The Work Ahead in Intelligent Automation: Coping with Complexity in a Post-Pa...
Optimizing Customer Support with Machine Intelligence
1. Optimizing Customer
Support with Machine
Intelligence
By applying learning algorithms, customer care
centers can arm agents with the ability to quickly
troubleshoot and proactively resolve customer
challenges.
2. 2 KEEP CHALLENGING December 2016
Executive Summary
For most product/service companies, a major portion of their
maintenance expenses are typically directed at customer care, with
customer support agents playing a crucial role in continuously
improving customer experiences by supporting and troubleshooting
device and systems issues. At the same time, many customer care
organizations struggle to answer common issues, such as how to:
• Reduce the time spent by customer care agents to resolve problems.
• Improve the customer satisfaction index.
• Expand the troubleshooting guide for new issues and symptoms.
• Create a user-friendly system to ease problem resolution for
customers.
• Reduce support center expenses without compromising quality.
3. OPTIMIZING CUSTOMER SUPPORT WITH MACHINE THINKING 3
We believe machine learning will play a major role in addressing
these critical customer care support challenges. Specifically, machine
learning can automate the process of identifying problems and
recommending fixes, either with or without agent intervention, which
speeds problem resolution, improves customer satisfaction and
reduces costs. Machine learning-based solutions can create an expert
system for agents that continuously evolves, based on historical and
current troubleshooting data.
This white paper reveals how machine learning technologies can be
effectively applied to resolve many of the challenges that roil customer
care support centers. It also details how our machine learning solution,
code-named Cognizant ASIMOV, can help address customer care
troubleshooting challenges.
4. 4 KEEP CHALLENGING December 2016
Customer Care Center Flow Challenges
In a typical customer care center, the workflow begins when the customer calls
a support agent, who collects symptoms of the problem, makes a diagnosis,
collects more clues, and refers to the knowledge base/troubleshooting guide for
solutions. The knowledge base/guide should contain detailed steps that the agent
can recommend to the customer. This is generally an iterative process, involving
multiple steps until the problem is resolved (see Figure 1).
In the archive, troubleshooting data is recorded in the form of log data that contains
information about the conversation. Generally, this content is neither structured
nor straightforward, making it difficult to summarize without human assistance.
The ensuing data complexities include:
• Troubleshooting data is represented in a chatty question-and-answer
pattern. The data is typically a dump that includes Q&A sessions between agents
and customers, diagnostics information, responses from devices, actions per-
formed to resolve the issue, and the outcomes of those actions.
• The troubleshooting flow differs for each product, based on the issue observed,
symptoms and line of business.
• The overall volume of troubleshooting data that is collected for each interac-
tion is typically very high.
The Virtuous Customer Care Cycle
Figure 1
Verify Problem
Diagnostics
Recommendation
Resolution
5. OPTIMIZING CUSTOMER SUPPORT WITH MACHINE THINKING 5
Applying Machine Learning to Customer Support
Using pattern recognition and computational learning, machine learning involves
the construction of algorithms to learn from and make predictions on data.1
Machine
learning algorithms have existed for decades but were used only for limited scientific
purposes because of their cost and dependence on high levels of computing power.
In recent years, as per Moore’s Law, computing power has increased as provision-
ing costs decreased, and with the advent of big data ecosystems, organizations can
now process and analyze large volumes of data with machine learning algorithms.
Today, practical machine learning solutions have transcended high concept to
become near-term reality across multiple disciplines and industries.
With customer care, the primary focus of the machine learning solution is to equip
the system with:
• Expert knowledge to understand the troubleshooting log.
• The ability and skill of support agents to resolve customer problems.
In summary, machine learning helps gather the knowledge of hundreds of agents
and subject matter experts (SME), and feeds the knowledge to support agents
to accelerate time to resolution and effectively solve customer challenges (see
Figure 2).
Figure 2
Human vs. Machine Customer Care
SME
Support
Agent
Machine
LearningUnderstand the insights in unstructured
troubleshooting logs for a limited
volume of tickets.
Understand the insights in
unstructured troubleshooting logs
for a large volume of tickets.
Understand the current problem
of a customer and correlate it with
the limited learnings of the agent.
Recommend an effective solution
based on a large volume of
history tickets.
