Commonality of Problem Structure

We look at several systems for delivering complex services over the web, and compare the components of Tupai's technology needed to build them.

Wealth Management for individuals
Reinsurance for businesses
Medical Diagnosis

Each boils down to - Know the client, know the products, match the need with what's available.

In a little more detail:

Wealth Management

A leading investment bank has decided to leverage its 100 years of wealth management expertise and world-renowned brand to launch a major effort to reach a much broader group of affluent individuals in the United States. This market has doubled in the last five years and now exceeds seven million households.

Research has shown that the service that clients seek most is a trusted advisor who integrates a full range of investment, tax planning, brokerage, banking, and generational planning services. The firm intends to deliver these through a choice of channels, including the Internet and telephone contact as well as face-to-face interaction with private bankers.

In his Letter to Shareholders, the bank’s Chairman declares:

"If we could have imagined a world in which the value of our brand would rise, this is it. Our role as trusted partner to our clients, providing unbiased advice with a global perspective and solving the most complex problems by integrating knowledge and service of the highest quality, is more relevant than ever. Now, technology enables us to take elements of this approach on line, allowing us to aggregate sophisticated information and provide integrated services through customized, single points of access."

The bank is intent on exceeding its clients’ expectations by bringing online the same level of subtle sophistication that has been its hallmark for generations. The bank realizes that the planned system must deal with a dynamic environment in which client, market, and bank are continuously evolving and changing. Consequently, in its interaction with the market and its clients, the bank must continuously adjust its actions based on the client’s current needs, the bank’s current objectives, and available alternatives.

In providing wealth management advice, many competing influences must be taken into account. Many of the decisions going into the advice will never be verbalized since the client expects the adviser to know what to do without having to ask unsettling questions. Realizing that the rules of the game are continuously in flux, the bank wishes to implement a system that will continuously monitor the market’s and the client’s behavior, integrate its findings with the bank’s practices and policies, identify possible alternatives, and project them according to their ability to satisfy the client (and perhaps advance the bank’s agenda as well).

Solution Overview

In the following description we focus on the elements involved in building a system for wealth management. Our intent here is to highlight the fact that providing wealth-management advice requires the integration of knowledge from multiple areas and multiple types. As we mentioned earlier, this ability to integrate the knowledge process is a cornerstone of Tupai’s technology.

eCognition addresses the fluid interaction of client needs, market behavior, and current portfolio holdings by constructing a representation (or model) of the situation at hand, and then performing situational analysis on it. In conducting its analysis, the wealth management system is able to research dealings with people in similar positions in the past, and use the knowledge to guide its actions. The bank then decides on the makeup of the portfolio, and under some circumstances, may suggest other actions, such as recommending a change in the client’s drawdown rate.

An eCognition model for wealth management comprises of a number of interacting components – the client, their portfolio, and the market. Each of these components is modeled initially, as well as the effects of their interactions.

Model Components

I. The Client

The client is a complex amalgam of psychology and financial need. Some parts we can analyze precisely, some parts we can analyze using probability, some parts we can only estimate using influences.

Some of the elements of the client’s life position:

Familiarity with wealth
Lifestyle
Age
Dependants
Societal obligation
Self generation
Tax position
Drawdown rate

Familiarity with Wealth

Some investors build up a fortune from nothing and know what it cost them, others inherited and know no other life, still others have became millionaires overnight from some dot.com. Each of these people will have a very different outlook on how their money should be invested:

wisely for the future,
as long as it pays the bills and covers inflation, or
who cares, I can do it again.

Some clients will have little knowledge of investments, leaving even the decision of what to do to an attorney, while others will want chapter and verse on each proposed investment. A high level of volatility in their wealth will horrify some clients, while others will be happy to ride a roller coaster in order to maximize gains.

In determining what decisions the system should make on behalf of the client, he system must determine the client’s desired level of involvement in the decision making process and their level of sophistication.

Lifestyle

Some clients will be modest and reclusive, some will have a giddy lifestyle, and some will literally not know how to spend their wearth, being too busy making more. Taking over a poorly performing portfolio and greatly increasing the returns in the good times may lead to an increase in lifestyle which is unsupportable in bad times.

