Download Presentation - Intelligent Business Systems

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Nonlinear dimensionality reduction wikipedia , lookup

Transcript
CSI , CIO CLUB
Software Gets Smarter
Artificial Intelligence in the Enterprise
Dr Kaustubh Chokshi
CEO, Intelligent Business Systems
Cutting-edge Artificial Intelligence techniques are ensuring that enterprise
software today can dynamically adapt to rapidly changing business
environments and provide realistic decision support and trend
forecasting for enterprise managers to act on
*******
AI-based software dynamically learns from experience and adapts to the
environment, as it is equipped to acquire knowledge from the data
generated within the organization, as well as from expert opinion and
external data sources
•
•
•
•
Artificial Neural Network
Bayesian Statistics
Genetic Algorithm
Artificial Immune System
•
Artificial Neural Network (ANN)
•
Process info in a manner similar to biological nervous systems such
as the human brain.
•
Large number of highly interconnected processing elements.
•
ANNs learn by example, much like humans.
•
An ANN is initially “trained” with large amounts of data and rules
about data relationships.
•
Once trained, a neural network becomes an “expert” in its specific
area of operation and can offer projections and trends based on past data
and answer “what if” queries.
•
Artificial Neural Network (cont.)
• ANNs operate using several different techniques, including gradientbased training, fuzzy logic, genetic algorithms and Bayesian methods.
• ANNs can derive meaning from complex or imprecise data, thus
recognising patterns and determining trends that humans or other
computer techniques would most likely fail to notice.
• Neural networks can also learn temporal (time-sensitive) concepts and
can thus be used in signal processing and time series analysis.
•
Bayesian Statistics
• Bayesian Statistics enables calculation of the probability of a new event
on the basis of earlier probability estimates of an event or events in the
past, derived from existing empiric data.
• According to Bayesian logic, the only way to quantify a situation with an
uncertain outcome is through determining its probability.
• Thus, the knowledge of prior events is used to predict future events.
• Bayesian Statistics (cont.)
• This is an iterative or learning process and is the preferred
method for designing software that learns from experience.
• Pattern recognition is based on Bayesian inference, and this
forms the basis of varied applications such as spam detection,
fraud detection, intelligent search, unstructured text mining,
etc.
•
Genetic Algorithm
• A genetic algorithm (GA) is an algorithmic model that tests a set of results,
each represented by a string, and selects the best fit among them.
• In this methodology, of a number of possible programs or functions within
a program, only the most effective survive and compete or cross-breed with
other programs, with the intention of evolving into an ever-better solution
to a particular problem.
• Used to find approximate solutions to difficult-to-solve problems.
•
Artificial Immune System (AIS)
• Like other biologically inspired techniques, AIS tries to extract ideas
from a natural system, in particular the vertebrate immune system, in
order to develop computational tools for solving engineering problems.
• Used for pattern recognition, data analysis, data clustering, function
approximation and optimisation.
•
Linear
• "linear" in its scientific sense; that is to say, as implying parallelism
between the magnitude of a cause and the magnitude of its effect.
•
•
•
•
One of the characteristics of
today’s business environment is
that it is definitely non-linear.
Non-linear
systems
exhibit
unpredictable but non-random
cause-and-effect relationships.
Edward Lorenz' "butterfly effect"
provided an illustration of this
behaviour by postulating that a
butterfly flapping its wings in
Brazil might cause a tornado in
Kansas.
In complex systems, even a very
small change in initial conditions
can rapidly lead to changes in the
behaviour of the system that
appear counter-intuitive in both
nature and magnitude.
•
•
•
Decision Management using AI is a systematic approach to automating and
improving decisions across the enterprise.
Businesses using AI gain much greater control over the results from highvolume operational decisions.
AI-based Decision Support Systems aim to increase the precision,
consistency and agility of these decisions while reducing the time taken to
decide and the cost of the decision.
•
•
These decisions are typically those that an organisation uses to manage its
interactions with customers, employees and suppliers.
Computerisation has changed the way you approach decision-making by
enabling decisions to be based on historical data, prior decisions and their
outcomes, corporate policies and regulations.
Data sampling
Data cleaning
Analysis
Decision
making
Data completeness
Bayesian
Network
Customer
Classification
Data normalization
Tourism
Dataset
Neural
Network
Data Division
Prediction
•
•
•
•
•
Operational business decisions - those taken in large volume, every day.
They are differentiated from "strategic" decisions such as where to open a new
store or when to drop a product line that are rarely the same twice and that
simply do not happen that often.
Clearly these are important, needs to be automated and make them in "realtime".
Many examples, such as approve/decline, next-best-offer to make a customer,
authorisation of a sale, fraud detection in a claim, account application
processing, etc.
These "tactical" decisions determine the way in which you will manage
processes and customers such as decisions about which segments of a customer
base will receive which precise offer.
