by: Max J. Pucher
Photo: res.sys-con.com
The proponents of BI however hope to fulfill the vision of Laplace; who suggested: '... with an intelligence sufficiently vast to submit all comprehensable data to analysis ... nothing would be uncertain and the future and past present to his eyes.'
Business Intelligence is not a productive system, such as accounting, stockkeeping, or supply chain management. Customer care or relationship management should first be consodered as a tool to manage and track your customer service, but is often seen as an analysis tool for customer behavior to improve for example cross-selling rather than service quality.
Business Intelligence is complex and expensive software for gathering and analyzing masses of data that supposedly will be of help to make better business decisions. It relies on a faith in mathematics that were obtained by Bernoulli, Coombs, Edwards, Neumann and Morgenstern. They saw heuristic approaches to decision making as defective because it takes resource saving short-cuts.
The proponents of BI now claim that it provides the computing power to optimize decision making by calculating probability for maximum utitlity, as described by Simon (1955). Optimisation however, relies on a number of restrictive assumptions, such as that the process of decision analysis has to be followed precisely and that the data available are correct and relevant. Let's for a moment assume that the garbage-in problem has been solved and the data given to the business executives are good.
Klein (1999) provides a list of these restrictions and requirements that were identified in many studies :
1. The goals must be well defined in quantitative terms.
2. The decision makers values must be stable.
3. The situation must be stable.
4. The task is restricted to the selection of options.
5. Exhaustive generation of alternatives.
6. Optimal choice must be possible in reasonable time.
7. Thorough comparison of options.
8. Use compensatory strategy
9. Probaility estimates must be coherent and accurate.
10. Failure prediction must be exhaustive.
11. Evaluation must be exhaustive
There are many studies about human decision making and most of them come to the conclusion that 'less is more'. Less information about a subject makes for better decisions. Bi can be used to gather and consolidate information that then seems to be simpler and easier to use for decisions. The problem is one of comprehension and trust. Can the user comprehend what the data values truly mean? Do the metadata make sense to the deciding person? Can past averages, means, standard deviations and periodic data be used to predict the future? I propose that just a few people within any organization might even sensibly comprehend what those data could mean. The old adage of garbage-in-garbage-out still holds. Who knows if the mathematical methods used to process the data are well chosen.
Klein (1999) proposes that forcing people to give up their heuristic approach to decision making puts them into 'information overload' and questions optimization as the gold standard for decision making.
Now that users of Business Intelligence data do not find them too helpful and seem overwhelmed, the new idea is to aid or replace human decision making with predictive analytics, using probability calculated from past data. Probability computing about future events is the next illusion that BI proponents sell.
Here is a list of what the executives and managers really need for decision making and don't get from Business Inteligence:
- what customers want.
- what to do to be competitive.
- where business has to innovate.
- how the market will react to current and upcoming changes.
- what competitors are currently doing.
- how employees really see the company.
- how to impove the profitability of the business.
- how to improve internal communications.
- the quality of business processes (not the quality of execution).
You might recognize a common element in the above list: Knowledge is not about knowing a lot of data, or taking decisions based on data. One has to come up with an ACTION or a list of alternative actions and then take a decision which one to perform. Finding out from BI that revenue is dropping only says that the management is out of touch with the business.
Simon (1972) was concerned that optimization was not practical in a field setting because of its restrictions. I propose that there is no proof available that BI solves the problem of the restrictions listed above. Business Intelligence can enforce the optimization process for the decision maker and seemingly create the conditions necessary. It can however only propose options based on the data available to the system. It can not propose to the user to go outside the system and analyze other information as it brakes the optimization process. As a consequence it BLINDS the decision maker to the real world. Klein (1999) proposes that it the enforced optimization process stops the decision maker from gaining experience for future benefit. Outside opportunities and constraints would be totally ignored.
Business Intellilgence enforces a better decision making process but not a better quality of decision making.
Timely and good quality information is not like having a crystal ball. It is abstract information and will tell the user nothing that he doesn't know. We can only make decisions based on analogies to previously perceived patterns that have to be fairly simple. Statistical software does not take better decisions just because it can process more data. Any data given to the user that is not truly relevant turns into noise that reduces the quality of the communication channel and obfuscates the important information. Less data means less noise.
