Instacart has the largest online grocery product catalog globally, with over 5 million unique items and more than 500 million item listings stemming from 300 unique retailers, from over 25,000 stores in North America.
To keep such a vast amount of data updated Instacart updates 500 million lines of product information daily.
So let’s have a look at the content each product on Instacart should have and how the entire product content system works.
At the basic level, every item in the Instacart product catalog should have a subset of the following product attributes: UPC, Price, Location, SKU, and name.
Universal product code or UPC for short is a kind of code that’s printed on a product packaging to help in identifying the specific item.
A UPC usually consists of two segments: the unique 12 digit number under it and the machine-readable barcode. The machine-readable barcode is a series of unique black bars.
UPC’s make it seamless to identify certain product features like the item’s brand name, color, and size at the point of checkout. UPC’s can also be used in tracking inventory within a warehouse or a store.
The stock-keeping unit or SKU for short is a scannable barcode that you can often find printed on a retail product’s label.
The purpose of this label is to allow vendors to track the movement of their inventory automatically. The stock-keeping unit comprises a set of alphanumeric combinations, which has eight or more characters.
The code, which is made up of the characters tracks the product details, the manufacturer, and the price.
Location data refers to where the retailer’s product is available. This will determine what users can place an order for the item.
This product attribute is simply the price of the retailer’s product. This can be dynamic or static for extended periods.
This is the name of the retailer’s product.
That said, Instacart can also accommodate a more detailed attribute list such as seasonal information, multiple images, dietary labels, (vegan, gluten free, or kosher) nutritional information.
So the more product attributes the retailer provides the more fine-grained and flexible the data taxonomy and search can be for customers and shoppers.
The numbers show that catalog data points on Instacart are updated even more than the number of global tweets every day.
For example, prices, inventory, and any other attribute change every minute depending on the needs of the retailer. And the platform does a great job in reflecting these changes in their product catalog, as inaccurate listings can result in disappointments for the retailers’ customers and shoppers.
Instacart right from day one included tools that enable them to receive regular updates about the retailer’s products.
So when a product’s price changes at a specific retailer’s location or if the items aren’t available, as a retail partner you can alert Instacart in various ways depending on the size of your inventory and the specifics of your technical infrastructure.
However, if a retailer just wants to carry out a one off edit on any of the product attributes, Instacart offers a user-friendly web portal that has a form-fill that permits retailers to include all updated product attributes whenever they choose.
As a retailer, you have the option to place CSV files on the Instacart SFTP server with your updated product attributes. You can drop or place product attribute updates onto the SFTP server as often as necessary, however, many retailers choose to drop it once a day.
Besides this, if you have a sophisticated in-house inventory system Instacart product catalog allows you to provide them with real-time streams of your “balance in hand” data, which provides Instacart with second-by-second updates about the items on your shelf.
Instacart’s API allows retailers to send them changes in product attributes and availability in real-time.
Once Instacart receives all of such product attribute data how do they clean it up? How does Instacart determine the most accurate, the best, and most consistent item attribute to use for listings in their product catalog?
Choosing the best and most accurate attributes is done by Instacart comparing sources against themselves and then the inputs are prioritized.
For example, product availability, which is a store-specific attribute has a huge effect on the shopper and customer experience.
So if the product data Instacart received the previous night in a CSV file drop informs the platform that “Florida oranges” are on the shelf at a store in Sacramento, California but a customer who is on-ground states in their mobile app that the product isn’t available or out of stock, Instacart immediately weighs every input against each other and decides in split seconds, which input is the right one.
Instacart created a system for this, which compares each data source’s reliability at macro levels to know the right answer to the query of if “Florida oranges” are actually on the shelf at the store. These heuristics according to Instacart are created using simple human intuition.
For instance, when multiple customers report a product as out of stock at a particular location, that is considered more reliable most times than a third-party retailer’s data.
Also, product data content updates provided in the last 30 minutes via the retailers portal are most times more accurate than the CSV nightly update.
However, the situation can get more complex.
For example, if a customer and retail manager provide conflicting product data simultaneously what happens?
Because of this Instacart’s system moves beyond the approach just mentioned.
That is because heuristics based on human intuition can only work at low scales and not at the scale the platform currently operates.
These heuristics were created when Instacart had a few product listings in their catalog to handle. Now the brand deals with massive data ingest daily and they have a lot more product updates coming from retailers, shoppers, and brands.
Instacart now uses machine learning models that help to discover signals from the data noise and create fine-grained and better prioritization models at massive scales.