Today, practical machine learning solutions
have transcended high concept to become
near-term reality across multiple disciplines
and industries.
6. 6 KEEP CHALLENGING December 2016
Introducing ASIMOV
ASIMOV is a machine-learning-based solution developed by Cognizant’s Global
Technology Office High Performance Computing Labs (HPC Labs). The system
contains software and processes to address common customer problems and
overcome customer care troubleshooting challenges (see Figure 3).
ASIMOV’s machine learning software can help customer care organizations do the
following:
• Extract insights from complex troubleshooting data.
• Correlate various symptoms, problems, actions, fixes and resolutions in the trou-
bleshooting logs.
• Recommend fixes and solutions based on existing symptoms to quickly resolve
customer problems.
ASIMOV works by creating a mathematically-based predictive machine learning
model using the data gathered and analyzed by subject matter experts. Using a
Making Customer Care More Cerebral
Figure 4
ASIMOV’s Foundation
Figure 3
ASIMOV generates machine intelligence by gathering and
applying agent and SME insights.
= +
TROUBLESHOOTING
HISTORY DATA
SUPPORT AGENT RECOMMENDATIONS
INSIGHTS
Customer Support Troubleshooting Assistance
CLASSIFIER DECISION TREE RECOMMENDER
7. OPTIMIZING CUSTOMER SUPPORT WITH MACHINE THINKING 7
large volume of troubleshooting historical logs, the system generates a more gen-
eralized machine learning model for predicting problem resolutions and recom-
mendations to customer issues. When a support agent feeds the system with a clue
or looks up a current customer problem, ASIMOV quickly predicts and recommends
the best solutions based on what it has learned from previous customer challenges
and related fixes (see Figure 4, previous page).
This way, agents can access a solution that has proved effective for other agents, and
can navigate the recommended path to quickly resolve the issue. The result: Agents
can recommend solutions with more confidence and resolve issues more quickly.
ASIMOV is designed to work on top of a big data architecture, enabling it to effi-
ciently handle large data volumes in a distributed manner (see Figure 5).
How ASIMOV Improves Customer Care
Troubleshooting
Consider the current experience of buying a product on a retail site. Many sites
typically highlight and recommend options based on prior customer interactions
and transactions to help customers choose the best product. In a similar manner,
a customer care agent with only minimal knowledge of a specific problem can be
assisted by ASIMOV to quickly choose the most accurate and specific resolution.
ASIMOV uses defined processes and algorithms, from data ingestion through data
extraction, modeling and validation. It also uses a recommendation engine interface
to guide agents for faster resolution.
Figure 6, next page, depicts a hypothetical use case for how ASIMOV uses recom-
mendations to resolve a specific customer issue involving low-speed connectivity.
The system begins with broad recommendations using the initial symptoms
provided, and when more symptoms are injected into the system, it offers increas-
ingly personalized recommendations, based on the history of continuous learning
it has with the ticket data.
ASIMOV MACHINE LEARNING PLATFORM
Customer
Support Solution
Retail-Specific
Customer
Support Solution
Healthcare-Specific
Customer
Support Solution
Prediction
Engine
Insight
Extractor
Recommendation
Engine
Figure 5
ASIMOV’s Technological Anatomy
8. 8 KEEP CHALLENGING December 2016
Looking Forward
ASIMOV’s machine learning solutions support big data ecosystems and cloud-based
environments to address a multiplicity of customer care challenges for call center
optimization. ASIMOV also provides accelerators to bring machine learning-based
solutions more quickly to market.
ASIMOV has been successfully piloted for a communication services provider, and
has produced numerous insights for resolving customer support challenges. This
has helped the company optimize customer complaint troubleshooting.
With its deep understanding of customer care support problems and customized
machine learning support software, ASIMOV can efficiently address critical business
goals that challenge customer support centers, including the following:
• Make more intelligent decisions to reduce time spent on customer support
calls. ASIMOV can suggest solutions for resolution, based on the initial diagnosis
of troubleshooting. As a result, it can speed problem resolution for agents and
customers anywhere from five to 15 minutes, on average. Agent time is directly
proportional to support center expenses, and reducing resolution time across
millions of customer calls will save substantial spending in support.
