Case Study — Grubhub

Merchant AI Agent: Reimagining B2B Restaurant Operations

Executive Summary

Facing low merchant satisfaction and steady churn, we knew we had to make managing a restaurant on Grubhub significantly easier. To solve this, we built the Merchant AI Agent right inside Grubhub For Restaurants (GFR) — a conversational assistant designed to take the friction out of daily operations, cut down on support calls, and help our restaurant partners thrive.

Project Overview

Running a restaurant is hard enough on its own — juggling the kitchen, staff, and customers while trying to keep the business healthy. Grubhub For Restaurants (GFR) added to that load by asking owners and operators to manage routine tasks that pulled their attention away from the floor. With partner satisfaction scores trailing our goals and a desire to reduce churn, restaurant owners and operators needed an efficient, low-friction way to manage their daily workflows.

I championed this initiative — leading the teams in product design strategy and overseeing the execution of the Merchant AI Agent — an LLM-powered assistant integrated directly into GFR. The initiative aimed to transition the platform from static task execution to proactive, conversational business management, reducing support care contacts and improving overall merchant satisfaction.

Generative User Research & Discovery

To uncover the root causes of merchant friction and validate our hypothesis before writing a single line of code, we conducted deep generative user research.

On-Site Merchant Visits: Our design and product team spent extensive time embedded in local restaurant kitchens and back offices during peak operational hours. Observing merchants in their natural environment revealed the severe physical and mental toll of high-stress rushes.
Uncovering Cumbersome Workflows: Through contextual inquiry and merchant roundtables, we discovered that operators rarely had the bandwidth to navigate multi-layered web dashboards. Simple routines — like updating prep times, pausing orders when overwhelmed, issuing customer credits for missing items, or checking end-of-day financial data — required too many taps and clicks, pulling attention away from food preparation and hospitality.
Translating Insights into Product Vision: Merchants expressed a strong desire for an intuitive assistant that could "just handle things" through simple conversation. This qualitative validation directly shaped our product strategy: building an agent that mimics a reliable floor manager rather than a complex enterprise software tool.

The Problem & Opportunity

Restaurant operators face intense operational pressure during peak hours, leaving little patience for navigating complex B2B dashboards.

The Pain Points: Merchants faced high friction when executing basic tasks — such as updating store hours, pausing orders during rushes, parsing financial summaries, and adjusting menu item availability. Low baseline engagement with GFR compounded these challenges, leading to high support ticket volumes.
The Strategic Shift: Competitors were rapidly deploying conversational AI tools. To retain merchants and improve partner health, we needed a solution that allowed operators to bypass complex UI paths entirely via natural language prompts.

Core Hypothesis

If we introduce a conversational AI agent into GFR that handles high-frequency operational tasks, merchants will complete workflows faster, reducing support care contacts and driving an increase in merchant NPS and platform engagement.

Design Principles & Strategy

To shape the user experience, our design team anchored the product development around three core pillars.

soap

Cleanliness

B2B tools are inherently complex. We designed a polished, distraction-free interface that minimizes cognitive load during high-stress service hours.

emoji_objects

Intuition

The agent was built as a true efficiency tool, guiding users proactively to their next best actions with seamless conversational flows.

sentiment_very_satisfied

Joy

We infused subtle micro-interactions, thoughtful color accents, and responsive conversational UI states to elevate a traditionally sterile B2B experience.

Scope & Prioritized Requirements

Working closely with product and engineering partners, we structured our rollout into clear phased milestones based on merchant frequency and operational impact.

P0 — Must Have (Core Operational Workflows)

Open and Close Store: Rapidly toggle store status or set timed pauses during unexpected rushes.
Update Store Hours: Modify operating schedules dynamically through conversational prompts.
Update Settings: Instantly adjust baseline configurations like prep times without digging through nested submenus.
Search Help Articles: Query knowledge base documentation on-demand without leaving the active dashboard.
Read and Respond to Reviews: Monitor diner feedback and issue appropriate replies directly through the bot interface.
Provide Credit for Order Issues: Streamline guest recovery by authorizing order credits instantly via conversational commands.

P1 — Next Up (Advanced Efficiency & Growth)

Mark Items Out of Stock: Quickly toggle inventory availability when specific ingredients run low.
Export Financial Documentation/Reports: Query and pull revenue data and financial summaries effortlessly.
Filter Through Orders: Slice and dice active order streams using custom criteria.
Actionable Next Steps: Guide users intelligently when a prompt exceeds the agent's direct execution capabilities.
Create Promotional Campaigns: Set up marketing goals and discounts through guided dialogue to hit target order volumes.

Solution & Key Workflows

The Merchant AI Agent intercepts complex UI navigation by offering a persistent, context-aware conversational overlay within GFR.

Timed Store Pause & Settings: Instead of navigating through multiple settings submenus, a merchant simply prompts: "Put my store in busy mode for 2 hours." The agent prompts for confirmation, calculates the timestamp, and executes the state change in the background.
Proactive Decision Support: Beyond executing commands, the agent surfaces actionable insights based on common merchant requests — such as issuing customer credits, managing item replacements, or handling diner reviews.
Bulk Editing & Timed Promotions: The agent also makes bulk editing menu items and setting up timed promotions effortless — turning what used to be a tedious, multi-step process into a single conversational request.

Marking Items Out of Stock

Marking an item out of stock with the AI agent

Running a Promotion

On Desktop — Pausing the Store

Operator pausing their store from the Grubhub for Restaurants desktop dashboard

Measurement & Impact

Success is being measured across a balanced scorecard of engagement, efficiency, and sentiment metrics.

North Star Metric

Improvement in merchant NPS is being tracked among active AI Agent users.

Operational Efficiency

We're seeing an estimated 8–10% reduction in support care contacts and call volume.

Product Engagement

We're tracking session frequency (averaging ~10 sessions per merchant per week) and capturing prompt logs to continuously refine our feature roadmap and priority backlog.