Efficiency analytics

Efficiency analytics show how well people use AI, not just whether they use it. PromptGlass looks at prompt structure, conversation patterns and outcomes. It identifies teams that get value quickly, teams that struggle, and the use cases that deserve templates, training or a better tool.

Signals

Signal	What it measures	Why it matters
Prompt quality score (0–100)	Context, a clear task, constraints, an output format, examples. Scored by a rubric model.	Low scores track with re-prompting and abandoned sessions
Re-prompt rate	Share of conversations where the user rephrases the same request within a few turns	High rates suggest the tool or the prompt isn’t working
Turns to resolution	Median turns before a conversation ends or the output is copied	Fewer turns usually means a better fit
Copy-out rate	Share of responses copied or exported (extension)	A proxy for usefulness
Model fit	Tasks sent to a large model that a smaller model handles equally well	Feeds cost optimization
Use-case mix	Distribution of clustered topics	Shows where AI is creating value

Time-saved estimates

PromptGlass can estimate time saved per use case. The model is transparent and you configure it. For each topic you set a baseline (minutes the task takes without AI) and an assist factor. The estimate is reported with its inputs, never as a measured fact.

timeSaved:
  topics:
    email-drafting:        { baselineMinutes: 8,  assistFactor: 0.5 }
    code-review:           { baselineMinutes: 20, assistFactor: 0.3 }
    contract-summarization:{ baselineMinutes: 45, assistFactor: 0.6 }
  excludeAbandoned: true   # don't count sessions with no copy-out and many re-prompts

Calibrate the factors with a short time-and-motion study or a user survey, then lock them so trends stay comparable.

Turning insight into action

• Prompt templates: publish templates for the top use cases. The extension can suggest them inline when it detects a matching topic.
• Targeted enablement: send teams with low quality scores and high re-prompt rates to role-specific training. This also produces evidence for AI literacy programs.
• Tool fit: when one team’s re-prompt rate is high in tool A but low in tool B for the same topic, standardize on tool B.

Privacy

Quality scoring needs prompt content at evaluation time. With metadata-only storage, PromptGlass scores the prompt inline, stores only the score and discards the text. Individual scores are visible only to roles with analytics.individual. Managers see team aggregates.