Intelligence into
Business Value.
Today.
2
Turning Artificial Intelligence
into Business Value. Today.
By Cyrille Bataller and Jeanne Harris
3
Introduction
4
Figure 1: Examples of how Artificial Intelligence technologies can be integrated into compelling business solutions
Audio Identifying, recognizing and analyzing sounds Speech recognition technologies integrated into call centers
processing and speech automate the identification of callers.
Sensor Processing and analyzing information from In an agricultural setting, sensors in the field can be integrated
processing sensors other than cameras and microphones with software to deliver “precision agriculture”— sensing and
communicating status about temperature, humidity, etc., enabling
more precise care for crops.
Comprehend
Natural language Understanding and generating language in Personal assistants on consumer smart phones provide guidance
processing spoken and/or written form and services using natural language. Increasingly, search
capabilities include the ability to understand the meaning of
what a person is saying, not just recognizing key words or doing
statistical retrieval.
Knowledge Depicting and communicating knowledge to Knowledge-based tools provide the capability to link a particular
representation facilitate inference and decision making search or piece of content to other relevant content on the web.
This is done by tagging all content and then mapping it to a larger
representation of knowledge. For example, a search for “Da Vinci”
will link one to particular paintings and creations, as well as to
Italy, to the Renaissance, and so forth.
Act
Inference Deriving answers from a static knowledge base Solutions can apply rules to make automated loan approval or credit
engines such as business rules decisions, or granting of visas. Such capabilities can deliver accurate
decisions in a fraction of the time of manual decision making.
Expert Emulating the decision-making ability of a Medical diagnostics as well as legal research can be significantly
systems human expert and solving complex problems aided by the ability of expert systems to sift through millions of
by reasoning with the information available in data sources, synthesize information and present it to a user.
its knowledge base
Machine Altering the decision process based on experience Software tools and personal agents can learn from users to
learning improve productivity—for example, by sorting email, then
extracting calendar entries and action items.
5
Artificial Intelligence:
Systems that can sense, comprehend and act
Artificial Intelligence technologies such as multispectral
image analysis (scanning passports using
The system (2) asks follow-up questions
and (3) stores facts in working memory,
systems can be self-learning. infrared and ultraviolet light), extensive then (4) takes the facts of the case and the
information databases and matching knowledge it has of medicine and cases
They are like bright students algorithms, an integrated solution here can (stored in a knowledge base) and infers a
who are given educational improve security by identifying people on solution or treatment. Finally, (5) the system
unauthorized entry lists or others posing presents a conclusion or suggestion to the
materials and then can learn a risk.4 Video analytics is another sensing doctor, who uses it as expert input into a
by themselves. technology that can automate observation
and incident detection by video surveillance
final diagnosis and treatment plan.
Learn
6
For example, self-learning Artificial At the consumer level, an example of a Technology advancements are continuously
Intelligence solutions are already in use by self-learning application is Google Now, improving the sophistication of computers’
banks to detect credit fraud. The banks use a personal assistant. It learns from its learning capabilities. A form of machine
machine learning models that understand user’s activities and interactions to find, learning called “deep learning” is being
previous spending patterns of a customer and collect and present relevant, personalized used today to develop systems that can
predict the transactions a person will make, information on a smartphone. Google Now perform more complex activities using
flagging unusual activity. These systems are is also designed to constantly improve neural networks to mimic the human brain
also given real examples of fraudulent and based on user feedback and its own structure. Such systems are increasingly
legitimate transactions so that the models learning methods. able to do things like recognize patterns
can learn to recognize new patterns and and objects, and generate descriptions of
evolve as fraudsters alter their tactics. images in natural language.
Figure 3: A medical diagnostic expert system comprehends input from a doctor, searches a knowledge base,
infers solutions and suggests conclusions or treatments
1
The user asks the
system a question
User Interface
The means by which the system interacts with
the user. Can be through a website, a graphical
user interface (GUI), speech recognition, etc. The system generates an
explanation for the solution
and relays it to the user.
2
4
The system gathers facts and The Inference Engine takes the facts
determines what follow-up from the Working Memory and all
Inference Engine
questions to ask the user. relevant rules from the Knowledge
The “brain” of the system: it leverages the
Base. It then infers a solution
information contained in both the Knowledge Base
and the Working Memory to draw conclusions.
The facts are stored in
the Working Memory
7
Realizing business value: A framework for
identifying Artificial Intelligence opportunities
One way to classify and explain the range On the other hand, many other activities
of available Artificial Intelligence solutions do not (and perhaps will not ever) lend
is to think of them in terms of both themselves to automation, so it is better to
automating and augmenting. think in terms of augmentation. As in the
medical example cited earlier, expert systems
Automation of routine tasks can improve are better than humans at sifting through
overall productivity and performance. In gigabytes of data related to a particular
financial services, for example, many routine medical condition. However, many patients
policy decisions—such as determining in a healthcare setting would be reluctant to
whether someone qualifies for a loan or for receive diagnosis, prescription and treatment
insurance benefits—can be time consuming from a machine. Similarly, smart machines
and technical for humans. Automated are already augmenting the work of
systems can make these decisions quickly, surgeons and fighter pilots, but it would be
accurately and consistently. inappropriate or even impossible to automate
that kind of work—where real-time human
analysis of a situation and the application of
judgment are essential.
