Next-best-offer

Drie artikelen van begin deze eeuw over de waarde van analyse en voorspelmodellen in het benutten van inkomende klantcontacten. In ‘Online Scoring’ wordt uitgelegd wat de mogelijkheden van voorspelmodellen zijn en hoe ze het beste kunnen worden ingepast in de toen gangbare outbound CRM benadering. ‘RealTime Interaction Management’ beschrijft mijn toenmalige visie op wat voorspellingsmodellen zouden kunnen betekenen in het benutten van klantinteracties en voeren van klantdialogen, over alle kanalen en media heen. En Gartner’s Spaarbeleg case is de vroege erkenning (2000) van onze unieke oplossing en van wat later een interessant verdienmodel bleek te zijn.

De door ons ontwikkelde oplossing is later bekend geworden als ‘RealTime Recommendations’ en ‘Next-Best-Offer’. Het werd het fundament onder de latere ‘RealTime Marketing’, ‘1-op-1 Marketing’, ‘Inbound Marketing’ en ‘Personalisatie’. Tegenwoordig heeft elke website en mobiele app wel een (zelf)lerend algoritme om individuele bezoekers te kunnen benaderen met de best passende propositie en het meest kansrijke aanbod. Specifiek de maatwerk software die we in het Spaarbeleg project gezamenlijk hebben ontwikkeld, is na de implementatie door Data Distilleries teruggekocht, en doorverkocht aan SPSS (Clementine) en later IBM (Modeler).

Onderstaande artikelen weerspiegelden mijn geloof in een toekomst met analytics en modellen/ algoritmen. De Spaarbeleg ervaring en deze artikelen waren voor mij het startpunt van een continue zoektocht naar het creëren van (klant)waarde met marketing analytics en voorspelmodellen.

Gartner's verhaal over de Spaarbeleg case

Spaarbeleg’s Contact Center Scores in Real Time (Gareth Herschel, maart 2000)

printversie: Gartner Spaarbeleg

By providing real-time customer information to contact
center agents, a Dutch bank is able to generate strong leads
and add value to customer interactions.

Many marketing organizations are familiar with the application of
customer scoring to the improvement of outbound telephone
sales and direct-marketing campaigns. The use of this technique
to achieve returns on inbound calls is gaining attention. One
company applying customer scoring to good effect is
Spaarbeleg, a Dutch financial services enterprise.

Problem: Spaarbeleg was faced with the challenge of increasing
sales to its customer base. Aware of the danger of oversaturating
its customers with unsolicited messages, and with most inbound
callers using automated channels, Spaarbeleg was looking for a
way to improve the conversion rate of inbound service calls to
generate new business.

Objective: To improve the conversion rate of inbound calls,
Spaarbeleg decided that its customer service representatives
(CSRs) needed a clearer indication of which customers were
good prospects for cross-selling. The return on investment (ROI)
would come from two primary areas:
• Increased sales volumes generated from more-effective
targeting of customers for cross-selling during the interaction;
Spaarbeleg estimates that a 0.1 percent increase in
successfully cross-sold inbound calls will result in a $4.4
million (NLG10 million) increase in sales
• More-profitable sales, as a result of identifying the optimal
product for cross-selling and eliminating the cost of outbound
communications or resellers.

Approach: Since the project’s goal was to provide CSRs with
better customer information, Spaarbeleg needed to perfect the
technical aspects of making customer insight available to CSRs,
as well as address the issues related to handling the CSRs’
know-how. Early in the project, the company decided to use realtime
systems to generate customer scores. This decision was
made because the alternative would be to batch score the entire
database, which would not be a scalable option as the number of
scoring models grew. With this decision made, the enterprise
began to evaluate data-mining vendors, with a focus on finding
one that had a vision that matched Spaarbeleg’s own. The
eventual choice was Data Distilleries (see Note 1).
Ensuring that a technical capability is present does not guarantee
that it will be used effectively. Thus, ensuring that a system
works well in the contact center environment is crucial to such a
project’s success. The skill set of the Spaarbeleg contact center
determined that the project could be run with the active
participation of the CSRs. Three organizational decisions were
made: 1) Use of the system by CSRs is voluntary, but an
incentive plan provides benefits for those successfully using it. 2)
The contact center manager can define how strong a prospect
must be before the CSR is prompted to cross-sell. During busy
call periods, only the best prospects are displayed, which limits
the impact cross-selling has on satisfying routine customer
inquiries. 3) The final element of the organizational effectiveness
of the project was the creation of an intuitive interface for the
CSR. In this case, a simple bar chart represents the customer’s
propensity score, and the data-mining models used can be easily
explained to the CSR.