The wealth effect from rapid stock market gains over the last few years, may have ratcheted up the client’s lifestyle, or led them to acquire overpriced non-income bearing assets. The client may be loath to change their lifestyle, even as speculative gains disappear. Here is a difficult choice for the bank to make – propose a reduction in the budgeted rate of outflow and lose their business, or keep the outflow and expect to eat into the assets, or increase the risk level on the liquid investments while leaving the illiquid investments in reserve – investments which may be near worthless in a downturn.

Age

Age usually carries with it a decreasing appetite for investment risk, almost inversely with the risk the portfolio could tolerate. Age can also carry with it a required change in who the portfolio should be managed for, from the client to the client’s intended heirs.

Dependants

The number of dependants, their positions in their life cycles, and any special considerations – education, travel, or wedding plans – leads to the need for modeling of the cashflows of dependants and potential future dependants.

Societal Obligation

Old money often puts something back into society by way of college endowments or gifts to museums. Some clients will want every drop of publicity from a high profile bequest, while others may prefer to be anonymous, modulating the investment strategy deliberately into under-performing investments to achieve a similar beneficial result with no publicity.

New money may feel the need to be in hot new technologies, support green issues, or give new entrepreneurs a helping hand through Venture Capital trusts. In providing wealth management advise, the system has to anticipate these requirements as part of the service.

Tastes for societal obligation change over time, and the wealth management package builder needs to be just as sensitive to how the client feels about their wealth as to how much the wealth can grow over time.

An eye needs to be kept on potential publicity exposures - an investment in a company using Asian sweatshops may be good financially, but extremely bad in terms of the negative publicity it could generate for a client with a high public profile.

Self Generation

The client may be on a high salary, or have a large options package from the current employer. The bank may need to estimate the continuance of the salary or the viability of the options, based on its understanding of market conditions.

The rate at which the client is able to generate wealth by their own efforts will change the risk appetite for the wealth management package – not just that a higher risk can be tolerated, but it becomes necessary to keep the business by showing that the assets are performing adequately in the shadow of the client’s own wealth generation.

Tax Position

Wealthy clients will usually have a complex tax position, which will require significant modeling in its own right. The combination of the tax position and likely variation in drawdown rate over time may require considerable shaping of the income stream.

Drawdown Rate

We need to assemble a model of the client’s particular financial circumstances – their other income producing assets, their rate of expenditure, their future liabilities, their tax position. The model is automatically assembled from components that interact with each other – an existing dependent child suggests a heavy education cost in the future.

The model needs to be run into the future, against the potential of different investment strategies. To successfully manage wealth requires at least a five or ten-year outlook, during which the market may change considerably, as well as the client’s position and needs. A ten-year outlook will almost certainly cross a downturn in the business cycle, with critical effect on some investment returns.

The above hardly exhaust the factors involved, but they do emphasize the difficulty of obtaining a simple financial analysis of all of the factors. The task is much more than shunning certain sorts of investments, then optimizing an investment strategy within the perceived risk tolerance of the client. The risk tolerance needs to be understood more broadly, and in most cases a broader risk strategy will be decided by the bank, after considering the other influences on the package.

Sometimes providing financial advice is straightforward; sometimes it is more like walking several tightropes at once. The respected investment name comes from handling the difficult cases without fuss.

Understanding the client requires a mixture of analysis and experience operating smoothly over a number of dimensions and combining into an integrated picture. Client feedback needs to also be incorporated into the model.

II. The Portfolio

When a new client presents their portfolio of assets, valuable information can be gleaned from it as to their sophistication. Once the bank has the portfolio under its control, the portfolio needs to be monitored to see that it is meeting the goals set for it.

 III. The Market

The bank maintains a large research team to give it sensitivity to the various asset classes. The market information being passed to the Wealth Management system is dense and complex, and may need to be analyzed by the system for consistency. Other information is required about the costs of changing assets within classes and across classes.

 IV. The Bank

The Bank component of the model acts as the decision-making element. The information coming from the research area needs to be transformed into a form that is relevant to Wealth Management – the returns from the various asset classes need to be seen from the viewpoint of the specific client, with their particular tax and income shaping requirements. The bank is matching the client’s needs against the assets available in the market. Optimization of the portfolio is possible once all the relevant influences have been identified. There may be no acceptable solution, so the decision element would need to search for alternative strategies that would permit a solution.

