Bound by technology, driven by data, and equipped with the smartest devices, everything is just a matter of a click. It feels amazing when we get instant fixes to our queries; be it a refund from an ecommerce website or canceling a flight with the help of a chatbot. The seamless experience is enabled and accelerated by the wide umbrella of digital services offered by a business.
Today, most customers engage with a business or brand is through its digital services and applications. The digital service of most organizations is driven by efficient IT operations team/customer support teams delivering quick fixes to customer request. In order to increase the delight value, niche technologies like AI, NLP and machine learning aid the IT operations. This gives a unique combination of 2 technologies working in tandem: Artificial Intelligence and IT operations (AIOps)
AIOps makes use of the data deluge around us by using big data, machine learning and automation technologies to automate IT operations (automation, monitoring and service desk operations).
Why Is AIOps the new necessity?
Gartner predicts that large enterprise use of AIOps tools to monitor applications and infrastructure will rise from 5% in 2018 to 30% in 2023.
The AIOps market was valued at USD 1.64 billion in 2019 and is expected to reach USD 6.88 billion by 2025, at a CAGR of 27% over the forecast period 2020 – 2025.
Hence, integrating business and IT operations with AIOps is the need of the hour to derive increased potential value that includes:
To efficiently handle complex data generated from varied data sources and filter out meaningful patterns
Root cause analysis of commonly occurring issues and to solve them without human intervention
Work smarter and faster by resolving data and digital issues much before customer or business is impacted
This integration will prevent outages, failures and assure delivery of quality services
It also ensures reduced MTTR, proper SLA compliance and improved CSAT
In a similar scenario, Brillio democratized a CSS support system. The following approach was undertaken to solve the problem statement in concern.
The objective was to Identify an optimized solution through Root Cause Analysis for Power BI Tech Support by enabling the support team for faster trouble shooting thereby reducing TAT and improving CSAT score for tech support services.
Current Client’s Business Scenario
Currently the help desk is assisting its customers via phone and email support. The ticket volume can be expected to be around 4K per month. All the tickets are logged in Ticket DB along with their status. KMDB(Knowledge Management Data base) and KEDB(Known Error Data Base) are updated and stored from time to time. SQL DB is used for storing the tickets and documents.
Scenarios in Scope:
Complete information about the problem is provided
Partial information about the problem is stated
Problem is unique or first of its kind
Auto Resolution – For most common scenarios faced
Solution for the above-stated problem is tackled in 5 phases and discussed in detail in the coming sections. Firstly, the Input data is received via email or voice call, if its voice call then the customer side conversation is transcribed and sent as input. Secondly, we would go ahead with the data preprocessing where the data is cleaned and modified according to our needs. In the third phase we would tokenize the issue vector and search the relevant documents and past tickets from our database and the most relevant documents and tickets are passed to the next phase. In penultimate phase, fourth phase, we would summarize the agent notes from the previous tickets and scarp the relevant section related to the problem from the documents which were identified in the previous step. Finally, we send the summarized agent notes and documents to the agent to resolve the ticket quickly. In case of partial data furnished by the customer, suggestion is given to the agent to ask the follow up question. If the ticket is unique then the agent solves it and updates the knowledge documents so that for the next time this issue is easily resolved.
Detailed Phase Wise Approach:
In this phase the input email or the transcribed voice call of customer side is being fed into the model
Data, which is obtained must be cleaned, so we will start with converting the entire text into lower case and then perform stop word removal, where the commonly occurring stop words are removed. Then the words are converted into their root form by lemmatization and finally converted into tokens
This is one-time activity where the keywords are extracted from each document and converted into tokens with the help of the vectorizer. In this phase each set key words- unigram, bigram, and trigram- are assigned certain weightage based on their occurrence within the document and their occurrence among the rest of the documents. This process is repeated after a certain predefined time frame to include the new tickets and updated knowledge documents.
Missing Aspect addition:
Our model based on the past ticket data will try identifying the correlation between the words and based on this association probable missing aspects are added to get a better result when a search is performed.
Search based on Similarity:
Once the documents and tickets are converted into vectors, using the same vectorizer the issue statement is converted into vectors and mapped in the same vector space. Based on the similarity search such as cosine similarity the nearest vectors and the corresponding knowledge documents and the previous tickets are selected. In case of more than a single ticket for the issue is identified then the ticket with least resolution time, high CSAT score and latest ones are picked
Text summarization and Document Scraping:
The agent note of the top similar tickets are sent into text summarization model where the extractive summary of the agent notes is sent to agent. In case of knowledge documents the similar documents are loaded which are then converted into raw format by parsing the document data. Extraction of the relevant content from the parsed content is done and then transformed into a required format.
Finally, the agent is provided with the summary of the previous tickets and the relevant section required to resolve the ticket, which facilitates in faster resolution of the tickets.
Reducing the amount of technical expertise required for a agent to resolve the ticket with the help of the model which in turn reduces the ramp up time from months to weeks.
Reduce the ticket resolution time from days to minutes by reducing the number of iterations.
Improved CSAT score by faster resolution
Improving the efficiency of the agent enabling the agents to handle higher ticket volume there by cost per ticket going down
We chalked out the above detailed approach which included the integration of machine learning and NLP to automate the search process for the agent, thereby recommending integration of AI and CSS support. Thus, riding on the wave of digital transformation and cloud adoption, AIOps equips modern businesses and organizations to deliver value and adopt to ever dynamic market changes seamlessly.