What information should it be that truly represents a company's competitive position in the market? Past sales data and comparisons of market share? What decision will that offer? The best way to find out about the competion is to ask a customer who decided for another product why. A competitor is a friend who helps the business to improve what it does. Without competitors companies would become complacent. Inteviewing five of those customers who switched - ideally face to face - will provide much more decision making input than statistics. That certain groups of randomly classified people spent a certain amount on randomly classified goods is not knowledge. A qualified manager who attends sales calls or walks into the store to speak with customers will see those changes in customer behaviour in real-time. He can ask about customer preferences at the time they happen and not months later in a filtered and watered down manner that is completely abstract.
March (1978) and Simon (1983) propose that people do not decide by calculating expected utility and question its mathematical foundation (that is used in BI) for real-world situations. Klein (1999) proposes that training people by exposing them to experience of decision making is more important than abstract optimization processes. This falls in line with Thomas Sowell in 'Knowledge and Decisions' who calculates the cost of knowledge by its practical usefullness and not by the amount of abstract eductation.
Antonio Damasio (1995) has virtually proven the influence of our emotional center on human decision making and Steven Johnson paints a wonderful picture of the power of our human mind in its connection of instincts, intuitions and emotions created by the link between the neocortex and hippocampus, and amygdala. Good decisions come from feeling good about a decision and mathematical optimization strategies fail to do that.
Massive business intelligence data do not eliminate the guesswork but create a substantial amount of new guesses that have no connection with the real world situation. Yes, gather business relevant data, filter it and link just the neceassary detail right into your business service, that will improve decision making. Our uncertainty is not reduced by knowing more, as we all have experienced. That is the approach we propose at ISIS Papyrus Software.
Bbliography:
A. Damasio (1995) Descartes' Error: Emotion, Reason, and the Human Brain
D. MacKay (2003) Information Theory, Inference, Learning Agorithms
G.Gigerenzer, R. Selten (1999), Bounded Rationality
G.Gigerenzer, P. Todd (1999) Simple Heuristics that make us smart
Thomas Sowell (1996), Knowledge and Decisions
S. Johnson (2004), Mind Wide Open
Max J. Pucher
Max J. Pucher is the founder and current Chief Architect at ISIS Papyrus Software, a globally operating company that specializes in Artificial Intelligence for business process and communication. He has written several books, frequently speaks and writes on IT and holds several patents.
Photo: res.sys-con.com
The proponents of BI however hope to fulfill the vision of Laplace; who suggested: '... with an intelligence sufficiently vast to submit all comprehensable data to analysis ... nothing would be uncertain and the future and past present to his eyes.'
Business Intelligence is not a productive system, such as accounting, stockkeeping, or supply chain management. Customer care or relationship management should first be consodered as a tool to manage and track your customer service, but is often seen as an analysis tool for customer behavior to improve for example cross-selling rather than service quality.
Business Intelligence is complex and expensive software for gathering and analyzing masses of data that supposedly will be of help to make better business decisions. It relies on a faith in mathematics that were obtained by Bernoulli, Coombs, Edwards, Neumann and Morgenstern. They saw heuristic approaches to decision making as defective because it takes resource saving short-cuts.
The proponents of BI now claim that it provides the computing power to optimize decision making by calculating probability for maximum utitlity, as described by Simon (1955). Optimisation however, relies on a number of restrictive assumptions, such as that the process of decision analysis has to be followed precisely and that the data available are correct and relevant. Let's for a moment assume that the garbage-in problem has been solved and the data given to the business executives are good.
Klein (1999) provides a list of these restrictions and requirements that were identified in many studies :
1. The goals must be well defined in quantitative terms.
2. The decision makers values must be stable.
3. The situation must be stable.
4. The task is restricted to the selection of options.
5. Exhaustive generation of alternatives.
6. Optimal choice must be possible in reasonable time.
7. Thorough comparison of options.
8. Use compensatory strategy
9. Probaility estimates must be coherent and accurate.
10. Failure prediction must be exhaustive.
11. Evaluation must be exhaustive
There are many studies about human decision making and most of them come to the conclusion that 'less is more'. Less information about a subject makes for better decisions. Bi can be used to gather and consolidate information that then seems to be simpler and easier to use for decisions. The problem is one of comprehension and trust. Can the user comprehend what the data values truly mean? Do the metadata make sense to the deciding person? Can past averages, means, standard deviations and periodic data be used to predict the future? I propose that just a few people within any organization might even sensibly comprehend what those data could mean. The old adage of garbage-in-garbage-out still holds. Who knows if the mathematical methods used to process the data are well chosen.