• Provide recommendations on troubleshooting approaches and solutions.
ASIMOV creates a real-time social navigation system for agents to use when re-
solving issues. Agents can learn which solutions worked well for others and rec-
ommend those fixes when resolving similar issues.
• Create smart agents backed by machine learning and increase agent effi-
ciency. With the decision-making intelligence of ASIMOV, customer care centers
can reduce the dependency on human intelligence and the impact of lost knowl-
edge due to agent attrition.
Figure 6
Applying Clairvoyance to Solve Customer Challenges
ASMIOV provides
personalized
recommendation for
addressing the problem
RECOMMENDATIONS BASED ON CONTINUOUS LEARNING
SYMPTOMS-BASED ISSUES
• Type 1 : 1,000 issues were due to cell tower signal
booster failure
• Type 2 : 700 issues due to OS update problem in
specific mobile platform
• Type 3 : 200 customers had data cap limits exceeded as
per plan
ISSUE-BASED ACTIONS
• For Type 1 : 70% of the issues will be resolved by
scheduling repair on booster devices
• For Type 2 : 90% of issues can be solved by sending
carrier settings update
• For Type 3 : 100% of issue can be solved by
upgrading plans
GUIDE RECOMMENDATION
For Type 1 : Refer to guide <link> - 85% success ratio
For Type 2 : Refer to guide <link> - 78% success ratio
For Type 3 : Refer to guide <link> - 94% success ratio
SYMPTOMS ANALYSIS
1,900 customers had the same symptom in the
past 10 days within the same geographical area
AgentCustomers
Send customer
symptoms and
diagnostics to
ASIMOV
Complaint:
High-speed
wireless
connectivity is
not available
CASE ILLUSTRATION:
Customer support for a wireless telecom provider
ASIMOV
9. OPTIMIZING CUSTOMER SUPPORT WITH MACHINE THINKING 9
About the Authors
Aravindakumar Venugopalan is a Senior Architect in Cognizant’s HPC Labs and
is involved in a wide range of research and development activities in the areas of
data science, HPC, big data and cloud computing. In his 15 years of experience in
the IT industry, he has created assets that have been showcased in supercomput-
ing seminars (SC12) and won Cognizant innovation awards. Aravindakumar holds a
master’s of science in software engineering. He can be reached at Aravindakumar.
Venugopalan@cognizant.com | https://www.linkedin.com/in/aravindakumar-venu-
gopalan-95a23011.
Sivasubramaniam Renganathan is an Architect in Cognizant’s HPC Labs and has 11
years of experience in consulting and software development. He has worked on a
wide range of projects involving Microsoft.Net, HPC, cloud and big data technolo-
gies for various Fortune 500 clients. Sivasubramaniam holds a bachelor’s degree in
computer engineering from Anna University, and his prime areas of interest include
big data, machine learning and predictive analytics. He can be reached at Sivasub-
ramaniam.Renganathan@cognizant.com | https://www.linkedin.com/in/sivarenga-
nathan.
Rajarajan Gandhi is a Manager in Cognizant’s QE&A business unit and has over 12
years of experience in the telecommunications domain. He has worked on multiple
networking, server management and OSS/BSS solutions, as well as Agile projects
for various telecommunications clients. Rajarajan holds a master’s degree in
software engineering. He can be reached at Rajarajan.g@cognizant.com | https://
www.linkedin.com/in/rajarajan-gandhi-01848754.
Acknowledgments
The authors would like to thank the following associates for their contributions to
this paper: Senthil Ramaswamy Sankarasubramanian, Senior Director – Technology,
HPC Labs, Cognizant, and Srinath Narayanan, Senior Manager – Quality Engineer-
ing & Assurance, Cognizant.
• Automatically resolve customer issues through highly confident decisions
from machine learning solutions. ASIMOV can be used to automatically predict
solutions and provide resolutions through interactive voice response (IVR) inter-
actions before redirecting customers to a human agent.
Note: All company names, trade names, trademarks, trade dress, designs/logos,
copyrights, images and products referenced in this white paper are the property of
their respective owners. No company referenced in this white paper sponsored this
white paper or the contents thereof.
Footnotes
1 Wikipedia page on machine learning: https://en.wikipedia.org/wiki/Machine_
learning.