1、 外文翻译 Intelligent process analytics for CRM Material Source: BT Technology Journal Vol 24 No 1 January 2006 Author: B Azvine, D D Nauck, C Ho, K Broszat and J Lim Many organisations use decision models in their processes such as tables or trees to provide decision support to their operational divisi
2、ons. For example, in fault management, customer contact centre operators usually use a decision model in the form of prescribed interviews. Based on the answers given by customers, the operator navigates through the decision model to reach an assessment of the problem. In order to achieve customer s
3、atisfaction and operational excellence, it is very important to constantly monitor the performance of a decision model not only on an overall level, but also on the level of individual decisions. In this paper we present a configurable business process analytics tool, known as the intelligent Univer
4、sal Service Management System, that constantly monitors decision data and is capable of optimising the decisions based on high-level business objectives. We explain the various features of the software and show how it can be used to optimise decision processes. We also show how we can easily provide
5、 a customised version to monitor the performance of provision processes. 1.Introduction Business analytics systems monitor data generated in business operations in order to analyse performance based on key indicators, and present the analysis results to a wide range of users in a format that can be
6、grasped intuitively. Consider the following scenario. A customer calls BT to report a fault on the telephone line. The operator logs the customer details and fault information. The operator is informed that an engineer is currently in the area where the fault has occurred. The customer is informed a
7、nd re-assured that the engineer will attend to the fault as soon as possible. At the time the call is taking place, automatic analysis of the fault shows that there have been an unusually high number of faults reported in the same geographical area in the last two weeks. This trend triggers an e-mai
8、l to the analysis centre where a group of specialists work round the clock to predict and react to current and future problems. The result of their analysis is directly linked to information available to operators and engineers. All the operational processes are monitored and optimised according to
9、organisational targets and strategy. One of the specialists realises that a particular road is the location of most reported faults. This information is automatically sent to the mobile devices of engineers in the location above. One of the engineers, while looking at the scene with analysis results
10、 superimposed as shown in Fig 1, notices a burst pipe in the area where there is some construction activity. On examination of the underground line, minor damage is detected and repaired. The customer who called and all other customers in the area are informed of the repair and the cause of the faul
11、t. Clearly most organisations would wish to turn data into information into action as described in the scenario. However, the barriers to realising such a vision are significant. One of the most important barriers is the manual process involved in gathering, analysing and distributing the results of
12、 the analysis. Overcoming this barrier is the primary goal of the intelligent process analytics platform described in this paper. One can divide the development of business analytic systems into four generations. The first-generation business information systems were based on centralised batch proce
13、ssing. Analysis of data contained in such systems was difficult and required experts. The second-generation data warehousing systems based on client/server computing were a major step forward, but, if the information was not readily available in the warehouse, it could not be found easily by busines
14、s users. The third-generation systems saw the introduction of OLAP, data mining and Web deployment. Packaged analytic applications in third-generation systems accelerated this deployment. However, detailed analysis of the data still required specialist skills and a high level of familiarity with the
15、 available components. As competition requires companies to respond ever faster to customer needs and rapidly revise their internal and external processes, real-time closed-loop processing of data is becoming a necessity. The integration of business analytics into the overall business process can be
16、 achieved by building a closed-loop decision-making system in which the output of business analytics is used by operational managers in the form of recommended actions. The vision of intelligent business analytics is to extend this closed-loop process to the automatic adjustment of business operatio
17、ns, based on decisions made through analysis of available data in real time. A closed-loop enterprise analytics system that can support real-time processing represents a fourth-generation of business analytics software. In order to achieve this vision, we set out to develop a platform that would fac
18、ilitate the integration of the latest intelligent technology within business analytics applications. The objective was to reduce the skill barrier, reduce development time and cost, and take advantage of economies of scale by reusing the platform to develop many applications. 2.iUSMS In a unified sy
19、stem management scenario, different manual, semi-automatic and automatic diagnostic systems and fault reception procedures carry out tests on systems, services and equipment, and make decisions based on the outcome of their tests. A decision can be a diagnosis, e.g. the identification of the cause o
20、f a fault,or the trigger for an action, such as dispatch an engineer to a fault location based on a diagnosis and possible additional information. A decision can always be interpreted as the output of a decision model. While automatic systems require a formal decision model, many manual and semi-aut
21、omatic systems are still based on informal decision models. A formal decision model is a documented process of computing the value of one or more decision variables based on a collection of input variables. Input variables can describe process observables, policies, external conditions, historic mea
22、surements, etc. In its simplest form, a formal decision model can be represented in the form of a table that records a decision for each combination of input values. If such a representation is deemed to be too inefficient and difficult to understand, maintaining knowledge-based representations can
23、be an alternative. A knowledge-based representation generalises from individual input/output combinations and represents a decision model in the form of a rule base, a decision tree, or a set of equations, for example.If a decision model was created by an automatic machine-learning process, it can h
24、ave the form of ablack box that computes output values from input values and does not have any interpretable representation. This is, for example, true for artificial neural networks or support vector machines popular approaches for learning decision models in knowledge- poor but data-rich environme
25、nts. An informal decision model is not necessarily documented and can be simply the collective experience of engineers or process managers. Sometimes there may be a collection of anecdotal evidence in the form of manuals or briefings that describe decision scenarios in natural language. While formal