8
Figure 4: By sorting tasks into four activity models, particular types of Artificial
Intelligence solutions can be identified
Augment
Effectiveness Model Innovation Model
Unstructured, Support seamless integration and collaboration Enable creativity and ideation
Volatile,
High-Volume • Wide range of interconnected work activities • Original, innovative work
• Highly reliant on coordination and • Highly reliant on deep expertise,
communication experimentation, exploration and creativity
• Example solutions: Virtual agents for • Example solutions: Support for biomedical
consumers or for enterprise customer service; research; fashion design; music writing
collaboration or workflow management
Data
complexity Efficiency Model Expert Model
Provide consistent, low-cost performance Leverage specialized expertise
Automate
Routine, Ad Hoc,
Predictable, Work complexity Unpredictable,
Rules-based Judgment-based
9
Expert model Effectiveness model Innovation model
In this model, work is more likely to As its name suggests, the goal of the In this model, Artificial Intelligence
involve judgment and is highly reliant effectiveness model is to improve the solutions enhance creativity and ideation
on individual expertise and experience— overall ability of workers and companies by humans—activities and roles such as
activities performed, for example, by to produce a particular desired result. biomedical researchers, fashion designers,
doctors, lawyers, financial advisors and This class of workers typically requires chefs, musicians and entrepreneurs.
engineers. Decision-making and action is considerable knowledge of their industry, Humans make decisions and act, while
generally taken by humans themselves, company and business processes. Their technology helps identify alternatives and
while technology’s role is to augment success is highly reliant on coordination optimize recommendations.
human sensing and decision making— and communication and involves a wide
enabling analysis and offering advice and range of interconnected activities—work An example of how Artificial Intelligence
implementation support. such as administration, managing, sales and can augment creative behavior is an
so forth. In these solutions, technology acts intelligent software program from
So-called “expert systems” are examples of as personal assistant or agent on behalf Platinum Blue Music Intelligence. The
Artificial Intelligence solutions within the of humans at their direction; humans use program can analyze a song and provide
expert model. Such systems are capable of the cognitive tools to assist in scheduling, recommendations to increase the likelihood
searching vast data stores, and then making communicating, monitoring and executing that the song will be a hit (by fine-tuning
inferences and recommendations based on activities. Artificial Intelligence helps the bass line, for example).
that knowledge. An important distinction workers be better at what they do.
among these expert solutions, however, Billed as a tool to assist artists and
is whether the system is functioning as a Virtual agents, whether consumer or producers, Platinum Blue’s technology
primarily autonomous system—for example, corporate, are excellent examples of uses spectral sound-wave analysis to offer
“given your travel parameters, here are some Artificial Intelligence solutions within advice. For example, the tool was used to
itinerary recommendations”—or whether it is the effectiveness model. Consumers analyze the song “Crazy,” by Gnarls Barkley.
providing input to human judgment. are increasingly using agents such as The analysis found that “Crazy” belonged
Siri, Cortana and Google Now on their to the same hit cluster as several recent
The medical diagnostic system discussed smartphones. Virtual agents receive hit songs as well as older hits by Olivia
earlier is an example of an advisory system requests in a textual or verbal form, Newton-John and Mariah Carey. The data
that augments human understanding process them using technologies such as indicated that “Crazy” was going to be a
and judgment. Given the nature and natural language processing or speech huge success, which it was.6
importance of doctor-patient relationships, recognition, search their knowledge
medical doctors themselves would most repositories or the Internet, formulate
likely want to interpret the results of the hypotheses and then give the answers
expert system’s analysis and present a back to the users, again in text or speech.
personalized diagnosis to the patient. In
other cases (such as a car configurator At the corporate level, virtual agents are
on a manufacturer’s website), a personal beginning to handle many of the routine
relationship may be less important, so the questions that come into a customer
expert system might interact directly with service center. Only when questions
a user in natural language, without the are not resolvable from an automated
need for a human intermediary exercising knowledge base would a user be referred
interpretation and judgment. to a live agent.
10
11
Understanding the potential for Artificial
Intelligence solutions: Industry examples
As demonstrated in these figures, executives from different industries can use the
Artificial Intelligence framework to plot different work activities within the matrix. By
understanding the complexity of different work activities, as well as the complexity of the
data and information used for those activities, particular Artificial Intelligence solutions
can be planned, designed and implemented.
For purposes of illustration, we demonstrate how the framework could be used in the
healthcare and banking industries.
