About the company
Instacart is an American company founded in 2012 that provides delivery and pick-up grocery services. The service is accessed via both the mobile app and the website. The company originally started in San Francisco and expanded across the U.S.A. and Canada in 2017. Despite the early years of unprofitability, mainly due to being in the initial phase of the business (high ops cost, building network, etc.) and having competitors like Walmart and Amazon Fresh, they were able to turn around during the pandemic (COVID-19) in 2020 and gained popularity. Now they partner with companies such as Uber Eats and popular national and regional retailers to broaden their business.
The process of the service model follows these steps:
- Customers choose stores and shop for the items that they want
- They choose the date and time and whether to pick up the items or get them delivered to the desired address
- If picked up, they would head over and pick up the items
- If delivered, a personal shopper near the store would receive and accept the order, and give the items
- The customers pay the total price based on the number of items, distance, etc.
The company's business model comprises multiple sources: advertisements, a commission from contractors/retailers, delivery/service fees, etc.
Since our datasets only contained anonymized customer data, we considered the delivery/service fees a revenue factor. The service/delivery fees start at 3.99 USD and increase based on the number/cost of the items, distance, day of the week, time, etc. However, without having the actual service fees and logistics as part of the data, we set the fees at a constant 3.99 USD.
Goal
With the searched information and the given dataset, we aimed to analyze the data, position VIPs, and propose a strategy to increase revenue.
Data
The provided dataset contained six different tabular data. Upon investigating the data structure via ERD, we sighted the unique and foreign keys and merged them into one table. Then, we cleaned the data by checking and removing duplicates and null values. For instance, the order_id was the unique key in the order table and was the foreign key in the order_product_prior table, so we merged them.
After exploring the data, we saw that the orders mainly occurred between 9 AM - 4 PM and evenly throughout the week, but heavily towards Sundays and Mondays.
When we combined the day of the week and the time, the orders were largely processed on Mondays from 9 AM to 3 PM and on Sundays in the afternoon (12 PM to 3 PM).
Looking at the reordering status, we discovered that the customers were reordering items rather than ordering new items. This led to the idea that they ordered items consumed and ordered frequently, typically within 7 days. We confirmed that was the case by looking at the sales by department and the reordered aisle items, which showed that items from dairy, fresh produce, and snacks were in demand over pets, bulk, and personal care.
Positioning VIPs
Earlier, we saw that the customers requested fresh items within 7 days and used them frequently. The frequency increased revenue via service/delivery fees, allowing us to position VIPs based on the order number (number of times the customers reordered). Upon investigating the order number, we saw that the customers who ordered at least 40 times greatly impacted revenue. They were only 8.68% of people but accounted for 32.97% of total revenue. Through RFM (Recency, Frequency, Monetary), we cross-validated that our VIPs had a huge buying power, and ordered numerous times in short intervals.
Purchase patterns of VIPs
VIPs mainly purchased fresh produce, dairy products, snacks, etc. on Mondays between 9 AM and 11 AM within 7 days. Fresh products and snacks have to be consumed quickly and already play a huge role in revenue, so we wanted to focus on the products that are low in sales and not entangled with certain demographics like pet owners, families with babies, etc. We found out that household, personal care, alcohol, and bulk items were something that we could emphasize and deliver coupons on.
We initially propose to perform an A/B test on the following:
- Give discount coupons to our VIPs
- Discount coupons are only applied to the low-sale products that we initially chose
- Coupons are active from Sunday and expire in 48 hours
Leveraging loss aversion, we hypothesize that once the above steps are taken, VIPs will purchase low-sale items. By doing so we can expect an increase in revenue and an improvement in product life cycle.
Improvements
- Defining other problems and solving them would have been another challenge
- This leads to different approaches to the data
- Strategy based on non-VIPs could have been another option
- Could have approached the data in different ways and methods
- Investigate demographics
'AI > Project' 카테고리의 다른 글
| 2nd K-AI Manufacturing Competition (2) (0) | 2024.12.26 |
|---|---|
| 2nd K-AI Manufacturing Competition (1) (2) | 2024.12.25 |
| 2nd K-AI Manufacturing Competition (0) (2) | 2024.12.24 |