A Wealth Management package stands in for a Financial Adviser, so if it is accepted by the client, its successful operation carries considerable responsibility for the bank. Wealthy clients can afford litigation over any perceived lack of care, so an opaque "black box" approach to computerized investment management may be hard to defend.

Competition

The behavior of the competition will influence the bank directly, as well as through the behavior of the client. Competition may be through return on portfolio, level of fees or quality of service. The larger the differential on investment returns, the more the bank must show that its strategy for the client is based on a sound knowledge of the client’s life position. Loss of clients to the competition can be analyzed by the system and used in modifying its behavior to other clients at risk.

Interactions Among the Components

I. Client and Bank

This is the main interaction. Mostly it is one way, with the client providing information about their assets, needs and expectations. Occasionally the bank will attempt to influence the behavior of the client by suggesting a reduction in the rate of drawdown, or by requiring agreement before increasing risk.

II. Client and Market

The client may feel a strong "wealth effect" as the market goes up, and considerable fear and loathing as the market goes down. It is easy to insulate a client from the inevitable downturns, much harder to convince them of the wisdom of it in a long bull run. Successfully managing the market influence requires some estimate of the client’s psychology.

 Model Machinery

The elements of model machinery needed to implement the Wealth Management system are:

Model Construction

The model for a particular client needs to be constructed from components that are relevant to the wide variety of individual circumstances with which it must deal. These components are not stored as static templates or profiles, but as sub-models which, when connected together, form an active model. Machinery to construct the individual client model is part of the model, not an outside process.

Analysis

The system must analyze a wide range of situations. It has available to it all of the elements of numerical and logical analysis. The analytic operators in the model can operate directly on probabilistic values.

Simulation

Some of the decision-making requires simulation of financial aspects over time, with different influences being active over different time periods.

Probabilistic Knowledge

The knowledge about the future behavior of the market, and of the future needs of the client, are only known probabilistically. This knowledge is easily captured in distributions and relations, basic components of the model.

Experiential Knowledge

Much of the ability to estimate the needs and behavior of the client will come from experiential knowledge, also held in relations in the model. This knowledge can be automatically extracted from the client database or quickly introduced to the model on an ad hoc basis.

Optimization

The model can create a local area where optimization is possible. It then uses forward cutting to obtain an optimum result. The process of optimization can include any other analysis, as optimization is part of the system’s normal operation. If there is no solution, it can move to a new local area by changing elements within the model.

System Attributes

Modifiability and Extendibility

The market can change its behavior rapidly over time, and new asset classes emerge. Client behavior changes over time, and tax law can acquire special cases rapidly. The system has to be easily modifiable. The eCognition model can modify its own structure, and can accept new pieces of knowledge that extend or modify its network. Its experiential knowledge can be automatically mined from transactions. The network structure allows extension in any direction at any time.

Verifiability

The model network represents all the data and processing of the system. Influences can be followed from where they arise to their point of application.

Reinsurance for Businesses

This document describes a suggested application of Tupai technology at a Reinsurer. We have made several assumptions in describing the application:

Dense and Complex Information Flows

The reinsurer has a great deal of information about hazards – earthquakes, tornadoes, marine and air disasters, human error. This information is complex, much of it is domain specific, and it needs to be brought to bear on the evaluation of a particular risk and the assumption of some part of that risk. The information also needs to be used to onsell the risk to other parties who lack the sophistication to fully understand the risk factor research, but who need sufficient information to allow them to integrate the risk they buy into their portfolio.

Complex Position

The reinsurer has a portfolio of investments designed to protect the reinsurer against both rapid short term fluctuations in liquidity caused by several major events occurring together, and long term support of the risk portfolio. The reinsurer needs to put together a complex portfolio of risk, fairly priced. The reinsurer must be confident of its ability to weather the lean years that will inevitably come in this business. Sometimes new research will show that the position in some area is optimistic, even untenable, and needs to be slowly unwound. The more knowledge that can be brought to bear on each transaction, the more confidence that the overall portfolio is wisely constructed.

Complex Client

The client of the reinsurer will either be a large insurance company or a large corporate client capable of carrying some but not all of the risk themselves. If the payout would be to enable the client to continue their business, the reinsurer needs to be convinced that the client would survive the shock and continue to pay the income stream that allowed the risk to be assumed. The client may be large, leading to overexposure, even though the incremental risk is small. The client may be nave - not fully comprehending their operational risks, unable to see past a physical asset to the total business risk.

