Disagreeing With a Stakeholder

Managing Technical Conflict Constructively

In machine learning engineering, disagreements occur regularly:

• PM vs ML Engineer: Product Manager wants a 70B parameter LLM for maximum capability, but ML Engineer knows it breaches the 50ms latency SLA and hosting budget.
• Data Scientist vs Data Engineer: Data Scientist wants real time streaming features, but Data Engineer advocates for simple batch features to lower operational complexity.

Interviewers ask "Tell me about a time you disagreed with a stakeholder" to evaluate collaboration, empathy, influence, and business maturity.

Ineffective Approach: "I told the PM their idea was technically stupid and refused to build it." (Toxic!)
Effective Approach:   "I listened to their goal, ran a fast 1-day benchmark showing the 300ms latency impact, proposed a hybrid alternative, and we aligned on user metrics." (Professional!)

The 4 Step Conflict Resolution Framework

┌──────────────────────────┬──────────────────────────┬──────────────────────────┬──────────────────────────┐
│ 1. LISTEN & EMPATHIZE    │ 2. FRAME BY USER GOALS   │ 3. EMPIRICAL BENCHMARK   │ 4. DISAGREE & COMMIT     │
├──────────────────────────┼──────────────────────────┼──────────────────────────┼──────────────────────────┤
│ Understand the underlying│ Translate technical      │ Build a fast prototype   │ Once a decision is made, │
│ business goal behind     │ constraints into product │ to present objective     │ execute with 100% team   │
│ their request.           │ metrics (latency/cost).  │ data over opinions.      │ alignment.               │
└──────────────────────────┴──────────────────────────┴──────────────────────────┘

1. Listen and Empathize First

Acknowledge their objective. A Product Manager pushing for high accuracy is trying to improve user satisfaction, not make your life difficult.

2. Frame Arguments Around Shared Business Goals

Avoid technical jargon ("gradient explosion"). Frame arguments in terms of user experience and business metrics:

• "If we use this larger model, P99 search latency increases from 40ms to 250ms, which historically causes a 5 percent drop in conversion."

3. Let Data Settle Subjective Debates

Instead of arguing over opinions, build a fast 1-day prototype or A/B benchmark. Empirical data converts subjective arguments into objective facts.

4. Disagree and Commit

Vigorously debate during the planning phase. But once a final decision is made by leadership or team consensus, commit 100 percent to making the chosen path successful.

Say this out loud

Resolving stakeholder disagreements requires active listening, framing technical choices around business metrics, and presenting empirical benchmark data. Rather than arguing opinions, build fast prototypes to let objective metrics guide decisions. Once a team decision is reached, follow Disagree and Commit to support execution.

Followups to expect

1. What if data shows your original proposal was wrong? Celebrate the outcome! Eagerness to accept data over ego demonstrates senior engineering maturity.
2. When should you escalate a disagreement to senior leadership? Escalate only when an unresolvable impasse threatens core security policies, legal compliance, or major architectural stability.