The 17% problem
A March 2026 Medallia report found that only 17% of consumers believe their experiences are getting better, even as AI accelerates the pace of design and delivery. Organizations feel the momentum. Teams are shipping faster. Prototypes that once took weeks are taking hours. And yet the people on the receiving end are less confident, not more. Something is not adding up.
The Balanced Scorecard was designed for exactly this kind of disconnect. Robert Kaplan and David Norton introduced it in the early 1990s as a way to stop financial metrics from eating everything else. The framework asks organizations to measure performance across four lenses at once: financial outcomes, customer experience, internal process efficiency, and the capacity to learn and grow. The idea is that optimizing one lens at the cost of the others is not a strategy. It is myopia with a spreadsheet.
Right now, AI adoption is a myopia problem. Teams are measuring the internal process lens obsessively. Cycle time is down. Output is up. The workflow looks clean. The other three lenses are an afterthought, and in complex, high-stakes systems, that gap is not just a performance risk. It is a design failure waiting to happen.
The cost nobody accounts for
Here is where it gets specific. In regulated industries, the most common AI failure mode is not a bad launch. There is no launch. Features die in security review. Integrations stall in compliance assessment. Proposals get tabled because no one can honestly answer what happens when the model is wrong and the downstream consequence is a billing error for 40,000 customers or a missed inspection flag in critical infrastructure.
There is no incident report for a feature that never ships. No postmortem. No lesson learned entry in a project tracker. The Balanced Scorecard records nothing because nothing happened. But the cost is real. Engineering hours are spent. Design cycles are burned. Organizational trust in the AI development process quietly erodes. The accounting system was not built to see any of it.
This is the pattern I am watching develop across highly regulated industries. The applications were too complex, the regulatory surface too wide, and the internal confidence too low. Teams were not anti-AI. They were appropriately cautious in a domain where a wrong answer is not a bad recommendation. It is a liability.
What is encouraging is that some teams are starting to recognize this. Risk conversations that used to surface in legal review are moving earlier, into discovery and scoping. That shift does not eliminate the problem, but it changes the economics. A risk conversation at week two is a design constraint. The same conversation at week ten is a sunk cost.
The fifth lens
By 2030, I think mature organizations will add a fifth lens to the Balanced Scorecard: Security and Trust. Not as a gate that blocks AI, but as a measurement perspective that prices risk honestly before a decision is made.
Managerial accounting has a concept called relevant cost thinking. The idea is that good decisions require stripping out what has already been spent and asking only what changes from this point forward. What is the actual financial exposure if this AI integration fails? What happens to customer trust if it is breached? What does the internal process look like when the model produces a confident wrong answer and no human is positioned to catch it?
A fifth lens does not slow teams down. It forces the questions that are already slowing teams down into a structure where they can actually be answered. The teams figuring this out now will not just ship more. They will ship work that holds up. That is a meaningful competitive advantage, and it compounds.