Overview

A large reinsurer wishes to deploy its knowledge, built up over more than 100 years, more widely. Of late, that knowledge has become increasingly complex and specialized, as its experts become more skilled in forecasting the likelihood of earthquakes and other natural hazards. The events it insures are moving from the destruction of physical assets, a plane down, to the destruction of intangible assets - an oil spill, an extortion attempt on a pharmaceuticals manufacturer. The increasing scale of the shock of losing some physical asset also require the strongest stomach for risk and the coolest head – the A-3XX for example.

The reinsurer realizes it does not just buy and sell risk, but also sells its knowledge about risk, making it a "knowledge company". It needs an environment which will allow it to acquire increasingly complex and technical knowledge, and to filter and deploy it ever more widely.

To be successful, the reinsurer must understand its client (often a chain of clients, at each remove more loosely bound) and the risk, then make a proposal that takes into account the particular needs of the client (including their behavior if the insured event should occur), the position of the reinsurer vis--vis the risk, and a clear eyed assessment of the risk.

Research has shown that the service that clients seek most is a trusted and stable reinsurer who can crystallize the risk issues involved, and also provide support in the areas of financial and disaster planning. The company intends to deliver these services through a choice of channels, including the Internet.

As the reinsurer delves into the problem, the magnitude of the challenge begins to emerge. It quickly becomes apparent that the planned system must deal with a dynamic environment in which client, type of and knowledge about risk, and the reinsurer itself are continuously evolving and changing. Consequently, in its interaction with its clients, the reinsurer must continuously adjust its actions based on the client’s needs and the reinsurer’s knowledge and position.

Solution Overview

In the following description we focus on the elements involved in building a system for reinsurance. We will use a scattergun of problems that relate to different areas but emphasize the difficulty of analysis. Our intent here is to highlight the fact that providing reinsurance requires the integration of multiple types of knowledge from multiple domains. As we mentioned earlier, this ability to integrate the knowledge process is a cornerstone of Tupai’s technology.

eCognition addresses the fluid interaction of client, risk, position by constructing a representation (or model) of the situation, and then performing analysis on it. In conducting its analysis, the system is able to deploy all kinds of analysis at every stage. The analysis is not some external process, but flows naturally from activity within the model.

The overall eCognition model for reinsurance comprises of a number of interacting components – the client, the risk, and the reinsurer’s position. Each of these components is modeled, as well as the effects of their interactions.

Model Components

I. The Client

We can draw on generic risk probabilities. Do we need to modify those risks for this particular client – are they operating to industry-standard norms, or are there special conditions that increase the risk. The client is a complex amalgam of location, operation, psychology. Some parts we can analyze precisely, some parts we can analyze using probability, some parts we can only estimate using influences.

Some of the elements of the client’s position (we are assuming the industrial client is large enough for us to deal with them directly):

Location – are there geographic factors – wing icing, poor facilities
Sophistication of workforce
Reliance on others – do they maintain their own planes
Sophistication of market
Risk minimization strategies
Are they under excessive stress – will they cut corners in the future, or be taken over and operate completely differently

The above hardly exhaust the factors involved, but they do emphasize the difficulty of obtaining a simple analysis of all of the factors.

If we are dealing with an insurance company, have they understood and fairly transmitted the risk to us. Do we need to examine their risk portfolio before accepting any risk from them. Have they recently changed their business practices, so our history with them is of no use.

Understanding the reinsurance client requires a mixture of analysis and experience operating smoothly over a number of dimensions and being combining into an integrated picture of what may be a moving target.

II. The Risk

The reinsurer maintains ongoing research into risk. To a large extent it must be reactive. Some new risks without history need to be estimated based on similar phenomena. Old risks are changing their shape, with companies increasingly being asked to clean up after their disasters.

Information about risk comes in many forms. It needs to be combined into a meaningful whole, which can be applied consistently at any office throughout the world.

