From Gut Feel to Optimal: How Hop Labs Transformed Menu Planning at Gobble
Gobble
SUMMARY
Gobble selects 14 meals each week from a catalog of ~1,000 recipes. This menu generation decision directly drives how many subscribers skip their delivery or not.
Hop Labs built an AI-powered, chef-in-the-loop optimization tool that replaced a fragile, single-expert process with one that any team member can run, and that consistently produces better menus than manual planning did.
Outcomes: an 11% improvement in customer satisfaction and a 18% drop in meal costs.
THE COMPANY
Gobble is a meal kit delivery service built around convenience and culinary delight. Each week, subscribers choose meals from a curated menu, making the quality and composition of that menu a direct driver of revenue. A miss on the menu means more customers skip their delivery.
THE CHALLENGE
Every week, Gobble's culinary team faced a deceptively complex puzzle: choose 14 meals from a recipe catalog of roughly 1,000 options. But not just any 14 – the menu had to span at least six cuisine types, balance protein variety (poultry, red meat, seafood, and plant-based protein), meet dietary requirements, and do all of this while keeping costs manageable and prep time realistic for the kitchen team.
The stakes were tangible. Gobble tracked weekly "skip rates” – the percentage of subscribers who opted out of a delivery because the menu didn't appeal to them. A poorly constructed menu translated directly to lost revenue.
For years, the menu generation process lived almost entirely in the head of one highly experienced chef. It required deep institutional knowledge – understanding which meals cannibalize demand from each other, how combinations of dishes resonate with the broader customer base, and how to honor moments like holidays without derailing the rest of the week's offering. It worked – but it was fragile, slow, and impossible to scale.
The core question Gobble brought to Hop Labs: Can we do this smarter, faster, and without betting everything on one person's intuition?
THE APPROACH
Hop Labs approached the problem as a multi-objective optimization challenge. This is a type of vector optimization that can be helpful when optimal outcomes need to be achieved in the presence of trade-offs between two or more objectives that conflict at times.
Understanding the problem space
We began by separating two fundamentally different types of requirements – constraints and optimization targets.
Constraints are the non-negotiables – rules that every valid menu must satisfy. For Gobble, these included cuisine diversity, protein variety, and dietary requirements (e.g., minimum vegetarian and gluten-free options). These reflect how Gobble's product must work, and breaking these constraints isn't an option.
Optimization targets are where trade-offs live. Gobble had three – minimize meal cost, minimize culinary prep time, and maximize predicted customer satisfaction.
Making the math tractable
With ~1,000 recipes and a need to select 14, the number of possible menus is astronomically large – far too many to evaluate one by one. Our team solved this through careful problem structuring; by precisely calibrating the constraint set, we reduced the universe of valid menus to a population that could be fully computed in a reasonable time. That population is calculated once and cached, so that both the initial optimization and any subsequent chef-driven adjustments happen in real time.
“There’s a level of expertise and authority that you guys carry, and you used it to create something for our business that I couldn’t do on my own. If I was going to try to build my own team to do it, I wouldn’t be able to get good enough people for the price.”
Keeping the chef in the loop
This tool was not designed to replace the chef – it was designed to empower the chef to make more informed decisions. After generating an optimized menu, the system tells the chef exactly which swaps are available for any given meal and how each swap affects cost, prep time, and projected customer satisfaction. If the chef wants to lock in a specific dish – say, a ham-based meal for the holiday season – they can fix that choice and re-run the optimization around it. The AI fills in the rest intelligently.
The result is a system that encodes the analytical complexity of the problem while preserving the chef's judgment and domain expertise. Anyone on the team can now run the tool and produce a high-quality menu, not just the one specialist who held all the knowledge before.
THE RESULTS
“The work your team did gave us the leg up against our competition — our small and agile, nimble team was able to punch much higher than our weight class. ”
The impact of the tool Hop built for Gobble was measurable and significant:
11% average improvement in customer satisfaction — Fewer skips, more engaged subscribers
18% reduction in meal costs — Optimized selection translated directly to margin improvement
Real-time responsiveness — The chef can tweak, override, and re-optimize interactively, turning a multi-hour manual process into a fluid, iterative workflow.
Resilience and scalability — As more process knowledge is embedded in the tool, the level of expertise required to plan a menu decreases. Now, the chef and optimization tool together design better menus than the chef did alone. Continued refinement of the optimization process leads us to a place where any member of the team can design an effective menu.
Where does your org need to optimize and bring more resilience to your processes? Contact us to learn how Hop can help.