The travelling salesman problem was mathematically formulated in 1800s by Mathematicians W.R. Hamilton and Thomas Kirkman. Considered as a classic algorithmic problem in the field of operations research, it focusses on optimizing a complex operation. While the problem finds parallels across industries, the complexity often varies and so are the solutions.

One of the salient feature of this solution is to visualize the problem as a network graph. Consider this sample graph; Algorithmic techniques like dynamic programming, Greedy algorithms often try to compute the shortest path between any two nodes. In business context, it’s often the optimal path defined by business rules. With advent of AI/ML algorithms like Bayesian optimization we can not only diagnose weak links in the network but also predict failure points based on real-time operational data

Let’s try and apply this to a typical back-office process where workflows are often complex connecting multiple departments with manual and automated interventions to resolve a problem ticket. The workflow systems are usually process oriented i.e., when a billing complaint is created, the system will define the process flow that this should follow.

The above example is a simple illustration of a process flow. In general, a problem ticket flows through multiple nodes. There will be effort from several human operators and bots. The brain behind the workflow engine is an advanced decision system that will determine the next assignment based on a set of business rules and operational data. Queues also exists for each of these nodes, which means that there will be a waiting time before an operator can act on the ticket.

Challenges in the current State:

 Lack of strong analytics layer- Process defined workflows do not take into account the real-time data on capacity constraints, loops, bottlenecks. The reporting and visualization layer are often inadequate to handle the level of complexity

360-degree view of process flows – A typical retail bank will have between 1000-3000 nodes and 15000-20000 operators. Ticket inflow exceeds 250,000 per month. Lack of understanding in the interconnected paths for these tickets will result in sub-optimal performance

Customer Experience KPIs – While the regular operational KPIs like TAT, wait time, STP are measured, there is no linkage to customer experience. Hence, operational strategies are often linked only to cost and not to customer experience

Backoffice Optimization as a Graph Optimization Problem

 While the traditional process suffers from several challenges the solution to that might lie in the age-old Travelling salesman problem. Visualizing the entire backoffice operations in the form of network graphs will help to better understand the complexity and accelerate process re-engineering.

For consistency, let’s define the following terms.

Ticket – Issue/Incident relating to a customer interaction. This can be a problem or a service request.

Node – Departments where a particular type of work happens. E.g., Bill printing, Bill reviews, Change authorization

Operators – Each node will have one or many human operators who will do the work defined for that department. The operators can also be bots incase that work is automated. E.g., Statement scanning is an automated activity and needs no manual intervention

Edge – The movement from one node to another by means of Workflow assignment. This assignment happens based on a status change to that ticket imitated in the previous node by an human or Bot operator 

Status Change – Change in any of the parameters associated with a ticket – Operator, Node, Contact reason, activity timestamp

A typical retail bank will have between 1000-3000 nodes and 15000-20000 operators. Ticket inflow exceeds 250,000 per month. Lack of understanding in the interconnected paths for these tickets will result in sub-optimal performance. Below is an example of representing back office process as a network graph

The benefits of this solution is quite apparent.

1.      Lucid visualization of complexity: Process owners do not have to struggle with the various reports to deduce process performance. Network graph representation of the process gives an un-blurred view of the operations. One can easily identify the bottleneck nodes and process loops, happy paths by different dimensions such as ticket category, geography and customer segments. Root cause analysis on this data will provide the amount of variance in the workflows and reasons behind it thereby acting as a diagnostic layer.

For Illustration purpose, the above example shows a simplistic view of two processes – Account maintenance and billing. In account maintenance, the shortest path is represented by red colored edges. In the billing, the node D5 colored in blue is a bottleneck node.

2.      Continuous Improvement powered by data driven analytics:  Advanced algorithms leveraging AI/ML (Artificial Intelligence and Machine learning) will help in predicting an event – e.g., bottleneck based on real-time and historic data thereby enable process owners to have a proactive strategy to avert such events. This will lead to data driven process re-engineering and self-healing sustainable process in long run.

3.        Effective Front to Back strategy: Another important outcome of such a solution is an effective front-office to back-office strategy that will link process KPIs to customer experience. Happy path for a customer means that incident should be resolved in accordance with the SLA(Service Level Agreement). This dictates that the workflow should take the path where TAT is lower at the same time gets completely resolved in the first pass (First Time Right). Any exceptions can get flagged and handled in an optimal manner along with pro-active customer communication leading to best-in-class customer experience.

Solution Approach

Implementing this solution requires a multilayered approach. The data layer will have channel, transaction and customer data that can be be mapped to workflow transitions. Large-scale data wrangling requires big data technology as a foundational layer.  AI/ML layer has the algorithms that will diagnose and predict the sub-optimal nodes/edges.  Advanced visualization through dash boarding provides executive/analyst view to monitor KPIs and optimize the process

In conclusion, back-office process visualization using network graphs will not only provide 360-degree view of the process but also drive continuous process improvements. Following are some key benefit areas.

1.      TAT Reduction:  Through data driven process re-engineering, the process owners can reduce turnaround time by 5-10% YoY

2.      Cost Optimization: Improvement in operational efficiency through improved agent performance will drive cost reduction by 3-5% YoY

3.      NPS Improvement: Effective front to back strategy will result in enhanced customer experience  

 The solution will provide both predictive and prescriptive recommendations to process owners’ thereby paving way for sustainable long-term benefits to the process health and help organizations actualize best-in-class customer experience