Results: The project is in a pilot phase (see Note 2), with only
one product being scored; however, early results are promising.
The system is performing well, with a three-second response
time from the entry of the client’s account number to the display
(or not) of a cross-selling prompt. The flexible adoption of the
system into the operational life of the contact center means that
fewer than 20 percent of inbound calls generate a cross-sell
suggestion, and less than one-third of the suggestions are acted
upon by agents (who exercise their own judgment about the
prospects for cross-selling during that call). As a result of this
two-stage filtering system, more than 50 percent of prompts
generate a qualified lead to the sales force, and its conversion
rate is also around 50 percent. Overall, the project is currently
returning approximately 10 percent of the business case with
only one product (of the anticipated 15 to 20) being modeled and
offered to clients.

Bottom Line: Effective cross-selling requires the organization to
be able to generate and deliver accurate customer information to
the point of contact. The integration of customer analysis
systems with channels such as the contact center enables
significantly more-effective cross-selling when the customer
chooses to contact the bank. Spaarbeleg has demonstrated both
the feasibility of such a strategy and the significant business
benefits that can be gained.

RealTime Interaction Management

Gert Haanstra en Pieter W. Boelens, Defying the Limits, A Thought Leadership Project from Montgomery Research, Montgomery Research Inc., San Francisco, Volume 2, 2001, blz 145 t/m 148

printversie: 19 RealTime Interaction Management

Analytical Customer Relationship Management (aCRM) has become a vital link in creating, developing and maintaining long life customer relationships. Recent developments reveal that the potential of aCRM goes beyond its original analysis function. Looking for unexpected patterns in the database, explaining behaviour and producing predictive models has made way for direct and interactive use of the predictive customer information obtained. The new generation of aCRM is focused on making the dynamic interaction process with the individual customer predictable, with opportunities and threats being flagged up in real time and exploited immediately. The real-time exploitation of opportunities is achieved by generating the best propositions automatically, and distributing them via central control of the distribution channels. The development reveals that aCRM is even more the engine initiating interactions with customers, providing and promoting a personal response, everywhere and at all times.

A second interesting development is the change that marketing itself is experiencing. The marketing function is slowly changing from an “art” with a large measure of “gut feel”, into a controllable, manageable process which needs proven measurable results. This trend started with customer lifetime value marketing (CLTV) where companies opt for a clearly different approach to customers. The (expected) profitability of the customer and the (expected) lifespan of the customer relationship take pride of place with customer lifetime value marketing. CLTV marketing is designed to maximise customer relationships. It calculates the net cash value of the expected profit during the life of the customer or customer group. The company then tailors its investment per individual customer accordingly. In the process, the company seeks to achieve maximum profit at minimum investment.

Targeted approach

Companies are seeking to use analytical CRM as a means of making (outbound) marketing campaigns more effective. The first step is often to test out what is possible with the analytical CRM. What could analytical CRM mean for the company? Does it create money? If so, how much? What investment is entailed? Companies are accustomed by virtue of their profession to store large amounts of customer details. They understand that the data that they have stored on the various databases yield money. But the data only achieves real value when it is given meaning. ACRM is required for this “translation exercise” to enhance the effectiveness of marketing efforts, but also to discover new sales opportunities in the customer base. The database is examined for commercially interesting and surprising (customer) segments. The point is to achieve interesting commercial success in order to induce the company to further investigate the potential for analytical CRM.

