application-analytics

name	application-analytics
description	Analyze job application outcomes and identify what's working. Use this skill whenever the user asks "how's my search going?", "what's working?", "show me patterns", "why am I not getting interviews?", "which sources are converting?", "give me analytics", "pipeline health", "conversion rates", or any question about the effectiveness of their job search strategy. Also trigger when the user asks to compare application outcomes, wants to understand which resume tailoring approach works best, or needs data to decide where to focus next. This is the intelligence layer — it turns raw application metadata into actionable insights. Do NOT trigger for individual status updates (job-tracker) or strategic recommendations (search-optimizer). This skill provides the data; search-optimizer interprets it.
recommended_model	{"default":"sonnet","reasoning":"Pattern matching across structured data, calculating conversion rates, and summarizing trends. No creative writing needed — just analytical clarity.\n","upgrade_to_opus_when":"The user asks for deep strategic interpretation of the data (e.g., \"what does this mean for my career?\"). In that case, suggest invoking search-optimizer instead.\n"}

Application Analytics — What's Working?

Intent

1. Minimum data threshold before claiming patterns — with fewer than 5 applications in a category, say "early signal" not "trend"; extrapolating from 2 data points is worse than having no data
2. Honesty about confidence levels — flag when sample sizes are too small, when time periods are uneven, or when a pattern might be coincidence
3. Urgency hierarchy in reporting — overdue follow-ups and stalled applications surface before conversion rates and trend analysis; actionable now beats interesting later
4. Make the invisible visible — without analytics, job searching is guesswork; this skill reveals which sources, match levels, and strategies actually lead to interviews
5. Data-driven strategy over gut feeling — every recommendation traces back to a number, not an opinion
6. Feed the learning loop — analytics reports are consumed by search-optimizer; reliable data in means reliable recommendations out
7. Full pipeline scan completes instantaneously; confidence flagging mandatory on every metric — speed enables frequent check-ins, but every number must carry its confidence context

Why this skill exists

Without analytics, job searching is guesswork. You don't know if LinkedIn or referrals convert better for your profile. You don't know if stretch matches ever work or if you should focus on strong matches. You don't know that your follow-ups are 5 days late on average. This skill makes the invisible visible.

Prerequisites

• job-applications/applications/ with at least 3-4 application folders containing metadata.json files
• Applications at various stages (evaluating, applied, interviewing, rejected) provide the most useful data. A pipeline with only "evaluating" applications has nothing to analyze.

Workflow

Step 1: Scan all applications

Read every metadata.json in job-applications/applications/*/. Build a dataset:

For each application, extract:
  - company, role
  - status (current stage in lifecycle)
  - match_score.overall (strong/good/stretch/long_shot)
  - source (LinkedIn, referral, direct, etc.)
  - applied_date, follow_up_date
  - tailoring_intensity (if set)
  - rejection_reason (if rejected)
  - interview_date (if interviewed)
  - learning_flags (if any)
  - former_employer flag

Step 2: Calculate pipeline metrics

Pipeline health:

• Total applications by status (evaluating, applied, interviewing, offered, rejected, withdrawn)
• Velocity: average days from evaluating → applied → interviewing
• Stall detection: applications sitting in "applied" for 3+ weeks without movement

Conversion rates (when enough data exists):

• Applied → Interview rate (overall)
• Applied → Interview rate by match score
• Applied → Interview rate by source
• Applied → Interview rate by tailoring intensity
• Interview → Offer rate

Source analysis:

• Applications per source
• Conversion rate per source
• Average time-to-response per source

Match score effectiveness:

• Distribution of applications by match score
• Outcomes by match score (which scores lead to interviews?)
• Gap patterns: which gaps appear in rejections vs successes?

Step 3: Identify patterns and at-risk items

At-risk applications:

• Follow-up dates that have passed with no status change
• Applications in "applied" for 14+ days with no movement
• Applications in "interviewing" with no follow-up date set

Rejection patterns:

• Common rejection reasons (if logged)
• learning_flags that repeat across applications
• Match score of rejected applications (are stretch matches always failing?)

Positive patterns:

• What do successful applications (interviewing+) have in common?
• Which keywords appear in applications that move forward?
• Which sources produce the most interviews?

Step 4: Present findings

Structure the report in order of actionability:

1. Pipeline snapshot — Where things stand right now (counts by status)
2. Urgent items — Overdue follow-ups, stalled applications
3. What's working — Sources, match scores, or approaches with highest conversion
4. What's not working — Patterns in rejections or non-responses
5. Recommendations — 2-3 specific actions based on the data

Keep the report concise. Lead with numbers, not opinions. If there isn't enough data for a reliable pattern (fewer than 5 applications in a category), say so — don't extrapolate from 2 data points.

Step 5: Save the report

Write the analytics report to job-applications/analytics-report.md (overwriting any previous report). This file is consumed by search-optimizer.

Minimum viable analytics

With only 3-4 applications, the skill can still provide:

• Pipeline snapshot
• Overdue follow-ups
• Time-in-stage for each application
• Basic match score distribution

As the dataset grows (10+ applications), conversion rates and source analysis become meaningful. At 20+, pattern detection is reliable.

Edge cases

• No applications yet — Explain that analytics need data to work. Suggest starting with job-intake.
• All applications in "evaluating" — Show pipeline snapshot, note that analytics will be more useful once applications are submitted.
• Only rejections — This is the most important case. Focus on rejection patterns and what to change.
• Single source — Can't compare sources. Note this and suggest diversifying.

Skill composition

Upstream	This skill	Downstream
job-tracker (provides data)	application-analytics (finds patterns)	search-optimizer (makes recommendations)

Read references/analytics-metrics.md for metric definitions and calculation details.