I would like to give you a general order fulfillment work flow of an eCommerce company (Flipkart will be doing similar to this) here:

Step1-Order Download in OMS: Order is placed by the customer from the front-end (web store) which is then downloaded into an order management system (OMS). This OMS can be a part of your web-store or it can be a back-end Enterprise system where the customer order gets downloaded.

Step2-Inventory Allocation: As soon as the order flows into an OMS, the inventory from the Warehouse gets allocated to the order quantity. Thus the free quantity of that particular SKU (product) is decreased by the order quantity.

Step3-Order Picking: The operations/fulfillment team then start processing the order in the warehouse. First a picklist is generated against that order (usually its for multiple orders at one go and using wave management) and is handed over to a picker in the warehouse to pick that SKU from the bin/rack (in a zone). The picker picks that SKU from the location mentioned in the picklist and bring it back to the picking station (a stage location)

Step4-Order Packing and generation of labels: After the picking is done, the next stage is packing. At the time of packing required documents are printed that needs to be send along with the shipment package to the customer. The order is then packed in a packing box and reports like Invoice, Shipping label are printed and kept along with the shipment.

Step5-Order shipment: After the order is packed, it needs to be shipped out to the customer. The order gets assigned with the courier as per the shipping location (usually either at the time of order placement or at the time of packing) and a manifest is generated. Then the shipment is handed over to the courier guy who comes to the warehouse to pick up the shipment. Once the shipment is out of the warehouse the inventory gets reduced in the system.

Step6-Shipment Delivered: The shipment then gets delivered to the respected customer and the courier company updates the delivery details back to the company whose consignment it was shipping. There can also be the case of customer return or return to origin due to customer unavailability which I am not discussing here.

The below flowchart gives the above explanation in brief: The scenario discussed above is an ideal scenario where the inventory is stocked in the Warehouse. There can be other 2-3 possibilities where the inventory is not stocked in the warehouse, such as: 1. Back to back order fulfillment: In this case the operations download the orders and ask its runner to go to the vendor immediately and pick up the required SKUs from the vendor and bring it back to the fulfillment center to fulfill the order. 2. Drop shipment: In this case the orders are downloaded and handed over to the vendor directly to fulfill the orders and ship it to the customers. The warehouse doesn't have any control over this fulfillment. 3. Made to order: In this case, the orders are first taken from the customer and then the purchase order is raised to procure the SKUs of that order from the vendor. Once the SKUs are in the warehouse, the normal fulfillment cycle (as described above in detail) is followed to process the order.

Flipkart usually does the stocking model or the back to back order fulfillment model. Let me know in case you need to know other scenarios such as customer returns, procurement, etc. in detail.

——— http://www.lean-ecommerce.com/2013/12/product-returns-workflow-steps.html

More important than the "ecommerce platform", a large SKU catalog will need to rely on three major factors: 1. The Db platform: if you really have 5 million SKUs, then MySQL won't cut it. You'll need PostGreSQL or perhaps Oracle (which is pricey). I would stay away from Microsoft on this. 2. If you're going to drop ship, then a decent order broker is a key point. Sterling Commerce leads the pack here, but Oracle also has some systems, as well as ATG and GSI Commerce. 3. With so many SKUs, the site navigation becomes a crux point for conversion. This is probably the most appropriate point when considering an ecommerce platform-- because most platforms are really just a simple front-end engine and then some sort of navigation scheme and search built-in. For the most part, I've found comprehensive packages "okay" at this, but not likely to perform well with that many SKUs. Magento won't work, as most of it's navigation management is manual (there may be some plug-ins). Demandware could work, but it would be because Demandware has some partners that specialize in large-SKU catalogs and can be integrated in the system. ATG would work, (part of Oracle), but it's pricey. IBM Websphere would also work, but also is pricey. ——

We utilize predictive analytics methods (data mining, statistics, and machine learning) across client to identify pattern and correlations in customer and non-customer attributes, interactions and behavior. These capabilities are fundamental to real-time marketing, allowing us to target the right user segments at the right time with the right offers.

Predictive analytics also allows us to accurately measure ROI and optimize future performance and more effectively budgeting spend between channels, tactics, ads and keywords.

Typical Predictive Analytics projects for eCommerce: Customer Micro-segmentation and Targeting: Through undirected and directed clustering (using decision trees) of customer or subscriber databases, we identify key customer segments and recommend messaging based on purchase drivers, typically via email, and direct mail. Product & Content Recommendations: By applying Market Basket Analysis and Association Rule Mining to identify frequently co-purchased/co-viewed products and co- viewed content, we serve product and content recommendations to end users on product pages, landing pages, content articles, and via email. Forecasting: We use multiple linear regression modeling to forecast revenue and optimize media spend allocation. Attribution Modeling: We typically utilize logistic regression models to analyze converting and non-converting user paths, to provide accurate ROAS measurement and conversion credit by marketing tactic. Churn Analysis and Optimization: We usually quantify retention differences using survival curves, and optimize email and logged-in website content messaging to reduce customer inactivity and drop-off.

Other Predictive Analytics Use Cases for eCommerce and Retail:

Location-based marketing Natural Language Processing & Sentiment Analysis Assortment optimization Pricing Optimization Improved inventory management

Some interesting eCommerce Predictive Analytics nuggets from our experience:

Customers who return or exchange products with you are potentially your most valuable customers. Treat them well. For our eCommerce Retail clients, manual product recommendations (based on Merchandiser intuition) rarely, if ever, beat algorithmic based product recommendation strategies. If they do, it's always on sites with small catalogs (limited SKU quantity). Oftentimes, the most tedious or time-consuming part of a Predictive Analytics project is not creating and testing Models. Outcomes are generally delayed by a lack of agreement on the technology or templates to deliver the real-time messaging to users.

====

You could implement some price testing. One example of a retailer that tried to create their own demand curve through price testing is Victoria’s Secret. They sent out catalogs with different pricing, and even though customers ended up paying the same amount for products in the end, it was a way to measure a customer’s willingness to pay. Although this is a bit of a dated example, retailers still continue to experiment. Amazon performed some price tests as well. There can be backlash from consumers if it’s not handled properly or perceived as unfair, so that’s one thing to be careful of.

One thing to keep in mind is that you don’t have to look at competitors and consumers separately when considering the best pricing st