// case study · 2026

MetricLens — AI Product Analytics Reporting Agent

A cloud-first workflow that turns structured product data into KPI analysis, funnel diagnostics, segment insights, AI-assisted hypotheses, and automated decision-ready reports.

The problem

Product information can be spread across events, user attributes, feature usage, and historical metrics. Turning it into a repeatable product review can require manual analysis and interpretation.

Question: Can a lightweight AI workflow transform structured product analytics into a repeatable, decision-oriented report while keeping the underlying metrics deterministic?

Workflow

Google Sheets
Users · Events · Features · History
→
Apps Script
Aggregation + analytics
→
Gemini
Interpretation
KPI Engine
Metrics + funnel
→
Google Docs
Structured report
→
PDF + Drive
Archive

Analytics engine

Activation & KPI analysis

Deterministic period comparison before Gemini interprets the data.

Complete funnel

Six steps from signup to core action, with user counts, conversion rates, and step drop-off.

Segments

Activation by platform, acquisition channel, and plan.

Feature + historical context

Feature adoption and historical activation/retention provide additional product context.

01 · Signup
02 · Onboarding Started
03 · Step 1 Completed
04 · Step 2 Completed
05 · Step 3 Completed
06 · Core Action Completed

AI layer

Gemini is not the source of truth for the KPI calculations. MetricLens computes numerical evidence first, then uses Gemini to interpret that evidence.

Evidence

Calculated KPIs, funnel movement, segments, feature adoption, and historical metrics.

Hypotheses

Potential explanations are clearly separated from observed evidence.

Experiment

The report translates product friction into a measurable experiment recommendation.

Guardrails

Primary and guardrail metrics keep recommendations measurable.

Automated report

OutputPurpose
Executive summaryFast review of the main product signals
KPI + funnel tablesInspect numerical evidence
Segment + feature analysisLocate meaningful differences
AI analysisInterpret evidence and frame hypotheses
Experiment + guardrailsTranslate analysis into a testable next step

Limitations

The included portfolio dataset is synthetic. Pre/post comparisons are descriptive and do not establish causality. Gemini model availability is account/API dependent.

Stack

Google Apps Script · Google Sheets · Gemini API · Google Docs · Google Drive · Product Analytics · Funnel Analysis · Workflow Automation