Working with machine learning models helps Instacart arrive at a pinpoint accurate decision as to what product data to project to customers based on what they receive. And this makes it seamless for retailers to have the right product listing available. And not just retailers but shoppers as well.
Knowing how Instacart product searches work helps retailers to know how to phrase their product content.
For example, the average number of searches per basket a user on Instacart makes before checkout is 20. Instacart was built to help save the customer’s time, which means the goal is for the customer to get the most out of all their 20 searches.
However, turning the platform’s 500 million product listings across 20,000 locations into a seamless browsable catalog is tough work.
For example, if the customer inputs tomatoes in the search bar, Instacart needs to determine whether they want dried tomatoes, fresh tomatoes, or canned tomatoes and then rank such suggestions in the most relevant order for the shopper.
To achieve this, Instacart uses technology and years of catalog backlog as well as purchase data.
So using the Instacart search results system, let’s see how it works:
The first step the Instacart search system utilizes is identifying the intent of the user. So based on the search queries used, the platform determines if the shopper is using the app simply to browse their options for inspiration or if they are searching because they intend to buy a particular item that’s on their mind.
If a customer types “tomatoes” for instance, that can mean anything. We could assume that the shopper is exploring different kinds of tomatoes for their diet but if they input “organic green tomatoes” their intent is a lot clearer.
For every search query, Instacart gives a core product and in our example it is tomatoes.
So in this next level of search on the platform Instacart searches to see if the shopper has included product modifiers as these usually review a stronger intent.
In our example, “fresh” and “organic” can be product modifiers. There are also other strong item modifiers like particular departments or brands that help Instacart organize its results.
As a retailer when you put your inventory data it is expected that you organize it by the department but if you don’t Instacart includes it in the preset list of standard departments.
So after using their product modifiers to determine what the appropriate customer’s intent is by using technology for its product match-maker Instacart sieves through their catalog database to show the shopper the most relevant products.
The last step in ensuring that customers are shown the most relevant product is based on Instacart determining what items shoppers like to purchase and then it places such products above the others.
That said, when ranking product results in their app, Instacart uses a combination of historical purchase data, which means what items are included in customers baskets often and keyword relevancw to create an ideal ranking.
It’s important to state as well that the Instacart system learns as more and more products and more and more baskets are included. Using machine learning the system gets a fine-grained understanding of what items are available anf how much relevance thy have to a shopper for each query.
So, for example, when a customer searches for shrimp and Florida oranges, Instacart will use the above process to determine intent query, the catalog, and rank relevance to filter out “shrimp facemask” and “orange-scented dish soap”.
When a customer lands on the search results, before they place items into their basket, they’ll browse by price. So to determine the right price for every basket, Instacart combines logic provided by the retailer, CPG partners, and the customer’s profile.
This is to ensure that if there are particular deals to the product or coupons with the customer, they are calculated and applied in real-time.
This is important because across the Instacart marketplace there may be over 10 million discounted deals running, which can affect the total cost of the basket. So to show correct listing prices in the app, Instacart takes into account the various contextual details assigned to the customer, the retailer, the particular product brand in the shopper’s basket, and the number of items in the basket.
Various varying factors can impact the price of the product picked at the storefront and the larger basket total as a whole.
Here are a few of them:
Product-specific deals: Brand and CPG partners can offer sales coupons for particular products on the platform.
Loyalty discounts: As a retailer you can offer loyalty pricing to specific customers who input their member digits into Instacart user profile.
Deals based on location: Retailers can also set regional allocation specific coupons for sales.
Discounts on delivery: Instacart partners can offer free delivery for first-time buyers and can also waive delivery fees if a shopper puts a particular amount of products in their basket.
Discounts on bundled products: As a retailer when you give customers a particular dollar amount discount on your total basket price. Such data will be applied in the pricing.
To ensure that shoppers basket totals take into account all these details and deals Instacart utilizes a technology that helps search basic product attributes from their catalog (things such as image, listings, set price, item and name) and uses a logic system to apply dynamic price transformations to the products the customer is browsing and adding to the basket.
As soon as the transformations are included into the listing the adjusted product cost is factored into the basket totals of the shopper.
Simultaneously Instacart also pulls in other contextual data related to the customer like first-time order, loyalty cards, and express memberships and applies them in real-time to the basket total.
Enhanced content is product content that has been enriched and expanded above simple data. It is the kind of content that aims to resolve all informational needs and queries the shopper may have and give in-depth and accessible knowledge of the product.
While this isn’t currently available on Instacart it’s in the works.
Instacart product content for the most part is dynamic and the platform utilizes a quality system to ensure that it’s seamless and easy to use for retailers to update their listings and for customers to discover what they’re searching for.
