Scaling looked rational.
AI handled two-thirds of chats, resolution time fell from 11 minutes to 2, satisfaction was reported on par with human agents, and profit improvement was projected at $40 million.
I turn ambiguous AI rollouts into decisions teams can act on.
In this Klarna case study, the headline metrics made expansion look rational. I identified the missing constraint—human exception capacity—and translated it into a bounded decision, an evidence plan, and a reassessment rule.
A strong AI pilot hid a capacity decision.
I reframed the question from “Is the AI performing?” to “Is the operating system ready for the next commitment?”
The proof stopped at aggregate performance.
Complex-case quality and escalated backlog were not separately measured, even as the human capacity receiving those cases was shrinking.
Hold the next increment—not the AI program.
Reduce scope to demonstrated segments, protect exception capacity, and make further expansion conditional on segment-level evidence.
Turn uncertainty into a release plan.
Name the binding constraint, set clearing conditions, assign the intervention to the business case, and define when the decision should be reassessed.
What this demonstratesProblem framingSystems thinkingMetric architectureAI product judgment
A contemporaneous decision · early 2024
Klarna: make the call before seeing what happened later.
Only evidence available at the boundary is shown until your judgment is complete.
01 / 06 · Present the decision
⅔ chats handled by AI11 → 2 min resolution time$40M projected profit improvement
The AI pilot is outperforming expectations.
Should Klarna deepen the AI mandate while continuing to reduce human support capacity?
Make the decision as it appeared in early 2024. Later outcomes remain unavailable.
The supporting model
The structure I used to find the gap.
StratOS separates the enterprise question, the operating altitude, and the location of proof so conflicting signals cannot disappear inside one score.
Enterprise questionsThe primary conceptual axis
Does this system create sufficient economic value to sustain itself?
Can the organization deliver what it has committed to?
Can the organization keep adapting as its environment changes?
Operating altitudesArchitecture and mechanics stay distinct
What machine are we building?
Where is the running machine succeeding or failing?
Loci of evidenceWhere the proof lives
What is happening inside the organization?
What is the organization causing outside itself?
3×2×2=12 poles
How the model emerged
StratOS did not begin with 60 metrics.
It began with two existing structures: an L1–L5 enterprise strategy framework, introduced through an experienced product and operating leader, and a set of 12 common C-suite roles.
I mapped those roles by the resources and organizational capacities they governed, whether their decisive evidence lived inside the company or in the market, and whether their signals behaved more like leading or lagging indicators.
Why twelve?
The goal was not to invent a taxonomy. It was to discover whether the roles shared a smaller structure. They resolved into three enterprise questions, two operating altitudes, and two loci of evidence.
The final axis evolved beyond lead versus lag. Its more useful property was location of proof: internal condition and external consequence could diverge.
Action after diagnosis
Divergence says where to look. L1–L5 says where to intervene.
The 3×2×2 model identifies what dimension of enterprise condition is in question. L1–L5 identifies the level at which that condition should be governed or changed.
Strategy
What are we trying to accomplish?
Business case
What must be true before we commit?
Implementation
What must be true before we launch?
Operations
What must remain true while we run?
Audit
Did the claimed outcome actually occur?
Escalated work may accumulate while receiving human capacity shrinks.
The missing capacity gate belongs in the architecture of the commitment.
Explore the 60-cell accountability model
12 perspectives × 5 levels = 60 accountable outcomes. The matrix assigns every material question an owner, lifecycle stage, evidence requirement, and consequence.
The 60 cells are measurement depth, not the opening hook. They become useful after the 3×2×2 model has located the disagreement and the decision has named the intervention level.
Why I built this
I’m strongest at the front end of ambiguous product problems.
Finding the underlying structure, defining the right conceptual model, and turning it into something teams can build, measure, and make decisions with.
StratOS is my attempt to make complex organizational decisions legible by showing where strategy, operations, and evidence agree—and where they diverge.
Problem framingConceptual modelingSystems thinkingMetric architectureProduct judgmentAI product thinking
Product strategy · 0→1 product management · AI product management · technical product management · decision intelligence · platform product management · enterprise AI product · product operations strategy · applied research / product