The Input Certainty Framework

Not all cognitive work is the same. The most important variable determining how AI should be deployed — and how AI products should be designed — is not the industry, the function, or the seniority of the user. It is the degree to which the problem is defined before the work begins.

The primary axis

Cognition leverage is not a single mode. It operates across a spectrum defined by input certainty: the degree to which problem boundaries, constraints, and success criteria are specified prior to engagement.

At one extreme, input certainty is minimal. The mandate is open. The problem itself must be discovered, framed, and structured before it can be solved. This is the domain of generative discovery — where the cognitive task is to expand the possibility space, surface non-obvious options, and construct the frame through which the problem will eventually be addressed.

At the other extreme, input certainty is high. The mandate is defined. The constraints are known. The success criteria are explicit. This is the domain of deterministic execution — where the cognitive task is to compress the defined brief into a precise, high-quality output with maximum efficiency and minimum human time invested.

Two modes, distinct metrics

The distinction matters because the two modes require entirely different AI product designs and are measured by entirely different success signals.

In generative discovery mode, depth of engagement is a legitimate positive signal. The value is being created through the iterative process itself — through hypothesis generation, stress-testing, reframing, and synthesis across a wide option space. A management consultant building a new market entry thesis, a chief strategy officer evaluating a platform shift, a venture investor constructing a sector thesis — these professionals are paid to think through problems that do not have clean edges. For them, AI is a thought multiplier. The metric is output quality per cognitive hour invested.

In deterministic execution mode, engagement is a cost, not a signal. The value is created by the speed and precision with which a defined input is converted into a high-quality output. A capital markets analyst executing a structured mandate, a risk officer reviewing against a compliance framework, legal counsel working through a defined contract — these professionals operate against pre-specified criteria. For them, AI is a compression engine. The metric is non-engagement plus repeat rate.

The secondary axis

A second variable reinforces the primary one: who bears the cost of token consumption. In generative discovery, the individual typically absorbs the cost — the exploration is billable work, the consultant's time is the product, and the depth of engagement is directly monetisable. Institutional tolerance for open-ended AI usage is high.

In deterministic execution, the institution bears the cost. A capital markets desk, a compliance function, an operations team — none of these will tolerate token waste on well-specified problems. The institutional buyer's willingness to pay is calibrated directly against time saved per decision cycle, not richness of dialogue.

Implications for AI-native product strategy

The input certainty framework resolves a question that plagues AI product teams: what are we actually building, and how do we know if it is working?

The answer begins with an honest classification of the user's position on the spectrum. An AI product serving generative discovery users must optimise for exploration depth, idea quality, and synthesis richness. Engagement metrics are not the enemy — they are directionally correct. An AI product serving deterministic execution users must optimise ruthlessly for compression: minimal input required, maximum output quality, fastest possible path to a decision-ready artifact.

Confusing the two is a product strategy error with commercial consequences. A compression product that behaves like an exploration tool frustrates time-pressured institutional buyers. An exploration product with engagement-phobic design strips away the iterative process that generates its core value.

At Potentium, the classification is unambiguous. Decision infrastructure for capital movement sits firmly in the deterministic execution domain. The mandate, the instrument, the risk parameters, and the success criteria are all specified before engagement begins. The product's job is to absorb that specification, run the cognitive heavy lifting internally, and return a decision-ready output with minimum demand on the user's time.

The North Star is not engagement. It is the quality of the decision, the speed of its delivery, and the user's willingness to return with the next one.