III. The Reinsurer

The Reinsurer component of the model acts as the decision-making element. The information coming from the research area needs to be transformed into a form that is relevant to the client’s circumstances – the returns from the various asset classes need to be seen from the viewpoint of the specific client, with their particular tax and income shaping requirements. The reinsurer is matching the client’s risk against the reinsurer’s position for that class of risk and for the financial outlook of the reinsurer’s assets. Optimization of the business proposal is possible once all the relevant influences have been identified. There may be no acceptable solution, so the business must be declined.

There can be very large sums of money changing hands, so the successful operation of the system carries considerable responsibility. An opaque "black box" or sequential programmed approach to computerized reinsurance would be hard to defend in comparison with a knowledge based approach that can demonstrate a large number of influences being taken into account in the decision making process.

Competition

Competition may come from cost of risk or quality of service. The larger the differential against lesser reinsurers, the more the Reinsurer must show that its strategy for handling the client’s risk is based on a sound knowledge of the whole position over the long term.

Interactions Among the Components

I. Client and Risk

This is the main interaction. The client affects the risk, and the risk coming to pass may destroy the client. Occasionally the reinsurer may need to mandate the use of risk minimization strategies, sometimes only realizing their necessity long after the ink is dry.

II. Reinsurer and Risk

The reinsurer, by publishing his research, may define the risk wider than was originally thought, leading to potentially larger payouts. Investigating the possible effects of a 1000 year flood can lead the client to say "You knew all this – now you must pay".

Model Machinery

The elements of model machinery needed to implement a Reinsurance system are:

Model Construction

The model for a particular risk and client needs to be constructed from components that are relevant to the wide variety of circumstances with which the system must deal. These components are not stored as static templates or profiles, but as sub-models which, when connected together, form an active model. Machinery to construct the individual model is part of the model, not an outside process.

Analysis

The system must analyze a wide range of situations. It has available to it all of the elements of numerical and logical analysis. The analytic operators in the model can operate directly on probabilistic values. Dense and complex information can flow through the connections in the model and be analyzed.

Simulation

Some of the decision-making requires simulation of risk scenarios and financial aspects over time, with different influences being active over different time periods.

Probabilistic Knowledge

The knowledge about risks is only known probabilistically, and may span many dimensions. This knowledge is easily captured in distributions and multi-dimensional relations, basic components of the model that allow information to flow across the dimensions of time, risk and money.

Experiential Knowledge

Much of the ability to estimate the risk will come from experiential knowledge, also held in relations in the model. Much of this knowledge can be automatically extracted from research databases, or quickly introduced to the model on an ad hoc basis.

Optimization

The model can create a local area where optimization is possible. It then uses forward cutting to obtain an optimum result. The process of optimization can include any other analysis, as optimization is part of the system’s normal operation. If there is no solution, it can move to a new local area by changing elements within the model.

MicroScheduling

The network is managing the interaction of many influences. The individual operators in the network control their knowledge needs, passing requests for information through the network. Complex interactions are obtained by allowing the many small elements of the network to be responsible for their own states. New influences can be easily added without concern for sequence of operation.

 

Additional Applications

The Reinsurance system is intended for support of an underwriter or for interaction directly over the web where the involvement of a human analyst would not be cost effective.

The good model of the client is of advantage if and when the predicted hazard eventuates, allowing estimates to be verified and new lessons learnt to be quickly integrated into the system – a strong reason for having a knowledge based system.

Medical Diagnosis

 

Summary

Computer-assisted (or fully computerized) business services must be capable of responding to considerable diversity. They must smoothly combine analytic and experiential knowledge, and rapidly alter their behavior in response to new information. The implementation of such systems requires an across-the-board improvement in a wide range of capabilities, beginning with how knowledge is discovered, continuing through how it is used for simulation and analysis, and concluding with its deployment. Stated differently, the entire knowledge process needs to be integrated.

The implementation of point solutions to handle different aspects of the problem not only introduces integration challenges but may also sour management’s enthusiasm for the project by producing partial and unexciting initial results.

We propose that only eCognition offers the breadth of capabilities needed to build systems that overcome all current barriers and bring deals to their successful conclusion. No less importantly, Tupai’s quick implementations go a long way to getting projects up and running, demonstrating early successes and fortifying management’s support.

Finally, unlike other approaches for capturing and deploying knowledge, Tupai’s technology enables transparency and verifiability of its workings, thereby supporting the controls and audits that are essential to responsible business management.

Home