1. Courses
  2. Metrics & Product Analytics
  3. Capstone

Capstone project for Metrics & Product Analytics

Prove the 30-minute promise

The 30-minute "Now" delivery is Mangrove Market's headline promise, and the managing director wants one clear view of whether it is being kept and what to do next. Build the measurement system for it and use it. Start with a metric tree from weekly on-time orders down to inputs teams can move, including picking time, rider assignment time and stock accuracy, with owners, sensitivity and guardrails. Write the tracking plan for the events from checkout to the doorstep, deciding which come from the apps and which from the server, and how offline Rider app phones are handled. Use the sample cohort table in the retention simulator (or your own) to find which customers stop ordering and when. Design an experiment for "suggest a substitute before packing", where a picker offers the customer a one-tap swap when an item is out of stock, taking care that pickers and riders are shared between groups. Finally, open the diagnosis simulator, pick one of its weeks where on-time fell from 91% to 79% in one city, and write the one-page diagnosis, deciding between rainy evenings, a new app version and a data bug, ending with a recommendation.

What you hand in

  1. A metric tree for weekly on-time orders with today's numbers, owners, the +1% sensitivity of each input, causal links marked as hypotheses, and three to five guardrails with limits
  2. A tracking plan from checkout to doorstep: 8 to 15 events with exact triggers, properties, app or server source, a source of truth for each number, a settling window and daily data checks
  3. A one-page retention readout: definition, the typical curve and its floor, the cohort that differs and why, and one decision
  4. An experiment plan for substitutes before packing: unit and why, primary metric, guardrails, minimum detectable effect, sample size and duration, and the decision rule
  5. A one-page diagnosis of the 12-point on-time drop in the diagnosis order, ending with a decision, an owner, a date and a follow-up

Check your work

Tick each item when your project meets it. Be strict: this is the rubric the reviewer uses.

  • The north star counts value delivered (orders on time, not orders or sign-ups), and the tree multiplies back to it with consistent numbers
  • Arithmetic and causal links are distinguished, and every input has one owner, a source and a counter-metric where it can be gamed
  • The tracking plan takes counts from the order table, uses two timestamps and a device-made event id, and guards the 'delivered' event that the on-time metric depends on
  • The retention readout states the active event, period and denominator, and explains the odd cohort before averaging it in
  • The experiment chooses a unit that avoids interference (store-day or zone-slot, or customers with picker-time guardrails) and fixes the MDE, size and decision rule before starting
  • The diagnosis checks the data first, including an app-version split and a second source, separates mix from rate, and names the slice that splits cleanly
  • Every recommendation has an owner, a date, the metric it should move and when it will be checked

0 of 7 checked

Model answer

The model answer opens once your capstone passes the review, so you can compare your work with an outline of a strong submission.

Hand in your capstone

Your links are checked automatically, then the work is scored against the rubric above, item by item, with comments. A pass counts toward the course certificate and shows on your profile.

Loading…