The Situation
CookUnity was a chef marketplace for ready-to-eat subscription meals. Its advantage: the largest variety of dishes from local chefs. Variety was believed to drive retention — and early markets like New York, with the deepest menus, also had the best retention.
But as CookUnity scaled to LA, Austin, Atlanta, Chicago, Seattle, and Miami, the relationship between variety and retention wasn’t as clear. LA, with fewer menu items, sometimes matched or beat NY in churn. Variety carried high operational costs — so how much was enough?
The Problem
Retention drivers weren’t obvious: was it menu size, new cuisines, repeat meals, referrals?
The team tried many tactics, but lacked a way to see which habits truly mattered.
Catalog variety was costly, yet presentation and in-app experience made the difference between delight and overwhelm.
The Shift
We built a Next Best Action ML model:
Baseline Predictive LTV: First, accurately predict each customer’s expected revenue trajectory.
Intervention Layer: Overlay tactics — try a new chef, repeat a loved meal, explore a new cuisine, refer a friend — to forecast incremental revenue.
Training the Model: Business teams partnered on campaigns (e.g. “Try a new chef” emails) to generate clean data. Over time, the model became highly accurate.
The Insights & Actions
For New Customers: The most powerful driver was finding a meal they loved — endless variety wasn’t the answer.
For Mature Customers: Trying new cuisines boosted retention for adventurous personas. For “healthy-ish” or restricted-diet personas, curating the menu to their needs (e.g. vegetarians) mattered more than breadth.
For the Business: Menu variety is valuable, but only if paired with personalized presentation and habit formation.
The Outcome
+10% Net ARPU vs. business-as-usual tactics in initial tests.
Rolled out nationally, the uplift held: +10% net ARPU at scale.
CookUnity grew to $400M ARR, with retention outperforming competitors.
