How It's Made

Tooling to Find More Items, More Quickly

Instacart

Instacart

May 25, 2022

How Instacart’s Fulfillment Technology group is working to make picking all of the items on your list as painless as possible

Authors: Benjamin S. Knight & Jason Cromer

Introduction

Is that wedge of Wisconsin sharp cheddar you’re looking for sitting in the Dairy section, or the Deli section? 🤔 What about that bag of flour you needed — do you look for it in Baked Goods or the Pantry aisle? Whether it is a tiny vial of spices, or that special cold brewed coffee that’s temporarily being displayed at the end cap between aisles 8 and 9, all it takes is one unfamiliar item to derail what would have otherwise been a fairly quick trip in and out of your local supermarket.

At Instacart’s Fulfillment Technology group, our job is to support shoppers by building tools that make doing right by the customer as effortless as possible. One of the initiatives that we’re most excited about is in-store navigation. Our ultimate goal is to be able to tell the shopper the moment they step into the store where every item is they need to pick, and the most efficient route with which to pick them.

Figure 1. An Instacart shopper starts with the list of items that they need to pick, but quickly navigates to the map view. Freely able to zoom and pivot the map, the shopper is able to plan their trip accordingly.

An Ambitious Agenda

Instacart’s vision for in-store navigation encapsulates four broad stages. Stage I centers on refining the core technology required to acquire item location data at scale, and juxtaposing that data within a map of the store presented in-app.

Figure 2. In-store navigation is being built in stages. These are proof-of-concept (Stage I), improved scalability using product categories (Stage II), helping the shopper position themselves within the store (Stage III), and helping the shopper navigate the store (Stage IV).

With Stage II we invest in scalability. It is better to know today what shelf contains the yogurt as opposed to waiting weeks while we accrue the necessary data to map each individual brand of yogurt (see the image second from left in Figure 2). Inferring a specific item’s location is much more challenging (and time-consuming) than inferring the location of a broad category of products. The rate with which we accumulate item location data roughly follows a logistic curve, meaning that while getting from 0% to 50% coverage can occur in as little as a week, getting to 95%+ coverage of all the items within the store takes considerably longer.

Figure 3. While item-level location data is always the gold standard, if we insist on exclusively using item-level data then we will be in for a long wait.

Rather than suffer long onboarding times, we immediately collect and deploy both product category-level location data and item-level location data, making productive use of the former while building up our library of the latter.

Stage III of in-store navigation centers on helping the shopper position themselves within the store (see Figure 2 second image from right). This functionality can be useful in cases where the store has poor signage, is exceedingly large (i.e. big-box retailers), or both.

Figure 4. A typical layout of a grocery store. Note the dual entrances at the bottom. Typically, we would need to make an assumption on whether the shopper has entered through the East or West entrance. Incorrect assumptions may result in an inefficiently sorted shopping list. With shopper positioning we can avoid this dilemma.

However, the key contribution of this feature is that it unlocks Stage IV — shopper routing. With item location and shopper location data in hand, we can call a TSP algorithm on the shopper’s device, guiding the shopper through the store in the most efficient manner possible.

Figure 5: Our ultimate goal — shopper routing — is a dynamic solution for navigating the store in the most efficient manner possible.

Progress to Date

Today the pilot encompasses 300 store locations with plans underway to expand further. The Fulfillment Technology group is actively assessing shopper positioning technology with the aim of commencing testing of in-store routing (Stage IV) later this year. Once the core functionality is in place, we expect to shift our focus towards optimization. While larger, more complex stores might represent the more urgent use case, we also will need to factor for the idiosyncrasies of individual store locations. How often are products retired and how often are new products added to the catalog? Once a product is displayed, how likely is it to remain at its initially recorded location? These questions carry important implications for both the amount of data and the amount of compute required to ensure that in-store navigation is performant.

Acknowledgements

While much work still remains ahead, we’ve only been able to come as far as we have thanks to the efforts of Instacart’s tireless engineers, data scientists, designers, product teams, and operational specialists. Special thanks are in order for Conor Woods, Deborah Lee, Guangshi Chen, Hannah Gardner, Matt Goff, Mei Or, Nik Sawtschuk, Salmaan Ayaz, Saumitra Maheshwari, and Teodor Lefter without whom none of this would have been possible.

Instacart

Instacart

Instacart is the leading online grocery platform in North America, partnering with more than 750 beloved national, regional and local retailers, including unique brand names, to deliver from more than 70,000 stores across more than 5,500 cities in North America. To read more Instacart posts, you can browse the company blog or search by keyword using the search bar at the top of the page.

Most Recent in How It's Made

Griffin: How Instacart’s ML Platform Tripled ML Applications in a year

How It's Made

Griffin: How Instacart’s ML Platform Tripled ML Applications in a year

Instacart’s hyper-growth entails increasing machine learning applications and requires fast iterations of machine learning workflows. To meet these requirements, we built Griffin, an extensible platform that supports diverse data management systems, integrates with multiple machine…...

Jun 22, 2022
Data Engineering and Infrastructure at Instacart with Engineering Manager Abhi Kalakuntla

Taste of Instacart

Data Engineering and Infrastructure at Instacart with Engineering Manager Abhi Kalakuntla

“It sounds cliché, but data drives Instacart’s entire business—our logistics and shopping processes, as well as our ordering experience, are all based on data,” explains Abhi Kalakuntla, Engineering Manager on Instacart’s growing Data Infrastructure team.…...

May 26, 2022
Your Fully Personalized Grocery Store:  How Ads can improve your online shopping experience

How It's Made

Your Fully Personalized Grocery Store: How Ads can improve your online shopping experience

Written by Girija Narlikar, Engineering Director, Ads Quality Imagine your favorite neighborhood grocery store, except that it’s arranged especially for you on every visit. Instacart’s ads platform is powering a marketplace filled with digital stores…...

Feb 2, 2022