•
•
•
•
•
•
•
Alerts - alerting the user to a decision-making opportunity or challenge.
Problem recognition - identifying problems that need to be solved as part of
the decision-making process.
Problem solving - providing and evaluating alternative and/or
complementary solutions.
Facilitating/extending the processing of knowledge - overcoming some of the
human limitations of the speed and volume of information that can be
processed (e.g. acquisition, transformation, exploration).
Stimulation - stimulating the human perception, imagination, or creating
insight.
Coordinating/facilitating interactions - in multi-participant decision making.
Various other stages and activities in the decision-making process.
• Is a concept developed by Richard Hackathorn.
• Decision Latency is the time it takes to receive an alert, review
the analysis, decide what action is required, if any, based on
knowledge of the business, and take action.
• Operational decisions require very low decision latency.
• Overall Aim of a Decision Management System is to try and
reduce Decision Latency.
•
Precision. Increase revenues and improve risk management through greater
segmentation, more relevant offers and better risk management. Benefits
include:
– Higher revenue yield per customer interaction, through better targeting
and segmentation and through more timely responses to customers
– Lower losses from fraud and bad debt, through using analytics to
improve risk management
– Lower costs through refined targeting, such as eliminating "off
target" marketing messages or prospects that are unlikely to buy
Predicted
Noise
Years
Actual
1997
1996
1995
1994
1993
1992
1991
1990
1989
1988
1987
1986
1985
1984
1983
1982
1981
1980
1979
Arrivals
300,000
250,000
200,000
150,000
100,000
50,000
0
•
Consistency. Ensure that all decisions meet your rules, policies and
regulations, by automating 75% or more of your operational decisions.
Benefits include:
– Lower costs of making decisions through automation, reducing the
number of people and streamlining the processes needed to make or
process a decision.
– Lower costs of compliance, regulatory requirements, through
centralised and easy to update business rules management.
– Faster decisions that operate at the speed of the transaction, lowering
hand-off costs between systems and between people.
•
Agility. Meet new competitive and compliance demands by rapidly
changing your business rules, and instantly executing new strategies.
Benefits include:
– Improved strategic alignment, greater competitiveness through faster
response to market changes and regulatory demands.
– Greater return on new product & market opportunities, through faster
time to implement and change decision-based processes, change
approaches to the market.
•
Increase revenue and profitability by improving the consistency, relevance,
speed and precision of customer decisions, getting more value from every
customer interaction.
•
Improve customer relationships and retention through more targeted offers,
faster response to service requests and more consistent treatment.
•
Minimise losses through the use of analytics for more accurate and consistent
risk assessment and fraud detection.
•
Gain competitive advantage by being more nimble than the competition -- get
new strategic initiatives, products, campaigns and pricing to market faster and
with greater precision and consistency.
•
Ensure and demonstrate rigorous compliance with corporate and regulatory
policies.
•
Reduce the costs associated with decision-based processes, while
improving decision consistency, speed and quality.
•
Leverage existing investments in data warehousing and CRM—derive
more value from all corporate and external data sources.
•
Reduce ongoing maintenance costs required to change / tune models,
rules or strategies that are in production.
•
AI can be applied to virtually any business area that involves
high-volume, operational decisions, or the use of analytics and
business rules to improve decision strategies.
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
Customer acquisition and retention
Matching prospects and customers with product/service offerings
Core customer decisions-underwriting, channel selection, credit, pricing, etc.
Forms and data management
Work process control
Fraud detection
Claims management
Guidance and employee support
Customer response and service
Debt collection and recovery
Agency management
Network integrity assurance
Online recommendations
Product configuration and design
Regulatory compliance
•
The supply chain of an Enterprise includes the network of all suppliers and
activities involved in the process of transforming the requisite raw
materials into finished goods and delivering them to customers.
•
The objective of supply chain management (SCM) is the integration and
optimisation of all the components and processes involved.
•
The purpose of AI here is to evaluate the different options of transporting the
different products to different terminals based on demand forecasts and forward
prices so that the margins are maximised.
•
The economic optimisation is accomplished by using the Linear Programming
technique. The margins calculated include sales revenues as well as the costs of
purchase, production, inventory holding, transportation and materials handling.
•
AI plays a key role in Distribution Network by not only making the best use of
capacities in the system (asset utilisation), but also ensures that all forecast
demands are met (prioritises the distribution to make the most economic sense).
•
Product Customer Fit: which means marketing a product to a customer who most
desires/needs/uses it.
•
Demographics Profiling: Information such as age, gender, household size and
parental status offer the most basic understanding of who the customer is.
•
Geo-Demographic Profiling: Location and type of area in which people live.
•
Psychographic Profiling: Attitudes, Values, Motivations and Aspirations.
