Introduction

E-commerce and wholesale companies benefit from displaying delivery dates on their websites because it directly impacts :

• customer experience,

  • Transparency builds trust in the brand.

  • Avoids frustration caused by vague shipping timelines

• conversions,

  • Customers want certainty before making a purchase.

  • Helps decision-making in time-sensitive purchases.

  • Reliable delivery builds repeat customers and brand loyalty.

• operational efficiency.

  • Helps compete with other sellers who may have faster shipping.

  • By displaying delivery dates, companies reduce unnecessary customer service workload.

  • Delivery date estimates help align inventory, fulfillment, and logistics.

Problem Statement

For customers browsing the Ankorstore website, we want to show the estimated delivery date (PDD) for each product the customer views.

The PDD can depend on multiple factors. The goal is to take into account all these factors and provide a near-accurate estimated delivery date for each product that a customer views on the website in real-time.

After the feature is in production, evaluate the behaviour of customers and business value provided by this feature.

Design an architecture for the feature with millisecond latency and easy updating on business SLAs.

Implementation

Define PDD

• Plan workshops with data analysts, data scientists, logistics managers and front-end developers to define what a predicted delivery date is, and the relation between a customer and the PDD.

  • PDD is a date in the future.

  • It depends primarily on the customer (location, date of order), product (in stock / out of stock / location of inventory) and the brand that sells the product (location, how the brand ships products)

  • It also depends on days (bank and public holidays), logistics SLAs and business SLAs.

• Using these factors as data points, create analytics data tables (in dbt).

  • These tables are pre-calculated on every 12 hours. They will be used by the real-time analytics engine.

  • The tables are stored in BigQuery

Prepare data for real-time analytics

• Using Airflow, we send data from the data warehouse (BigQuery) to a real-time analytics engine that can process data with low latency and high throughput.

  • Clickhouse, MySQL, Apache Druid are some options for OLAP analytics engines.

• The data that is used by the processing query is stored in Clickhouse.

Prepare PDD API

• A REST API is called with a payload necessary to calculate the PDD date. The payload consists of :

  • customer id

  • brand id

  • product id

• The API queries the analytics engine which calculates the PDD within milliseconds.

• The API then responds to the request with this PDD value.

• The API logs are stored in BigQuery for monitoring of the feature

Prepare front-end for website

• The website has a section on which the PDD is displayed.

• When this section is loaded on the website, the API is called with the payload

  • customer_id of customer browsing the website

  • product_id of the product displayed on the webpage

  • brand_id of brand selling the product

• The API returns the response as a PDD date that is displayed within milliseconds

Diagram