A System, Not a People Problem

In March 2026, Victor Kiani completed a case study at New York University on why employees leave and what actually keeps them. Leadership wanted a mentoring program to fix retention. The data pointed to something they weren't ready to hear, and saying it required as much strategy as finding it.

The scenario

Global Growers, a mid-size company, wanted a mentoring program for new hires, which leadership believed would improve retention. Before recommending anything, Victor Kiani gathered feedback from three groups.

What leadership believed was working:

• Some leaders provide short-term cross-functional projects.
• Some encourage employees to use the company's online learning program at their own pace.
• All review internal promotion candidates quarterly. But they have not found suitable internal candidates.

What they weren't hearing, from current employees:

• Employees do not consistently discuss career development with managers.
• Employees feel they work in silos.
• Leaders support conference attendance, but employees pay all costs themselves.

Why people actually left, from former employees:

• They "did not find opportunities for growth."
• Managers "prioritized completing tasks instead of developing talent."
• Leaders "frequently required that employees work on weekends." Burnout was widespread.

Diagnosis

Leadership believed a new mentoring program would fix sentiment and retention. The data told a more complicated story.

He recommended three structural changes, grounded in Bauer & Erdogan's organizational-behavior framework:

1. 01Formalize the mentoring program: clear expectations, regular meetings, cross-team mentor matches.
2. 02Manager-led career conversations: regular development discussions, not just task assignments.
3. 03Address burnout structurally: weekend-work culture and lack of growth support are driving people out.

The challenge

His recommendations were already written; the risk was in the delivery. How do you tell leadership their development system is broken without them hearing "you failed"? So he used the delivery to test a question: could AI refine a hard message for a resistant audience, or would it just sand the truth down to nothing? Kiani stayed the editor, not the author, judging every output and overruling the model when it went soft.

Refinement: two prompts, three versions

He ran the revision through two prompts. The first optimized for tone and over-softened the message; the second pulled it back toward specifics and evidence:

1. 01Original (direct)"Will not succeed if leadership treats mentoring as a stand-alone solution." Clear and evidence-based. But could trigger defensiveness.
2. 02Gemini v1 (over-softened)"We have an opportunity to integrate mentoring into a more cohesive development framework." Diplomatic, but it lost its edge.
3. 03Gemini v2 (balanced)Acknowledges leadership's intent while redirecting toward a broader strategy. Evidence and citations intact.

Result

Design principles. Three choices turned the criticism into something leadership could act on:

Why it matters

• AI over-softens by default. Left alone, it sands the edge off a hard truth. Knowing when to push back on it is what keeps a message both honest and hearable, the difference between feedback that lands and feedback that gets ignored.
• Iterate; don't ship the first draft. Two passes beat one. The second is where a difficult message becomes something leadership can actually act on.
• AI is a revision partner, not the author. The thinking has to stay human; the model is for pressure-testing how it lands, not for having the thought.