Healthcare example
Augment
Effectiveness Model Innovation Model
Unstructured, Support seamless integration and collaboration Enable creativity and ideation
Volatile,
High-Volume Biomedical research
Patient flow management
Clinical workflow management Medical ethics rulings
Nursing handoffs e-Consulting
Operating room scheduling
Data
complexity Efficiency Model Expert Model
Provide consistent, low-cost performance Leverage specialized expertise
Robotics-aided surgery
Physician authorization requests Medical specialist diagnosis
Structured, Ordering labs Prescription advising
Stable, e-Prescribing
Low-Volume Appointment scheduling
Simple medical diagnosis
Automate
Routine, Ad Hoc,
Predictable, Work complexity Unpredictable,
Rules-based Judgment-based
12
Banking example
Augment
Effectiveness Model Innovation Model
Unstructured, Support seamless integration and collaboration Enable creativity and ideation
Volatile,
High-Volume
Account management New-product creation
Branch management
Marketing campaigns
Automate
Routine, Ad Hoc,
Predictable, Work complexity Unpredictable,
Rules-based Judgment-based
In banking, transactions are being driven toward greater efficiency through automation,
while greater effectiveness through Artificial Intelligence is the goal for activities such
as security and identity management. For work such as financial advising and retirement
planning, human interaction is essential; nevertheless such activities can be aided by
expert systems, automating some tasks and augmenting others.
13
Artificial Intelligence challenges
14
Conclusion
15
References About the authors About Accenture
1
http://blogs.cisco.com/news/cisco-to-boost- Cyrille Bataller is a managing director, and Accenture is a leading global professional
network-security-platform-with-acquisition- the domain lead for Intelligent Computing, services company, providing a broad range of
of-cognitive-security/ with the Emerging Technology group at services and solutions in strategy, consulting,
Accenture. He leads Accenture’s exploration digital, technology and operations. Combining
2
http://readwrite.com/2014/01/29/google- of the Artificial Intelligence field. Previously, unmatched experience and specialized skills
artificial-intelligence-robots-cognitive- he directed biometric identity R&D activities across more than 40 industries and all business
computing-moneyball within Accenture Technology Labs. functions—underpinned by the world’s largest
delivery network—Accenture works at the
3
http://www.xconomy.com/san- Jeanne Harris is the former managing intersection of business and technology to help
francisco/2010/06/14/the-story-of-siri-from- director of information technology research clients improve their performance and create
birth-at-sri-to-acquisition-by-apple-virtual- at the Accenture Institute for High sustainable value for their stakeholders. With
personal-assistants-go-mobile/ Performance. She led the Institute’s global approximately 373,000 people serving clients
research agenda in the areas of information, in more than 120 countries, Accenture drives
4
See an example of how e-Passport gates are technology and analytics. Jeanne is the co- innovation to improve the way the world works
using biometrics to speed up the traveler’s author of the business bestsellers “Competing and lives. Visit us at www.accenture.com.
experience: http://www.accenture.com/us-en/ on Analytics” and “Analytics at Work.”
Pages/success-london-heathrow-airport- Jeanne is currently on the faculty at Columbia
automated-border-clearance.aspx
About the Accenture Institute for
University. She is based in Chicago.
High Performance
5
For more on video analytics, see: http://
www.accenture.com/us-en/Pages/service-
Acknowledgments The Accenture Institute for High Performance
develops and publishes practical insights
video-analytics.aspx The authors wish to thank the following into critical management issues and global
people for their assistance in the research economic trends. Its worldwide team of
6
“What People Want (and How to Predict It),” and preparation of this report: Adam Burden, researchers connects with Accenture’s
by Thomas H. Davenport and Jeanne G. Harris, Maysun M. Hasan, David Light, Craig consulting, technology and outsourcing
MIT Sloan Management Review, Winter 2009. Mindrum, G. Scott Phillips and leaders to demonstrate how organizations
Abigail L. Terrill become and remain high performers through
7
This is sometimes referred to as the
“uncanny valley.” See http://spectrum. original, rigorous research and analysis. For
For more information more information, please visit
ieee.org/automaton/robotics/humanoids/
an-uncanny-mind-masahiro-mori-on-the- Contact cyrille.bataller@accenture.com http://www.accenture.com/institute.
uncanny-valley.
About Accenture Emerging
8
Pew Research Center, August 2014, “AI, Technology
Robotics, and the Future of Jobs.”
Accenture’s Emerging Technology
9
“FT interview with Google co-founder and group is responsible for incubating
CEO Larry Page,” by Richard Waters, Financial the next generation of Accenture’s
Times, October 31, 2014. technology services. By partnering with
Accenture Technology Labs, Emerging
Technology Delivery and the Technology
Industry Ecosystem, it provides
expert services to identify, pursue
and deliver advanced and emerging
technologies with pioneering clients.