---
title: "Decoding Customer Value: pLTV Models for India’s Quick Commerce Delivery Wars — part 1"
description: "Swiggy’s pLTV effort addresses early-stage lifetime-value prediction for Food Delivery and Instamart, where sparse signals, zero-inflated outcomes, long tails, and trial-driven acquisition make conversion models unsuitable. It encodes 350+ pre-order signals across 12 groups in a simple multilayer perceptron with three shared hidden layers, task-specific heads, a custom MAE loss, and Adam optimization. A four-head variant added order-count predictions for each vertical, producing marginally better performance with 135K rather than 363K parameters. Evaluation emphasized ranking through Spearman correlation and decile diagonal coverage, reaching above 0.75 Spearman and 70–80% diagonal coverage. In production bidding experiments, the in-house model delivered better retention and higher gross order value per acquired user at no incremental platform cost, while a third-party platform acquired more users per dollar."
---

# Decoding Customer Value: pLTV Models for India’s Quick Commerce Delivery Wars — part 1

[Swiggy](https://yomu.fyi/company/swiggy) · Jayshmi A · Jul 17, 2026

**Type:** Problem & solution

## Summary

Swiggy’s pLTV effort addresses early-stage lifetime-value prediction for Food Delivery and Instamart, where sparse signals, zero-inflated outcomes, long tails, and trial-driven acquisition make conversion models unsuitable. It encodes 350+ pre-order signals across 12 groups in a simple multilayer perceptron with three shared hidden layers, task-specific heads, a custom MAE loss, and Adam optimization. A four-head variant added order-count predictions for each vertical, producing marginally better performance with 135K rather than 363K parameters. Evaluation emphasized ranking through Spearman correlation and decile diagonal coverage, reaching above 0.75 Spearman and 70–80% diagonal coverage. In production bidding experiments, the in-house model delivered better retention and higher gross order value per acquired user at no incremental platform cost, while a third-party platform acquired more users per dollar.

## Context

The model had to predict customer lifetime value shortly after registration and before a first order, when transaction history was unavailable and early signals were sparse. The target was difficult because many users placed zero orders, value distributions were long-tailed, Food Delivery and Instamart had different purchasing patterns, and trial-oriented users could resemble repeat buyers at signup.

## Approach / What changed

The work used 350+ pre-order features spanning 12 groups, including acquisition context, device and fraud signals, geo-hierarchy metrics, complaints, Instamart category affinity, order behavior, socioeconomic indicators, name and address matching, and cart drop-off. A simple multilayer perceptron used three shared hidden layers and task-specific heads. A four-head version added order-count prediction for both verticals, with custom MAE training and evaluation based on ranking metrics and decile coverage.

## Takeaways

- A four-head model combining LTV and order-count prediction achieved marginally better performance with 135K parameters, compared with 363K for the two-head LTV-only model.
- The evaluation used decile quantile matching, diagonal coverage, and Spearman correlation because MAE and MAPE can be misleading for zero-inflated, heavy-tailed LTV targets.
- In production Google tROAS bidding experiments, the in-house model produced better retention and higher gross order value per acquired user at zero incremental platform cost, while the third-party platform delivered more users per dollar.

**Tags:** [Machine Learning](https://yomu.fyi/topic/machine-learning), [MLOps](https://yomu.fyi/topic/mlops), [Model Evaluation](https://yomu.fyi/topic/model-evaluation)

- Source: [Swiggy](https://medium.com/swiggy-bytes/decoding-customer-value-pltv-models-for-indias-quick-commerce-delivery-wars-part-1-d01a3e6015ba)
- Source URL: https://medium.com/swiggy-bytes/decoding-customer-value-pltv-models-for-indias-quick-commerce-delivery-wars-part-1-d01a3e6015ba
- Ingested by Yomu: 2026-09-09T13:57:18.008Z

[Read original post](https://medium.com/swiggy-bytes/decoding-customer-value-pltv-models-for-indias-quick-commerce-delivery-wars-part-1-d01a3e6015ba)