Owning the AI Roadmap: Prioritization, Saying No, and Stakeholder Alignment

AI work needs explicit research vs. product separation. Every no needs a why and a when. Making tradeoffs explicit (not just declining) is the skill that separates good EMs. The reversibility test for deprioritization decisions.

The AI team roadmap is a political document as much as a technical one. Every item on it represents a decision about what you are not building. The EM who can't say no ends up with a team that ships nothing important because they're busy shipping everything urgent.

The Prioritization Problem in AI Teams

AI work has two failure modes that pure software doesn't: uncertainty (you don't know if the approach will work until you try) and asymmetric effort (a 1% model improvement might take 3 sprints while a 10% improvement takes 1 sprint — you can't predict it from the outside).

• Don't commit to outcomes, commit to time-boxed experiments with clear exit criteria. Be explicit about what 'done' means before the work starts — accuracy target, latency SLA, fallback behavior. Reserve 20–30% of sprint capacity for unplanned work. AI systems generate incidents unpredictably. Separate research work from product work on the roadmap. Different risk profiles, different review cadence.

The Saying-No Framework

Every no needs to come with a why and a when. 'We can't do that' is a dead end. 'We can do that in Q3 after we stabilize the retrieval pipeline, and here's why that sequencing matters' is a conversation.

The Three-Bucket Prioritization

• Must ship: missing it has a hard external cost (compliance, contractual, safety). Should ship: clear ROI with validated user need and implementation confidence. Explore: uncertain ROI, time-boxed, explicit learning goal. Kill it cleanly if the hypothesis fails.

Stakeholder Alignment Without Capitulation

Product managers, business stakeholders, and senior leadership all have legitimate but different views of what the AI team should build. Alignment doesn't mean consensus — it means everyone understands the prioritization rationale and knows the escalation path when they disagree.

• Publish the roadmap and its rationale, not just the items. Transparency prevents the 'why aren't you building X' question. Have a standing deprioritized list with written reasons. When new requests come in, map them to what they'd displace. Make the tradeoffs explicit: 'If we add this, we drop that. Which do you want?' Ownstakeholder relationships separately from roadmap reviews. Surprises kill trust.

What the Lead/EM Interview Probes

Behavioral question: 'Tell me about a time you had to push back on a stakeholder request.' The interviewers are not looking for a story where you heroically said no. They want to see the reasoning process — how you assessed the request, what alternatives you offered, how you maintained the relationship.