Teams building LLM features use everything from informal 'vibe checks' to formal governance, but many cannot turn evaluation findings into concrete fixes — a persistent 'results-actionability gap' that slows LLM value capture; firms that invest in instrumentation and clear remediation pathways reap larger productivity gains.

Main Finding

Product teams evaluating LLM-powered features rely on a spectrum of practices—from informal “vibe checks” to organizational meta-work—to cope with LLMs’ unpredictability. The study confirms several known evaluation challenges and introduces a novel “results-actionability gap”: teams collect evaluation data but cannot translate findings into concrete, implementable improvements. Successful teams close this gap by systematizing interpretive practices and creating clearer pathways from evaluation signals to product changes.

Key Points

• Study scope: interviews with 19 practitioners across diverse sectors who build or evaluate LLM-powered products.
• Evaluation practices: the authors identify ten practices that teams use, ranging from lightweight, interpretive checks (e.g., “vibe checks”) to formal organizational processes (governance, KPIs, meta-evaluation). Examples include qualitative user reviews, red-team testing, A/B experiments, telemetry and log analysis, structured annotation, and organizational coordination work.
• Confirmed challenges: the paper validates several previously documented problems when evaluating LLM systems (e.g., model unpredictability, metric mismatch, high human-evaluation costs, and difficulty reproducing failures).
• New challenge — results-actionability gap: teams often produce evaluation outputs (tests, metrics, user feedback) but lack the mechanisms, processes, or technical levers to convert those outputs into actionable engineering or product changes.
• Interpretive practices are not merely sloppy work: the authors argue that ad-hoc, human-centered evaluation (interpretive checks, team sense-making) are rational adaptations to LLM behavior rather than methodological failures.
• Practices-to-formalization trajectory: the paper describes how effective teams evolve from ad-hoc interpretive methods toward systematic evaluation by (a) formalizing prompts/tests, (b) instrumenting outputs, (c) mapping failure modes to remediation paths, and (d) creating organizational decision rules.

Data & Methods

• Method: qualitative interview study.
• Sample: 19 practitioners across multiple industries and organizational roles involved in building or evaluating LLM-enabled products.
• Analysis: thematic coding of interview transcripts to identify evaluation practices, challenges, and patterns among teams that successfully move from informal to systematic evaluation.
• Outcomes: taxonomy of ten evaluation practices, articulation of the results-actionability gap, and recommended strategies drawn from observed successful teams.

Implications for AI Economics

• Measurement friction and value realization: the results-actionability gap creates a hidden cost — teams can detect problems but cannot cheaply translate findings into improvements, reducing the speed and ROI of LLM investments. This raises the effective cost of deploying LLM features and may slow monetization or product-market adaptation.
• Resource allocation and organizational structure: firms that invest in instrumentation, cross-functional processes, and remediation levers capture more value from LLMs. The findings suggest a comparative-advantage story: organizations with better evaluation-to-action pipelines will obtain higher productivity gains and market edge.
• Labor and task reorganization: the persistence of interpretive, human-in-the-loop evaluation implies ongoing labor requirements (annotation, sense-making, governance roles). This affects forecasts of automation and labor substitution in sectors adopting LLMs.
• Market for tools & services: there is demand for tooling that bridges evaluation outputs to actionable fixes (failure-mode libraries, standardized remediation templates, evaluation-to-priority mapping). This signals economic opportunity for third-party platforms and consulting services.
• Policy and compliance: regulators and standard-setters valuing transparency and auditability will need to account for the gap between evaluation results and actionable fixes; firms may require incentives or rules to ensure evaluation leads to remediation, not just documentation.
• Research priorities for economists: quantify the productivity gains from closing the actionability gap; estimate firm-level heterogeneity in evaluation capability and its effect on adoption; model investment trade-offs between building evaluation-to-action pipelines versus accepting reduced LLM performance.

If you’d like, I can: - extract the ten evaluation practices into a concise list with brief descriptions (based on the paper’s taxonomy), or - draft a short research agenda for economists studying the results-actionability gap (models, empirical approaches, data sources).