Develop Data-Driven Features with Integrated Analytics

This skill orchestrates a comprehensive data-driven feature development process, from initial data analysis and hypothesis formulation through implementation, A/B testing infrastructure, and post-launch analysis, using specialized subagents at each step and emphasizing modern product analytics practices and statistical rigor.

Phase 1 (Data Analysis and Hypothesis Formation) runs exploratory data analysis (a data-scientist agent analyzing user behavior, segments, and baseline metrics via Amplitude/Mixpanel/Segment), business hypothesis development (a business-analyst agent defining success metrics and expected ROI using ICE or RICE prioritization), and statistical experiment design (sample size and power analysis, control/treatment groups, randomization strategy, and consideration of Bayesian A/B testing for faster decisions).

Phase 2 (Feature Architecture and Analytics Design) covers feature architecture planning (a backend-architect agent designing feature-flag integration via LaunchDarkly/Split.io/Optimizely, gradual rollout, and circuit breakers), analytics instrumentation design (event schemas, segmentation properties, funnel tracking via modern SDKs), and data pipeline architecture (real-time streaming via Kafka/Kinesis, batch processing, warehouse integration via Snowflake/BigQuery, and GDPR-compliant governance).

Phase 3 (Implementation with Instrumentation) covers backend implementation (feature flag checks at decision points, comprehensive event tracking, error monitoring), frontend implementation (event tracking for interactions, Core Web Vitals, consistent control/treatment experience), and optional ML model integration (low-latency online inference, A/B testing between model versions, drift monitoring).

Phase 4 (Pre-Launch Validation) validates analytics implementation end-to-end in staging and configures experiment infrastructure (feature flag targeting rules, initial 5-10% traffic allocation, kill switches, randomization testing). Phase 5 (Launch and Experimentation) executes a gradual rollout (internal dogfooding, then 1-5% beta, then increasing to target traffic with automated rollback on anomalies) and sets up real-time monitoring dashboards, statistical-significance alerts, and guardrail-metric tracking via tools like Datadog or New Relic.

Phase 6 (Analysis and Decision Making) runs statistical analysis (significance testing with confidence intervals, segment-level effects, both frequentist and Bayesian approaches, multiple-testing correction), business impact assessment (actual vs expected ROI, cost-benefit including operational overhead, a rollout/iterate/rollback recommendation), and post-launch optimization (analyzing treatment-group behavior for friction points and planning follow-up experiments via cohort analysis).

Its configuration options cover experiment parameters (minimum sample size, confidence level, runtime days, gradual/fixed/adaptive traffic allocation), analytics platforms (Amplitude, Segment, Mixpanel), feature flag providers (LaunchDarkly, Split, Optimizely, Unleash), statistical methods (frequentist, Bayesian), and monitoring toggles (real-time metrics, anomaly detection, automatic rollback). Success criteria include 100% event-schema tracking coverage, valid randomization with sufficient statistical power and no sample ratio mismatch, rigorous significance testing, measurable business impact without degrading guardrail metrics, no p95 latency degradation with error rates below 0.1%, a clear go/no-go decision within the planned runtime, and documented learnings feeding the next development cycle.

experiment_config:
  min_sample_size: 10000
  confidence_level: 0.95
  runtime_days: 14
  traffic_allocation: "gradual"

analytics_platforms:
  - amplitude
  - segment
  - mixpanel

feature_flags:
  provider: "launchdarkly"

statistical_methods:
  - frequentist
  - bayesian

monitoring:
  - real_time_metrics: true
  - anomaly_detection: true
  - automatic_rollback: true

Use this skill for data-driven feature development tasks or workflows needing guidance, best practices, or checklists - building a feature guided by data analysis, instrumented with analytics, validated through A/B testing, and rolled out gradually with statistical rigor.

Not for tasks unrelated to data-driven feature development, or where a different domain or tool is needed.

Inputs: a feature to develop, existing user behavior data, and target business metrics.

Outputs: an EDA report and business hypothesis, a statistical experiment design, instrumented backend/frontend implementation with feature flags, a validated and gradually rolled-out experiment, and a statistical/business impact analysis with a go/no-go recommendation and follow-up experiment plans.

Amplitude, Segment, Mixpanel, LaunchDarkly, Split.io, Optimizely, Unleash, Kafka, Kinesis, Snowflake, BigQuery, Datadog, New Relic; specialized subagents including data-scientist, business-analyst, backend-architect, data-engineer, frontend-developer, ml-engineer, deployment-engineer, and observability-engineer.

Product and engineering teams building features guided by data insights, instrumented analytics, and rigorous A/B testing from hypothesis through post-launch analysis.