Indrani Poddar

An algorithm is a process or set of rules to be followed in calculations or step by step problem-solving methodology followed.
 
In our organisation we have built algorithms using Random Forest model on R, wherein the quality check process has significantly improved efficiency and accuracy of the process. Typically in a quality check process either samples are randomly picked without any logic or rules being or rules defined are typically; all employees with less than 6 months experience 100% audit to be done, critical process 100% audit, 100% audit of all transactions above a certain threshold value, 50% audit between a given value and another so on and so forth. This conventional audit process uses maximum effort and cannot ensure high coverage of errors. What we have done is basis past trend of errors for 1.5 years, identified critical variables and combination of errors and developed an algorithm that gives the list of transactions for the audit team to audit which has highest potential of error occurrence.  The model developed provides the audit team the transactions to be audited from the total number of transactions processed.  Thereby optimizes effort spent and has lead accuracy rate closer to 100% using this predictive analytics. This algorithm also has self learning capability with any new type of error occurrence. 
 
Another example where we have effectively used algorithm as predictive analytics again used Random Forest Model was to predict the number of emails that can be responded in first attempt within a shorter turn around time. In a query ticketing process we get queries from customers in a ticketing tool inquiring about invoice payment status, payment not received, deduction amount, disputed payments etc. The inflow of such emails is about 250 per day with an SLA of 3 days to be responded or resolved. The current process was facing extreme challenges with backlog building up on the queries to be handled leading to miss of SLA and high vendor and customer dissatisfaction. Also simpler queries were pending for a longer turn around time as the team was handling the queries one by one basis ageing. We developed an algorithm using past history of emails / queries closed to predict which queries can be resolved in first attempt and which queries would need more research and hence passed on to the resolution team. This has helped improve current TAT of emails being resolved, higher First Pass Yield,  reduction of backlog of emails without adding additional resource and team is having more qualitative time for research of complex queries or doing effective follow up with dependencies on customer organisation.