An AI model can meet all technical criteria (correctness, relevance) yet produce outputs that are tonally inappropriate or off-brand. Ex-Alexa PM Polly Allen shared how a factually correct answer about COVID was insensitive, proving product leaders must inject human judgment into AI evaluation.

Don't ask an LLM to perform initial error analysis; it lacks the product context to spot subtle failures. Instead, have a human expert write detailed, freeform notes ("open codes"). Then, leverage an LLM's strength in synthesis to automatically categorize those hundreds of human-written notes into actionable failure themes ("axial codes").

Product managers may lack the expertise to create comprehensive evals from scratch. A better approach is to generate initial outputs with a base model, have subject matter experts review them, and use their direct feedback to define what constitutes a failure. It's easier for experts to spot mistakes than to predict them.

The common mistake in building AI evals is jumping straight to writing automated tests. The correct first step is a manual process called "error analysis" or "open coding," where a product expert reviews real user interaction logs to understand what's actually going wrong. This grounds your entire evaluation process in reality.

Do not blindly trust an LLM's evaluation scores. The biggest mistake is showing stakeholders metrics that don't match their perception of product quality. To build trust, first hand-label a sample of data with binary outcomes (good/bad), then compare the LLM judge's scores against these human labels to ensure agreement before deploying the eval.

AI is great at identifying broad topics like "integration issues" from user feedback. However, true product insights come from specific, nuanced details that are often averaged away by LLMs. Human review is still required to spot truly actionable opportunities.

AI can generate rule-based "top-down" evaluations from a task description (e.g., word count). However, discovering nuanced "bottom-up" evaluations requires human intuition from reviewing many real-world data samples to find subtle, recurring failure modes.

A one-size-fits-all evaluation method is inefficient. Use simple code for deterministic checks like word count. Leverage an LLM-as-a-judge for subjective qualities like tone. Reserve costly human evaluation for ambiguous cases flagged by the LLM or for validating new features.

The most valuable evals aren't built with complex software but are often simple spreadsheets. Their power comes from deep subject matter expertise, which is necessary to create nuanced prompts and accurate scoring criteria that truly test a model's ability in a specific domain like clinical genomics or law.