Incidence routing & knowledge base recommendation


Client, a player in the hi-tech sector, maintained a knowledge base repository of past logs of incidences and resolutions for many of its internal business processes. They wanted to leverage this existing data repository to automate and expedite the routing of its internal operational incidences to the right entities by using ML, as well as reduce the overall turnaround time.

Incidence Routing & Knowledge Base Recommendation


Data Understanding

Data was explored in collaboration with different stakeholders from internal business processes to get a holistic view of relevant information and business rules involved in incidence routing decisions.

Information Extraction

Implemented models for NLP-based extraction of relevant information and keywords from incidences. Fuzzy matching algorithm was used to make the solution robust against minor misspells and variations.

Incidence Routing

The extracted information and keywords were processed by a hierarchical categorization and classification algorithm in alignment with business rules for identifying the right department. Information such as language and proximity were considered for identifying the right location for multi-location departments for a given incidence. In addition, existence of specific keywords in incidences was leveraged for identifying more fine-grained module within a department for quicker turnaround time.

Knowledge Base Recommendation

Existing knowledge base repository was converted into a vectorized form using NLP, which was then used for finding semantic similarity with vectorized incidence. The similarity score was used in conjunction with the incidence routing information derived earlier to recommend top 5 knowledge bases that are very similar to the current incidence, for aiding in quicker resolution.

Feedback and Refinements

Human-aided active learning was used for further refinements in the model, with the help of limited volume of manual feedbacks on the model performance made available by the client.


Solution was adopted and integrated into the client’s business process. Average turnaround time for incidence resolution came down from 3 days to below 6 hours

Case Studies