When Tolerance Is Not a Number

When a product team is eighteen months into a development program and the clinical team still can't get locked specifications from engineering, the problem rarely shows up on anyone's risk register. It shows up as a regulatory delay — and by then, the schedule has no room left to absorb it.

A mid-size hearing aid manufacturer was developing a behind-the-ear device with an AI-powered sound processing module. The regulatory path required clinical validation data, which meant the clinical team needed final design specifications to build test protocols against. Engineering kept telling them the specs weren't final yet. The AI research group kept improving the algorithm, and each improvement changed the hardware requirements.

The Real Problem

No one had told the AI research group that regulatory timeline was the constraint that governed the program — not performance, not feature completeness, not hardware elegance. The product manager responsible for the launch timeline was getting questions about delivery dates she couldn't honestly answer. The issue wasn't execution. It was that the organization had never decided, in writing, which constraint governed when they conflicted.

Making the Governing Priorities Explicit

An Intent Management™ session with the program's core leadership produced a governing outcome statement that reordered everyone's priorities in writing: deliver a clinically validated device that meets FDA clearance requirements by the target submission date, with AI performance that is competitive at launch. The ordering of that sentence mattered. Regulatory timeline was the constraint. AI performance was the goal within that constraint.

The algorithm version in use at a specific program milestone would be locked for regulatory purposes, even if subsequent iterations showed performance improvements. This was a significant concession from the AI research group, and it required executive sponsorship to hold — the CMO and VP of Engineering both signed off on it as a written boundary.

The firmware team's simulation environment had been generating tradeoff recommendations based purely on acoustic performance metrics. Once regulatory dependencies were mapped and fed into the simulation constraints, the system stopped suggesting algorithm changes that would have required re-validation.

What Changed

The clinical team got the design lock they needed eight weeks earlier than they would have under the previous approach — enough to preserve the submission timeline. The AI sound processing module that shipped was not the highest-performing version the research team had developed. It was the best version that could be clinically validated, manufactured, and submitted for clearance on the timeline the business required.