posted 1 May 1998 in Volume 1 Issue 5
Applying Customer Knowledge
Sam Korman, Abbey National Plc, discusses how presumptive knowledge can be derived and used to establish improved customer relationships in a Retail Financial Services setting.
In an economic environment that continues to be increasingly competitive, organizations must make maximum use of their corporate knowledge, by capturing, storing, managing and sharing it with all relevant individuals within the corporation.
Traditionally, this knowledge has been classified into two components:
|explicit knowledge, which is written down and encapsulated in the form of data inventories, formal procedures, best practice guides, and so on; and|
|tacit knowledge, which is the understanding that individual staff members often carry around on their heads or on localised records, and which enables them to undertake their work more effectively and more efficiently. Examples of this might be an informal list of expert contacts, insight into technical or work processes, general rules of thumb that have been built up from experience, and so on.|
As the latter can be specific to an individual and often not explicitly shared with others, it is at risk of being lost to the organisation should that individual member of staff leave or transfer to other responsibilities.
The practice of knowledge management, then, is to establish appropriate methods for codifying and recording the above knowledge, disseminating it among relevant staff (often within special interest communities inside the organisation and perhaps elsewhere), and maintaining the up-to-date currency of the knowledge. An effective strategy and implementation for this will then enable the organisation to become more efficient through the gradual adoption of best practice throughout, improving both its efficiency and time-to-market of new initiatives.
However, a company that deals with a large number of customers or is involved with a large number of similar processes where understanding or decisioning is required, has a further opportunity to derive useable knowledge. This would arise from analysis of previous instances, by associating the outcome of any case with the factors underlying the decision, and thereby determining those characteristics and their combination most likely to be indicative of favourable outcomes.
In this way, in any particular instance, the conditions applying to that case can be assessed and the likelihood of a favourable outcome estimated, and used as part of the decision-taking process. The knowledge that leads to this determination might be termed presumptive knowledge, in that it is neither explicitly indicated nor tacitly known and, indeed, might even be erroneous in individual cases - although on average it is more likely to be correct than not. Because of this aspect, it would also be relevant to establish and record the degree of correctness expected as part of the presumptive knowledge.
This article discusses how presumptive knowledge can be derived and used to establish improved customer relationships in a Retail Financial Services setting.
Abbey National Plc is the 5th largest Retail Bank in the UK, providing a full range of banking, savings, investment, insurance, and mortgage products to some 15 million customers via a number of distribution channels, including extensive Branch and Introducer networks, telephone and post.
As with many similar institutions, the financial marketplace within which Abbey National operates continues to become more competitive. There has been a recent pattern of new entrants coming into the market (or parts of it), often with newer computer systems infrastructure and with restricted distribution channels. Included among these are non-traditional providers such as supermarkets, as well as established credit operators from abroad who are expanding into other countries.
As a consequence, products are becoming more commoditized, margins coming under increasing pressure, product innovations being copied more quickly, and many customers are spreading their business and in general becoming more fickle.
Nevertheless, a combination of brand loyalty and the proactive initiatives that are enabled by “ownership” (or rather “stewardship”) of its large base of customers provides the established business with the opportunity to maintain and, indeed, enhance its share of the market.
A key approach towards utilising these strengths is to leverage the organisation’s data and systems in order to help build and maintain the most profitable relationships with customers.
An important methodology for applying this strategy is by implementing the Learning Cycle, as denoted in Figure 1.
Figure 1: The Learning Cycle
Before describing the elements of this cycle, it is worth briefly defining the key differences between data, information, and knowledge (as interpreted in the context of the developments described here).
|Data are elemental facts captured by, or otherwise known, within the business.|
|Information is obtained by aggregating data to produce meaningful and manageable categorisations. Typically these will summarise past events.|
|Knowledge is derived by combining information and data in order to provide insight that will serve as a basis for future actions.|
As an example in the case of data on customers, we might have among our many product records the facts that:
|Mr J Smith has purchased Product A|
|Mr John Smith has purchased Product B|
|Mr & Mrs J.A. Smith jointly hold Product C|
This is the elemental data.
A brief aggregation establishes that the address and date of birth for each of these product holdings is identical, which leads to the recognition that each of these customer records relates to the same individual. This is information.
Adding to this the fact that the typical customer holds less than 2 products, and establishing from the records how long these products have been held and how the accounts have been conducted, could lead to the conclusion that Mr Smith is a loyal and profitable customer.
This is knowledge, which can be utilised for future decision-making. Knowledge is therefore only worthwhile if it is actionable, whether explicitly or implicitly.
We now return to look at each of the elements of the learning cycle.
Data - Sources and Repositories
Data relating to customers, either individually or generally, can come from a number of sources. It will often be provided by the applicant/customer, be recorded automatically by transactional systems in consequence of an activity, arise from a general customer contact or inquiry, and so on. Other data might be purchased from external sources, such as credit reference agencies, geo-demographic segmentation vendors or third party aggregators of market data.
(This last category includes bureaus that operate on a reciprocity basis with those client organizations that provide sensitive market data, on the understanding that only the aggregates of all data will be provided to the other members, thereby enabling each business to obtain detailed market share data without exposing competitor confidentiality).
The data will be collected and used for one or more of a number of purposes, including regulatory, operational, analysis and provision of Management Information, strategic planning, and knowledge generation and modelling.
It might be noted that each of these uses has different requirements regarding accuracy, currency, level of aggregations, details of access, and so on.