•
Customer Loyalty: Value, Frequency and recurrence of purchases.
•
Behaviour Profiling: Profiling based on Consumer Behaviour focuses on issues key
to anyone seeking to sell or market products: what do your users buy? What are they
in-market for? How much do they spend, and where do they spend it?
• AI can help find answers to key questions:
– How can we know which specific marketing actions to take, based on
purchase behaviour and personal profile information, to maximise value?
– How can we mine the data to derive actionable insights into customer
segments and response patterns?
– How can we spend our marketing efforts more effectively, and minimise
waste?
– How can we create the messages and offers that are most likely to elicit a
favourable response without doing expensive in-market testing?
– How can we establish an ongoing dialogue and a deeper level of intimacy
with our best customers?
– How can we measure the sales and profit resulting from our investments in
data-driven marketing programs?
– How can we integrate customer data from all databases, channels and
touch points to create a 360-degree view of our customer relationships?
– How can we determine which customers account for the vast majority of
our profits (and future profit potential) and how can we map their “genetic
makeup” so that we can then market to others just like them?
•
Behaviour based profiling, behaviour, and for anybody concerned about what their
customers are doing
•
What AI provides is the ability to market your products to the right customer depending
on his buying patterns.
•
AI-based system can also encode expert opinion, for instance, what kind of red wine will
go with what kind of blue cheese.
•
This helps to promote other lines of needed products to relevant customers. Also, the
system has the ability to “learn” from data about customers’ preferences with respect to
wine and blue cheese.
•
Profiling using AI can then bring objective information to the marketing department.
This can be used to reduce the cost of the campaigns by selecting only the prospects that
have a high probability to reply positively, or they can be exploited for fraud detection.
•
In a nutshell, AI-based customer profiling and advertising integrates AI/CI platforms
needed to perform behaviour analysis, context-sensitive acquisition, cross-selling and
retention programs, enabling multi-product, multi-channel companies to drive more
efficient and profitable customer interactions.
•
•
•
•
•
Detect every single event fraud as it occurs.
Detect fraud trends more quickly.
Minimise the cost of manual labour.
Convert more valid orders easily.
Minimise the cost of customer service inquiry resulting from valid
order rejection.
• Control fraud risk tolerance.
• Detect known and unknown ways of frauds.
External Rules
Experts Inferences
Past Experiences
External Data
Merchant Analysis
Card Holder Profile
Time Series Data Analysis
(NN):
Numeric
Structure Text
Unstructured Text
Known Behviour
Detection
Output:
Risk Scores
Explanations
Decisions
Unknown Behviour
Detection
Transaction Data
•
Data Mining explores
•
Data mining applies sophisticated
mathematics and/or AI to data in
order to search for useful patterns
in large data sets.
•
Data mining is often one stage in
developing an AI-based system.
•
AI’s analytical
"What next?”
•
AI explores and uses data patterns
to
make
forward-looking
predictions, or to make complex
statements about customers by
evaluating multiple data patterns.
•
AI uses the patterns those represent
in the enterprise, in order to
"formalise" the relationships and
predict future behavior consistently.
power
answers
Importance of Variables for Model Prediction
7
6
Importance
5
4
3
2
1
0
Variables
POLINST
CPI
Exc. Rate
Income
Price of Oil
6.39743
2.81577
2.95374
2.85635
3.11219
Variables
•
BI delivers insight, predictive analytics delivers action
•
Traditional business intelligence (BI) tools extract relevant data in a structured way,
aggregate it and present it in formats such as dashboards and reports.
•
BI helps businesses understand business performance and trends.
•
BI focuses on past performance, predictive analytics forecasts behaviours and results
in order to guide specific decisions.
•
BI suites now include some analytics.
•
However, BI analytics almost always aggregates past customer data in a collective
sense - for example, how many of my customers are in a particular set so I can
forecast product sales by quarter?
•
BI tools are more exploratory than action-oriented.
•
Exploration is more likely driven by a business user than an analyst.
•
AI can help BI to focus on
– past performance
– predictive analytics
– forecasting behaviour
– Results/scenarios in order to guide specific decisions.
•
If BI tells you what’s happened, AI tells you what to do.
•
AI in BI is important in order to make better business decisions.
And finally…
About Intelligent Business Systems
Intelligent Business Systems (IBS) provides innovative business solutions
incorporating cutting-edge Artificial Intelligence (AI) engines. The company’s
core competence lies in Artificial Intelligence, with a significant focus on
Business Intelligence. IBS also has expertise in Robotics and
Bioinformatics.
IBS has a very strong research focus in everything it does
Established in the UK in 2003, IBS has just expanded into India in a very big
way
Check out the interactive presentation on the CD to learn more!
www.intelligentsystems.biz
Thank You!
Dr Kaustubh Chokshi
[email protected]
CSI , CIO CLUB