With advanced technologies such as Artificial Intelligence(AI) and Machine Learning(ML) widely adopted by businesses, there is a large push towards building solutions with modern tools and technical frameworks.. However, most of the time, technical advancements are perceived as growth in the technical maturity ladder for individuals, teams, IT or business organizations. Such situations often downplay the value proposition of embracing new transformations.

As a result, businesses tend to deboard such solutions which can cost precious dollars to the business. In several occasions, businesses have not understood the AI engine making decision on behalf of an experienced personnel who does it manually or overrides AI engine outcomes eventually pushing the developed solution to obsolete path. The trust factor of such AI algorithms is at stake with explainability being the centre of a hidden rationale.

This is the second blog in this series which attempts to discover the potential real-word use cases for AI infusion into our IBM Sterling Order Management suite of applications from a business perspective in our customer organizations.

Read the first blog here:

Edition 1: AI infusion opportunities for IBM Sterling Order Management suite of applications with real-world use cases | Persona: Call Center Agent

Edition 2

Persona : Fulfillment Manager

Problem statement

• In a supply chain, when inventory is assigned and reserved for an order, First-In-First-Out (aka FIFO rule) is the most common and widely used scheme. The IT applications model FIFO needs through a concept of inventory reservation which manages the queue of demands (i.e. orders) identified by demand date and match the supplies based on earliest available dates (aka ETA)
• It is very common for business lines, subsidiaries, partners, and stakeholders to take deviations in FIFO rules in certain circumstances due to various internal and external factors. However, deviating from the regular FIFO protocol would imply that certain orders skip the queue, race ahead and start stealing inventory while the orders which are pushed back in the queue lose that intended inventory allocation. This results in altering the original order promised dates to orders from which stock is claimed/stolen and has far reaching implications across multiple departments of business potentially affecting compliance scores, customer satisfaction, social media effects contract negotiations, etc.

Solution outline

• To address business problems like that defined above, businesses need a way to achieve the practical needs of FIFO queue jump which give them business agility and advantage, while safeguarding the interests of their ecosystem by minimizing the impact of the negative side effects explained above
• IBM Sterling Order Management manages the demand (orders) and allocates (reservation) the inventory based on available supplies in FIFO orders as required by business
  
  
  1. The solution gets triggered on the events like the following:
     • Inventory updates in the supply chain like Order Cancellation, Inbound and Outbound Shipment, Picking Operations like Short Pick, Corrections like Warehouse Count
       
     • External/internal Events, like natural disasters, weather, business acquisition and shipping calendar
       
     • Capacity changes, in the supply chain like Operations, Accidents and Unplanned Leaves
       
       
  2. The order management application also is cognizant of buyer/seller profiles on the orders placed. This is made available through third party application such as CRM or Vendor(supplier) management suite of applications:
     • If required, a real-time internet search can be performed using cognitive tools to know more about the profiles
     • The order management application continuously tries to optimize and right supplies for the right demands based on dates, inventory stock on hand, safety factors, etc.
     • During the continuous optimization of inventory allocation cycles, order management application understands the FIFO queue depth, similarity between orders with similar set of items ordered, order total dollar value on the, promised dates, historical item velocity of FIFO queue, etc.
     • Decision for the need of reprioritization is made, based on appropriate threshold/condition like
       • Threshold on inventory and capacity (Example: Trigger the system only if the change involves more than 100 units)
       • Item Type (Example: Trigger only for a set of items)
       • Price (Example: Trigger only when unit price is above $100)
         
         
     • Post the decision made by system to alter the FIFO queue, the solution leverages ranking algorithms to determine the best “Order” candidate to cause least disruption based on historical customer sentiments, past cancellation rates of customers, dollar value gain/loss probability values.
     • The order fulfillment application/optimizer employs machine learning techniques to generate insights on “What If” analyzer for system to determine the impact of jumping the FIFO queue for an array of orders.
     • The solution can use techniques like multi-linear regression to determine the relationship function of jumping the queue v/s negative implications.
     • For each order, the order level insights are used by the system to consider the order for reprioritization (re-arranging and recalibrating the supplies against demands).
       • This can be threshold driven (Example: Reprioritize only when the impact is more than $1000)
       • This can also be condition driven (Example: Reprioritize only for Gold customers)
       • Predicted business (Example: Do not reprioritize if expected order value within next 1 year is greater than $1m)
         
         
     • This is also explained to business personnel with reasoning why a particular “Order A’ was traded off with Order B. Basically, system also provides an opportunity for the right personnel (like Fulfillment Manager) to trigger/view the “What-If” from the UI, to offer visibility into actions and justifications of recommendations. Along with the indication of why order A was prioritized over order B, system also provides additional insights like what is the business gain out of it
       
       
     • Solution provides an opportunity to the personnel to override the recommendations with reason codes and uses this data as a feedback for reinforcement learning for subsequent recommendations
       • Overriding request can be conditional (Example: Ask only when the order total value of an order getting deprioritized is above $10000)
       • For automatic reinforcement learning feedback, the system is integrated with an appropriate system like CRM
       • For example, using NLP and Speech-to-text conversion techniques, system can determine if the customer, whose order was deprioritized, logged a complaint about the same.

Sample Use Case

1. “Your Home Dreams” is a large furniture retailer.
2. The company takes in orders at stores for kitchenettes, beds, sofas, large tables, dining sets, etc.
3. Due to the nature of furniture industry, which deal with bulky items having significant lead times for procurement, manufacturing, assembly and final delivery, business enforce to deliver together all items on the order together for deriving supply chain efficiency and reducing logistics costs.
4. Say, orders are placed for similar bed-set by n number of customers
   
   

5. Business personnel observes that some of the orders having this bed-set are large with multiple line items inclusive of non-bed-sets as well. This attributes to a large financial order value on few orders having the bed-set while few orders have just the bed set and hence order value is small.
6. Upon drilling down further, business analytics indicates the personnel that large order (Order n) for Customer N is behind the queue and has a potential risk of breaching promise dates and hence leading to cancellation.
7. Without the implementation of the solution
   • System will not give the earliest stock allocation to the large order behind the FIFO queue because of lack of visibility on negative implications.
   • If a human intervention is involved, the personnel uses manual intuition to take decision which is based on individual risk appetite and at times potentially biased as well.
   • If the decision is to steal stock, the system/personnel ends up in next question on which order to steal stock from, which is again a blind question in the situation.
   • If the decision is to not to steal stock, business risks a cancellation of large order reflecting badly on the financial books
     
     
8. With the implementation of the solution
   • System leverages the recommended “What-If” scenarios to understand the pros/cons of giving the earliest stock allocation to the large order.
   • System has data points on Order n about breaching of promise dates on an earlier order and financial impact score.
   • System determines that Order 3 by Customer C has the least affinity towards cancellation and dissatisfaction risk in the queue.
   • The system provides recommendation to steal stock from Order 3 by Customer C to Order n by Customer N.

Read the next edition of this blog series that details the use case for a Call Centre Agent:  

Edition 3: AI infusion opportunities for IBM Sterling Order Management suite of applications with real-world use cases | Persona: Customer Relationship Manager

Stay tuned for the next Edition of the use case!