Stand outside any busy sweet shop before Diwali. Plenty of people walk in, fewer reach the counter, and a surprising number leave at the billing queue with nothing in hand. The shopkeeper who only counts footfall will hire a better signboard painter. The one who watches the queue will open a second billing counter. The Zomato assignment was a lesson in becoming the second shopkeeper.
The brief: as a PM for Zomato food delivery in Tier-1 cities, improve engagement and retention across Discovery, Order and Delivery. I had no company data, so I built a declared-synthetic model and anchored it to public benchmarks. Every number below from my model is an assumption, not a Zomato figure.
Part 1: Seven stages, one priority
I mapped seven stages: app open, restaurant discovery, menu view, cart, checkout, order placed, delivered. In my model, 100,000 weekly opens ended in 17.6% delivered orders. Each stage got one drop-off hypothesis, because a drop-off without a hypothesis is only a complaint.
| Priority | Stage | Why here |
|---|---|---|
| 1 | Cart to checkout, checkout to order placed | The customer has already chosen the food. Fees revealed late and payment failures lose a sale that was nearly won. |
| 2 | Delivered and back to next open | A late or wrong order damages the next visit more than the current one. |
| 3 | Menu to cart | Weak photos, unclear portions and too much choice. |
The ranking follows intent. A user at checkout is the customer at the billing counter. Public data supports the focus: Razorpay has reported food and beverage cart abandonment of about 63.6%. My target was to move delivered orders from 17.6% to 26.4% of opens, mostly by fixing the last two steps before payment.
Part 2: Not all customers are the same customer
I compared three pairs of cohorts on month-1 retention in the model.
| Cohort pair | Weaker side | Stronger side |
|---|---|---|
| New vs repeat | New: 39% | Repeat: 82% |
| Low vs high frequency | Low: 47% | High: 91% |
| Discount-driven vs full-price | Discount: 52% | Full-price: 74% |
The discount cohort is the one that changed my thinking. With an assumed acquisition cost of ₹320 and a contribution of ₹36.8 per order, a customer must place about 8.7 orders before the company recovers its money. A customer who comes for the coupon and leaves when it ends never reaches that mark. Discounts buy first orders. Habit has to be earned some other way.
A coupon can bring a customer to the door. Only a kept promise brings them back.
Part 3: One North Star, many ladders
I chose on-promise delivered orders per monthly transacting customer, with a target of 3.1 to 3.6. "On-promise" is the important word. It counts only orders that arrived when the app said they would, so growth teams cannot move the number by pushing orders the kitchen and rider network cannot honour.
Under it sit thirteen solutions, seven for funnel stages and six for cohorts. Each has its own metric ladder: an input metric the team controls, an output metric the business sees, a leading indicator that moves in days, a lagging one that moves in weeks, and a guardrail that must not get worse.
What I took away
Funnels tell you where, cohorts tell you who, and the North Star tells you whether fixing both actually mattered. Doing only one of the three is like checking only the speedometer on a long drive. You also need the fuel gauge and the map.
Sources for benchmarks: Medianama (Eternal quarterly results), Business Standard, Razorpay and Growth-onomics. All funnel, cohort and unit-economics figures are synthetic model inputs.