The data for facilitating Information derivation and knowledge generation (as defined above) would often be stored in a Data Warehouse. This would typically be relational and customer-orientated, drawn from number of sources, up-to-date but also retaining accounts history, and would include transactional data or detailed summarisation’s thereof.
This primary repository might be complemented by specifically generated datasets for analysis purposes; which in essence would comprise comprehensive snapshots with history, supplemented by access to core ledger system data, and external data as needed.
There would also be facilities and tools for data manipulation activities, such as further data aggregations, generation of characteristics, data visualisation, analytical modelling, and knowledge mining generally.
Generating Presumptive Knowledge
There are a number of techniques for presumptive knowledge generation, such as statistical methods, neural networks, evolutionary algorithms, rule induction, and so on. Each has its own particular strengths and weaknesses and therefore preferred applicability in particular areas. However, the process by which they operate is similar, the objective being to derive a model (or equivalently a set of rules) which will predict whether a customer will exhibit a particular type of performance or outcome (for example to purchase an unsecured loan product within the next four months, in response to a contact).
The model is developed empirically by considering groups of customers whose outcome performance is known, making use of the principle that there is no perception in the absence of contrast.
Some of these will have taken out loans in four-month outcome period being assessed, and some not. One seeks to identify what customers details, including foremostly aspects of their behavioural data in the 12 months (say) immediately prior to the beginning of the outcome period, could have led us best to discriminate between customers with, respectively, ‘good’ and ‘bad’ outcomes. Using one of the above or similar techniques, one would then establish a model that would combine the different items of predictive data into one summary score or indicator of likelihood, thereby generating the presumptive knowledge.
To ensure the model is of general use - and not simply fitted to the training dataset - one would then check on a similar set of data from the same period and see how well the outcomes (unseen by the model but of course already known) can be predicted on these cases.
The successfully validated model will provide a segmentation of the customer population by probability ranking, and would then be applied operationally on new cases to utilise this presumptive knowledge to improve the decisioning processes.
Actioning the Knowledge
The model, therefore, is a tool that will generate knowledge as to the likely requirements and performance of individual customers, by objectively quantifying the relevant factors in a decision, seek to predict the future rather than reflect the past and provide a measure for policy evaluation, which in turn facilitates better control.
To use this tool to best advantage, the organisation would require access to a computer-based (because of the volumes) system that can apply parameterised rules for decision taking to the whole of the customer portfolio. The rules should be able to be appraised before live implementation (so that their effects can be estimated), easily monitored, easily altered, and able to be applied in differing variants to various groups of customers concurrently, thereby allowing differential treatments to be implemented and tested.
At its most fundamental, the customer portfolio would be (randomly, as we will see below) segmented into different groups, the customers in each of which will be subject to a particular decision-making rule-set or strategy.
An example of a rule-set that might be applied to one segment within a group of customers would be:
IF the customer has been with the Bank for more than 2 years,
AND does not already hold the Bank’s credit card product,
AND is currently profitable above a certain threshold,
AND whose likelihood to default is less than x%,
AND whose propensity to respond to a mailing is more than y%,
THEN mail out a credit card offer.
Customers within the same group who have been with the Bank less than two years, or did not meet one or more of the other conditions might have a different treatment applied. Similarly, other groups of customers would have a different range of treatments applied to them.
The specifics of presumptive knowledge that might form part of the customer treatment strategies could include: profitability score, potential profitability score, segment indicator, product cross-sell propensities, retention propensities, risk score, channel response score, and so on; which would be applied across areas such as upselling or cross-selling to current customers, retention activity, customer acquisition, and arrears management and collections.
Monitoring and Refinement
To facilitate the subsequent monitoring analysis, it is important that each group be randomly selected to be representative of the whole of the customer population. Otherwise the selection of the group itself could influence the outcomes, which would overly complicate the subsequent feedback and assessment. Similarly, it is relevant to include a control group (again selected at random) for which no proactive action is undertaken. This will allow the analysis to control for the effects of external factors (for example advertising campaigns by competitors), as these would apply equally to all groups including the control. Appropriate conclusions would then be able to be drawn solely on the basis of the organizations’ strategies, as is desired.
Once the customers have been grouped in this way, the operational and financial results for each segment can then be simply monitored in order to decide on the most effective or profitable strategies. These can then be either expanded to cover a larger part of the portfolio or dispensed with, in an evolutionary manner of survival of the fittest (also known as the champion-challenger approach), where at any time the majority of customers would be subject to the champion strategy.
New challenger strategies will be continually devised, based on observations and analysis of performance, some of which might simply be using different parameters within current strategies.
A knowledge-based system in use will therefore have a number of treatments running in parallel on different groups of customers; with new proposals being tested, perhaps initially on small groups, before wider application as merited. By this means, the organisation will enter a virtuous circle where customer and related strategies are continuously being improved by a process of learning while doing.
The overall system can be depicted as in Figure 2.
Figure 2: Knowledge Based Adaptive Control
Benefits of Customer Knowledge Management
We can therefore see that effective use of presumptive knowledge for perceiving and meeting the needs of the customer in a high-volume business will allow the organisation to maintain income stream from customers, help avoid transfer of stewardship of customer, and maintain and increase share of wallet.
In addition it can facilitate loyalty programs and communications, and perhaps engender the becoming of ‘point of first call’ for new requirements, thereby assisting in the development of new products and services.
Finally, the increase in further business will provide additional customer-specific data, which in turn will enable better decision taking, thereby again reinforcing the positive feedback loop.
Business generation then becomes more efficient as well as more effective, leading to improving profitability overall.
Sam Korman is Head of Customer Analysis & Modelling at Abbey National Plc. He can be contacted at: