E-commerce and serviceability: Yet another story of breaking down the monolith at TataCLiQ

Introduction

If you have tried to purchase a product on any of the e-commerce websites, one of the must-do steps is to enter your zipcode/pincode and check if the product can be delivered to your location are. This is the point where the customer intent is being changed from informational to transactional. Why is this moment so important? Because there are 3 things happening at this moment…

1. The delivery capability of e-commerce platform/seller is being determined.
2. The customer is getting to know by when he is likely to get the product (i.e., if he is likely to get the product)
3. “Lead” is getting very specific for the platform — providing details of demand by location and product

The Complexities…

1. Multiple pick-up locations: At Tata CLiQ a significant volume of products get serviced from customers' closest locations. This capability of ours increases the lanes in which we must operate.
2. 3rd Party logistics partners: We utilize the capabilities of different delivery partners. Some of these delivery partners have been delivering products/documents throughout the country for years now. Each of these partners has strengths in certain lanes and for certain product categories and our job is to put their prowess to the best of use for our customers.
3. Limiting serviceability: Some of our partners, wary of damage during transit, want to limit delivery items to close by regions. These restrictions are also a result of brands being able to provide warranty and related services.
4. Spread of TAT: At Tata CLiQ we not only focus on adherence to promise a delivery date, but we also closely track our accuracy to that promise. While in some parts of the country the spread of delivery can be tight in others this can really spread out with attempts starting from as early as 3–4 days before the promised date.

The Challenges…

With the above complexities, the data model over the period of time evolved into a huge monolith and presented the following challenges…

1. Feasibility vs Capability: The model started drifting towards feasibility ( degree of convenience) rather than capability
2. Lack of visibility: In the past couple of years we have grown at breakneck speed with thousands of integrations and new stores getting onboarded. This in turn exponentially increased the data and visibility of this data became a big challenge.
3. Friction for customers: A lot of our customers would see errors that were unbelievable for them — “you don’t deliver in Andheri” was our nightmare.
4. Unclear accountability: The error messages shown to customers or tracked on our data platforms had some grey area in terms of ownership — operations vs category.
5. Scalability challenges: The growth in the size of data was exponential rather than being logarithmic or linear at worse.

Impacted metrics:

1. # of serviceability requests declined on the platform
2. Accuracy of Estimated Delivery Date (EDD)
3. Cart add rates
4. Customer contacts for order status
5. Perfect Order %

The change….

1. Breaking down of the monolith: We decided to breakdown the monolith into 3 different components based on importance of each in the customer journey…

a. Serviceability decider engine

b. EDD engine

c. Allocation engine

2. Focusing on clusters instead of specific pincodes: On doing analysis we figured that variability in the performance of delivery performance is a factor of the cluster rather than a specific Pincode.

3. Removed grey areas in messaging: We improved our logic to have unique customer messaging for each of the reasons why we would be unable to service the product to customer.

4. Separating flexibility and capability: For each of the destination pincodes in the country we would have the capability of our logistic partners. In the earlier logic team was using the capability parameters for flexibility. This again created lot of confusion and errors on the platform.

5. Model replication: The same model is replicated in case of reverse operation as well.

How did we fare?

1. # of serviceability requests declined on the platform -20%
2. Accuracy of Estimated Delivery Date (EDD) +10%
3. Cart add rates +0.5–1%