Deep Learning for a Deep Supply Chain

It was early 2004 and I was catching up with a friend from my college days.  He was the regional head for a Consumer Goods major in India, and told me about changes they were making to streamline operations. That included consolidating the number of stockists and wholesalers. I was puzzled because the company was known for its reach even in remote rural areas, enabled in large part by these very channel partners they wanted to terminate. Our conversation went something like this:

  Me:       This is counter-intuitive.  Your revenues will tank.

  Friend:  We will make up a good chunk of it by expanding the territories of the partners we retain. 

  Me:       But why are you doing this?

  Friend:  The wholesalers we are terminating have poor IT capabilities, and struggle to keep us posted on the inventory they hold. Forecasts have also been poor. There will be major cost savings. We can reduce inventory and wastage.

  Me:       You could have asked them to upgrade their systems.

  Friend:  The investment may not be worth their volumes.

It came across as a rational decision – information-flows are the lifelines of effective supply chain operations. Results in subsequent years vindicated the company’s decision.

I wonder if a similar situation today, in 2017, would lead to the same decision.  Let us not forget that these partners bring some unique capabilities to the value chain – convenient deliveries, credit availability, local knowledge, unique products, attractive prices etc. Can today’s machine learning technologies allow smaller partners at the edges of the supply chain to communicate better with their brand-owners?

Let us start with inventory.  What if the wholesalers could just take snaps of their warehouse from different angles and upload them to an app provided by the brand-owner?  Deep Learning algorithms are pushing the frontiers of image recognition.  It should be possible for them to infer the SKU-wise inventory from the snaps. All that the wholesaler needs is a smart-phone.  Expensive RFID readers and packaged software systems can wait until the business grows.

Coming to sourcing, Food & Beverage companies are dependent on the farmers’ ability to supply the committed quantity and quality. Today big orchards use drones that go around trees and count the fruits at various stages of growth. This enables the farmer to forecast her harvest with unprecedented accuracy.  This is possible not just because of the modern day drone’s ability to fly around tight spaces. Improvements in image/video recognition also play a big part. The marginal farmer may not be able to afford a sophisticated drone; but like the small stockist at the other end of the supply chain she too has a smart-phone handy.  Once again, the brand-owner just needs to have a means of extracting information from videos and pictures the farmer uploads.  Not too difficult, with today’s deep learning techniques!

For brand-owners/ retailers, sourcing from smaller manufacturers with less robust practices brings in the risk of unpredictable quality. This often means having a team of people check the received goods. In many cases, these are just visual checks. An algorithm can easily do them today.  The manufacturer just has to send photographs of the goods before shipping them.  The algorithm can clear or reject the lot even before it leaves the manufacturer’s premises.

For a long time, managers have managed the deeper ends of their supply chain with trepidation. They use aphorisms like “flying blind” to describe the way they manage tier-3 or even tier-2 suppliers/ distributors. They hang on to these partners mostly due to lack of reliable (read larger) alternatives. Very soon, this could be a story of the past. The power of Deep Learning is already being harnessed to make it easier and less expensive to manage deeper supply chains.