Klein (1999) proposes that forcing people to give up their heuristic approach to decision making puts them into 'information overload' and questions optimization as the gold standard for decision making.
Now that users of Business Intelligence data do not find them too helpful and seem overwhelmed, the new idea is to aid or replace human decision making with predictive analytics, using probability calculated from past data. Probability computing about future events is the next illusion that BI proponents sell.
Here is a list of what the executives and managers really need for decision making and don't get from Business Inteligence:
- what customers want.
- what to do to be competitive.
- where business has to innovate.
- how the market will react to current and upcoming changes.
- what competitors are currently doing.
- how employees really see the company.
- how to impove the profitability of the business.
- how to improve internal communications.
- the quality of business processes (not the quality of execution).
You might recognize a common element in the above list: Knowledge is not about knowing a lot of data, or taking decisions based on data. One has to come up with an ACTION or a list of alternative actions and then take a decision which one to perform. Finding out from BI that revenue is dropping only says that the management is out of touch with the business.
Simon (1972) was concerned that optimization was not practical in a field setting because of its restrictions. I propose that there is no proof available that BI solves the problem of the restrictions listed above. Business Intelligence can enforce the optimization process for the decision maker and seemingly create the conditions necessary. It can however only propose options based on the data available to the system. It can not propose to the user to go outside the system and analyze other information as it brakes the optimization process. As a consequence it BLINDS the decision maker to the real world. Klein (1999) proposes that it the enforced optimization process stops the decision maker from gaining experience for future benefit. Outside opportunities and constraints would be totally ignored.
Business Intellilgence enforces a better decision making process but not a better quality of decision making.
Timely and good quality information is not like having a crystal ball. It is abstract information and will tell the user nothing that he doesn't know. We can only make decisions based on analogies to previously perceived patterns that have to be fairly simple. Statistical software does not take better decisions just because it can process more data. Any data given to the user that is not truly relevant turns into noise that reduces the quality of the communication channel and obfuscates the important information. Less data means less noise.
What information should it be that truly represents a company's competitive position in the market? Past sales data and comparisons of market share? What decision will that offer? The best way to find out about the competion is to ask a customer who decided for another product why. A competitor is a friend who helps the business to improve what it does. Without competitors companies would become complacent. Inteviewing five of those customers who switched - ideally face to face - will provide much more decision making input than statistics. That certain groups of randomly classified people spent a certain amount on randomly classified goods is not knowledge. A qualified manager who attends sales calls or walks into the store to speak with customers will see those changes in customer behaviour in real-time. He can ask about customer preferences at the time they happen and not months later in a filtered and watered down manner that is completely abstract.
March (1978) and Simon (1983) propose that people do not decide by calculating expected utility and question its mathematical foundation (that is used in BI) for real-world situations. Klein (1999) proposes that training people by exposing them to experience of decision making is more important than abstract optimization processes. This falls in line with Thomas Sowell in 'Knowledge and Decisions' who calculates the cost of knowledge by its practical usefullness and not by the amount of abstract eductation.
Antonio Damasio (1995) has virtually proven the influence of our emotional center on human decision making and Steven Johnson paints a wonderful picture of the power of our human mind in its connection of instincts, intuitions and emotions created by the link between the neocortex and hippocampus, and amygdala. Good decisions come from feeling good about a decision and mathematical optimization strategies fail to do that.
Massive business intelligence data do not eliminate the guesswork but create a substantial amount of new guesses that have no connection with the real world situation. Yes, gather business relevant data, filter it and link just the neceassary detail right into your business service, that will improve decision making. Our uncertainty is not reduced by knowing more, as we all have experienced. That is the approach we propose at ISIS Papyrus Software.
Bbliography:
A. Damasio (1995) Descartes' Error: Emotion, Reason, and the Human Brain
D. MacKay (2003) Information Theory, Inference, Learning Agorithms
G.Gigerenzer, R. Selten (1999), Bounded Rationality
G.Gigerenzer, P. Todd (1999) Simple Heuristics that make us smart
Thomas Sowell (1996), Knowledge and Decisions
S. Johnson (2004), Mind Wide Open
Max J. Pucher
Max J. Pucher is the founder and current Chief Architect at ISIS Papyrus Software, a globally operating company that specializes in Artificial Intelligence for business process and communication. He has written several books, frequently speaks and writes on IT and holds several patents.
No comments:
Post a Comment