XX.V. Decision Chain
The 5th process architecture pattern: a routing process with multiple destinations
The previous chains that we discussed were mainly for processing incoming ingredients and producing a quite standardized product according to a specific recipe at a certain end-point. Now we will consider a decision chain with multiple destinations to which an item may be delivered, depending on the routing logic of the process.
Decision chains represent distribution and decision-making pipelines. In a physical distribution pipeline, physical objects or people are routed to a physical destination. At each gateway of this process the system needs to choose where a given physical item will be dispatched next. In a logical decision-making pipeline, information requests or data travel through a logic scheme to the logical destination — the final decision, a choice out of two or more options.
Examples of physical pipelines: Complex marketing funnels, complex manufacturing lines that assemble different products out of the same components, and a retail distribution network.
Examples of logical pipelines: Software and digital products, data pipelines, decision-making process, Decision-to-be-Made logic, and insurance claim processing.
Research landscape of the decision/distribution chain
There are four main research spaces to be explored: Gateways, Destinations, Factors, and Flows.
Gateways
A gateway is the step in the pipeline where a routing choice is made on where the item or data will travel next. In fact, this ramified chain is a network of simple chains connected through the gateways. Each gateway receives the output of the upstream chain, decides where to send it next, and redirects it to the chosen downstream chains. Your research team will need to understand this routing logic during your quest. These routing choices may be made according to a simple rule, or to a more sophisticated algorithm that needs to be explored.
Examples of pipelines with gateways: Mortgage application decision process at a bank, economy and business class lanes at the airport, customer service call routing, image classification pipeline, a distribution center delivering FMCG goods to the retail stores, quality inspection at the factory, insurance claim processing logic, and packages sent via delivery service.
Destinations
Destinations are the final points at which items arrive at the end of the process. Physical destinations are places that receive physical items.
A logical decision-making pipeline is designed to make a choice between options: approve or reject, buy/sell/hold, choose between candidates, choose between suppliers, and so on. A logical destination is one of these alternatives. When a new question or request that requires the decision arises, it goes through the decision-making pipeline and “arrives” at the destination — the final decision.
Examples of logical destinations: Go/no go decision of an investment committee (2 destinations/choices: “go” and “no go”; projects seeking investment are items that travel through the pipeline) and choice of an accounting firm for a new jurisdiction (multiple logical destinations/choices: a list of top accounting firms; a new jurisdiction will become an item that travels through the supplier selection pipeline).
Some destinations are more business-critical than others. The architecture we consider, the decision chain, is designed exactly to provide accurate routing through gateways in order to make sure every item is dispatched properly.
Factors
When an item arrives at each gateway, how does the system make a routing choice? This choice is made based on the attributes, or factors, associated with the item. These factors may be observable or hidden, distinct or fuzzy, predictable, or uncertain. If the wrong routing decision is made, the item will arrive at the wrong final destination, which is usually bad.
Examples of routing decisions made based on factors: Delivery of packages based on the QR code, mortgage application decision based on the loan amount and the credit score of the applicant, quality inspection based on visual defects and sensor signals, and hospital treatment based on the medical test records.
Flows
When an item arrives at each gateway, not all routing options are the same. Most items may be sent one way, while a small number of others may be sent to other quite rare or even exotic pathways. Thus some branches in a complex pipeline are mainstream, others are thin flows. Some transmit 60, 80, or 90 percent of items, others may carry only 10, 5, or even less than 1 percent. But that does not mean thin streams are insignificant. For example, a thin flow may contain carefully selected items with high importance for the business. If so, it is even more important to be mindful of accurate routing at each preceding gateway.
Examples of thin and thick flows: Fraudulent vs normal transactions at a bank, defective vs normal products at the factory, positive and negative patients for a rare disease, high-rated aspirants vs not-so-successful at the top-rated University, and hot startups vs others in the venture fund pipeline.
Problem analysis
The main four problems in a decision/distribution chain are usually:
- Wrong routing decisions and the eventual inaccurate delivery to the final destination
- Delays on certain steps and gateways in the process
- Low throughput of the system
- Effectiveness and cost of the processing on each step as in any other chain
When an item is delivered to the wrong final destination, it is a mistake of the whole pipeline. These mistakes can cost a lot. For physical pipelines that means expensive reverse logistics, for a logical pipeline that means a wrong business decision.
In an ideal world, your quest givers want to have zero mistakes, but that is almost impossible. However, what you can do is to identify the gateway where the wrong routing decision was made in the pipeline and reengineer them, ensuring that they make more accurate routing choices in the future.
There can be two types of mistakes:
- False negative error: The destination point does not receive an item that was intended for this point. Business is usually very sensitive to this type of error if the items that are expected at destination are expensive or valuable.
- False positive error: The destination point receives an item that was designated for another point. This type of error is usually less critical for business but it implies unwanted expenses to redirect the items back to the pipeline so that they will be processed again and delivered properly.
Examples of false negatives: A damaged bottle was not identified as a defective one during quality control, an infected patient was not diagnosed properly and was sent home, the payment transaction did not go through into your account, and your luggage was sent to another city.
Examples of false positives: A normal bottle was mistakenly identified as a defective one during quality control, a healthy patient was diagnosed with a disease and was sent for additional clinical testing, someone else’s payment transaction by mistake came into your account, and unclaimed luggage was found at the airport.
In so far as not all destinations of the pipeline are created equal, your goal is usually to identify the most business-critical and understand what types of mistakes are most harmful. Also, you should explore the frequency of mistakes and volume of flows that dispatch the items to the destination.
The eventual mistake accumulates the inaccuracies made on each gateway during the dispatch. Knowing that, you need to identify how inaccurate routing on the preceding gateways contributes to the eventual mistake.
Common flaws and illnesses of a decision/distribution chain:
- Delays and long wait times
- Errors in routing that make some portion of the items arrive at wrong destinations
- Low throughput of the system
- Expensive processing at some steps
Leveling guide — Decision/distribution chain
If you’re designing and building a process from scratch:
1. Create Placeholders of all steps of the pipeline. Identify the steps that are necessary to deliver items to the main destinations and upgrade them to the Quick & Dirty level (Low Fidelity, Medium Quality). Some steps may still remain placeholders. No worries, we will upgrade them later.
If you are improving an existing process, skip phase 1 and go to the next phase.
2. Target discovery. Identify the main destination that currently suffers from mistakes, wrong deliveries. If the destination point does not receive parcels destined for this point (false negative error), or it receives too many items that belong to other points (false positive error). Explore which gateways in the pipeline make the most contribution to these errors. These will be your targets for improvement.
3.Analyze the flows and frequencies. Which branches of the pipeline are supposed to be mainstream and which are thin flows of rare items. Reprioritize targets based on this knowledge. Having a more detailed knowledge of flows you may now calculate a more precise contribution to the eventual deliveries and mistakes.
4. Upgrade the target gateways to the Good Enough level. Analyze the factors that determine routing choices. Get rid of insignificant and amplify the most meaningful factors. Improve routing algorithms based on those factors.
5. Find sources of additional data and enrich the set factors. Continue improving the routing algorithms. Focus on the steps of the pipeline that are on the way to the most business-critical destinations.
6. Continue identifying bottlenecks and hindrances. Further enrich the set of factors, upgrade the gateways to Good Enough and the most important ones to the State of the Art level.
Always evaluate not just improvement of the individual steps, but also the performance and the accuracy of the system as a whole.