26、 decision models can be analysed and optimised based on their representation a decision table, for example, can be checked line by line for correctness informal decision models are very difficult to analyse. Analysis is only possible if, for each decision, the available data is collected, i.e. the i
27、nformation that is used to make a decision, the decision itself, the information used to determine if that decision was correct, and if not what the correct decision would have been.This form of data collection is also required to analyse the performance of formal decision models. It also provides u
28、s with the opportunity to turn any informal decision model into a formal one by using data analysis and machine learning techniques to derive a model from the collected data. If a formal model already exists the data can be used to optimise it. In order to achieve operational excellence and a high l
29、evel of customer satisfaction, it is necessary to collect all available data related to decisions, and to regularly evaluate all decision models in terms of accuracy, cost and impact on processes.The intelligent Universal Service Management System (iUSMS) is a flexible and highly customisable softwa
30、re platform for analysing process data of any format in a host database or data warehouse. iUSMS provides visualisation tools to enable the user to drill-down into data and to analyse and understand decisions made by people, processes or diagnostic systems. A decision, for example, can be a test res
31、ult requested by a diagnostic system under certain conditions. iUSMS provides reports and statistical information about the decisions made, e.g. their accuracy and cost. These derived statistics are kept in the database for further analysis or comparison. Users can analyse the conditions of individu
32、al decisions or groups of decisions. The view of the database can be easily customised to improve its quality by cleaning, conversion and transformation. New fields or derived statistics can be added, fields can be hidden, and values of fields can be grouped as necessary for further data analysis.On
33、 top of these analysis facilities, iUSMS contains a learning and knowledge discovery module. By comparing feedback data with recorded decisions, iUSMS can identify decision models that do not perform as expected. The module can then automatically use process and feedback data to generate different t
34、ypes of decision models that can be used within processes. Some models, like decision trees or rules, can also be used to describe the patterns derived from the data. The accuracy and cost of each individual pattern can be computed and compared against actual decisions recorded in the data. Decision
35、 models can be deployed in real time into systems and processes in such a way that formal decision models can always be kept up-to-date and highly optimised. The visualisation facilities of iUSMS contains features to set up intuitive interfaces like dashboards that help the user to spot any unusual
36、changes quickly, e.g. a degrading trend in a customer satisfaction indicator, or rising operational costs. The system allows the user to easily configure the dashboard to display any information available in the data, like key performance indicators (KPIs) or some critical system measures. The syste
37、m can raise alarms by producing an ad hoc report of analysis results or generate messages for operational staff who could then be automatically contacted by e-mail or SMS. 3.The architecture iUSMS is an extension of our Intelligent Business Analytics (IBA) platform , and follows the same design logi
38、c by making use of the following features: if the same application, i.e. the same business logic, is used for another customer the code should not change, any SQL database can be used to store the data, according to the requirements of the application domain, suitable data analysis methods must easi
39、ly plug in, it should be possible to exchange the business logic but still retain a lot of the code, an application should be Web-based such that thin clients can be used, but it should also be able to run as a stand-alone version that does not require network access, it provides a library of intell
40、igent data analysis (IDA) such as decision trees, rule generators,Bayesian classifiers, hidden Markov models, neural networks, etc, it provides a number of reusable GUI modules (reports, charts, maps, etc). Like IBA, iUSMS has a three-tier architecture. A database server holds the operational data a
41、nd the data structures of the business logic. Java servlets provide data management, data preprocessing and business analytics. The client side is implemented as a Java application and provides the GUI to the application. The client connects to a Java-enabled Web server like Tomcat or Weblogic that
42、in turn executes the servlet part of the application. In comparison to IBA, iUSMS provides the following extensions. Generic data preprocessing iUSMS assumes that the data for each decision process is available in a single relation (data table). Multi-stage decisions will be represented by multiple
43、relations. This approach was used to prevent the necessity for time-consuming separate runs for data analysis and model building. Because the data is frequently provided in some legacy format, iUSMS provides import routines for text-based data representations. Once the data is uploaded, it is availa
44、ble as a database table and preprocessing can commence. The preprocessing steps comprise data type conversion, computation of standard statistical information for each attribute, and attribute value grouping based on a target (decision) variable to reduce data complexity. Model visualisation Decisio
45、n models resulting from an intelligent data analysis process can now be visualised as trees, rule sets, or specialised charts. Report generator The iUSMS client can capture any chart or analysis result and include it in an HTML report. Drill-down charts The new graphical user interface allows the us
46、er to drill down into charts by implementing a grouping functionality. User management Apart from a standard log-in facility, iUSMS has the capability to manage user types to grant or restrict access to any function within the program. When we use iUSMS to provide an application for a new domain, we
47、 can easily introduce new functionality in four ways: by providing a new dialogue on the client we can implement customised charts and reports if the report does not require additional analysis beyond that which is already provided by the server, no further implementation is necessary, if a new func
48、tionality requires customised analytics,a new method is implemented on the server side to provide results for the new dialogue on the client,while on the other hand, if no new data structures or IDA models are required, no further implemen-tation is needed, if a new analytical method requires custom
49、ised data structures, they are implemented in the database schema of iUSMS, the implementation being done by providing the necessary SQL statements in the servers XML configuration file this reduces the implementation overhead in the Java classes and allows data representations to be changed without altering the code of the server, if a new analytical method requires a new IDA algorithm, this algorithm is incorporated into the IDA library of iUSMS and is then available to all analytical methods, while on the other hand, if the IDA algorithm pro
