Swiggy has developed an in-house predicted lifetime value (pLTV) model to estimate the long-term value of new customers across its food delivery and Instamart quick commerce businesses. The model is designed to generate a useful signal before a customer's first order, allowing Swiggy to use predicted value in advertising bid optimization rather than relying on short-term conversion metrics.
Swiggy pLTV model architecture (Source: Swiggy Blog Post)
The prediction problem is complicated by sparse early signals and a highly skewed target distribution. Swiggy notes that a large share of new users place no orders during their first 30 days, while a relatively small group generates substantially higher lifetime value. The model therefore needs to distinguish between users who initially appear similar but develop different ordering patterns over time.
Swiggy's approach uses more than 350 features grouped across areas including acquisition channel, device and fraud signals, geographic behavior, complaint history, Instamart category affinity, order behavior, payment patterns and socioeconomic indicators. The features are generated from information available before a customer's first order, addressing the cold start requirement for advertising systems.
Soumyajyoti Banerjee, Staff Data Scientist at Swiggy, highlighted the trade-off in a LinkedIn post,
The twist: adding a second prediction task shrank the model by 63% — and made it more accurate.
The model uses a relatively simple multilayer perceptron rather than a more complex architecture. Three shared hidden layers learn representations across the prediction tasks, followed by task-specific heads for Food and Instamart. Swiggy also introduced order count as an auxiliary prediction task alongside lifetime value. The resulting four-head model produced marginally better results while reducing the parameter count from 363,000 to 135,000, a 63% reduction.
According to Jayshmi A, an engineer at Swiggy, conventional regression metrics such as mean absolute error and mean absolute percentage error were not sufficient to evaluate the model because of the large number of zero-value users and the long tail of customer value. Instead, Swiggy evaluated whether the model could correctly rank customers by value. The evaluation divides predicted and actual values into ten quantiles and measures how often users fall into the corresponding predicted and actual groups.
The problem wasn’t producing a score; it was producing a score that could survive sparse early signals.
The model achieved a Spearman correlation above 0.75 for both Food and Instamart, with 70% to 80% diagonal coverage across the value deciles.
pLTV Decile Coverage or Prediction Heatmap (Source: Swiggy Blog Post)
In production, Swiggy uses the pLTV signal with Google's target return on ad spend (tROAS) bidding. The intended workflow is to rank customers by predicted value and allow the advertising system to differentiate bids based on that ranking.
Swiggy also compared the in-house model with a third-party pLTV platform in production A/B experiments. The company's model delivered higher retention and gross order value per acquired user without additional platform costs, while the external platform acquired more users per dollar spent.
Swiggy said the next stage of the work will move from point estimates to probabilistic pLTV predictions, aiming to represent uncertainty in customer value and incorporate it into bidding decisions.