analytical infrastructure that adapts to specific decisions, includes the stakeholders who must execute the outcome, makes assumptions and trade-offs explicit, and leaves behind something the organization can use again.
That is what is now scarce.
That is also what is now possible, with discipline and the right scaffolding, to do well.