The effectiveness of marketing-initiated (outbound) activities is enhanced by looking to see which customer best fits with a particular produce or service. Or to put it another way, finding the right customer and the right offering. The benefit of such score models is that the company can approach its customers in a much more targeted manner. In this way, costs are reduced and at the same time a larger volume of contracts is signed. Score models are drawn up for this purpose to predict who in all likelihood will purchase the specific product. The models are produced by comparing customers who give a smart response with those who do not respond. This enables customer characteristics to be identified that have the greatest predictive value for evoking a response. The same commercial result, or even a better result can be achieved using a score model, with significantly lower runs and/or contact frequency. Higher sales are therefore achieved, at lower cost. It also offers the major benefit that the customer is subjected less frequently to campaigns, which she wasn’t sitting waiting for.

Closed-loop Multi-Channel Marketing

The success of aCRM is enhanced if it becomes a structural part of the marketing and sales process. With the rise of new distribution channels such as the Internet and mobile phones, companies are expanding their distribution channels and integrating aCRM with the existing marketing and sales process. On the one hand to gain maximum benefit from the revenues by consistently using aCRM in the multi-channel marketing process. On the other, to be continuously evaluating and further refining forecasting models that have been developed. Integration ultimately leads to major improvements in the marketing and sales results.

“Closed loop marketing” is the aim. This entails ongoing improvements to the quality of the models, to the creation of more models for new target groups and for other products and earning higher sales volumes. Or, to put it another way, continuity in optimising the commercial process.

The right mix of customers, products and channels is sought with a view to coming into contact with the customer through a wide range of channels for integrating aCRM into the multi-channel concepts. Marketers will then no longer make open-ended use of score models for approaching client groups. Instead, for each campaign there will seek out the highest-scoring target group in the database. Their aim is to optimise campaigns as far as possible by using score models.

By making customer contacts structurally more effective, aCRM clearly adds greater value compared to operational applications (often already in place) such as campaign management, sales force automation, call centre and web applications). As the customers with the highest likelihood of conversion or response are identified, more and profitable products can be sold to the most profitable and loyal customers. The maximum result is obtained where there is seamless technical integration between the aCRM application and the existing infrastructure.

The predictive quality of existing models is enhanced by using them and testing them against old models. Ultimately, these cycles lead to a set of good-quality prediction models for most “standard” computerised campaigns. The ultimate result is a smooth-running campaign management process, ongoing monitoring of the quality of selections and effective distribution of customers among marketers’ campaigns. In short, an instrument for processing existing customers in an effective and efficient manner.

Example Closed-Loop Marketing

An example of a successful application of aCRM concerns a large financial institution which uses direct marketing as its main means of serving its eight million-plus customers.

Objective: to optimise the direct mail processes by means of an integrated aCRM solution.

Application: Data Distilleries software is used in combination with campaign management software. As a result of the integrated aCRM solution, database marketers are actually in charge of defining customer selections for direct mail campaigns. Data Distilleries software rapidly analyses the whole customer database and presents the results in the form of clear customer profiles. The solution offers database marketers the possibility of analysing customer data and incorporating the results of these analyses more rapidly in new campaigns.

How is aCRM used?
The bank uses DD/ Marketer, a special module for marketers, in order to analyse the whole customer database more quickly and to export the results with the associated scoring probabilities quickly and easily to the campaign database. The integration has made the building of mail models a simple process and one which can be carried out independently by the database marketers. In order to select the most promising customers from the database, the database marketers apply various types of analysis. For example, they can first carry out a trial mailing, and analyse the response to this mailing. The Data Distilleries software then generates a model which specifies which customers will show most interest in the bank’s financial products. Alternatively they might use the so-called propensity analysis. Here the software analyses product purchases by current customers, in order to produce a model for target group selection.

In the example of this Dutch bank:
Direct mail process before aCRM was used:
– the building of models took over 4 weeks,
– this was very labour-intensive and required effort on the part of several departments,
– only 30% of the mailings sent out were based on models,
– no customer knowledge was built up by the database marketing department.

Direct mail process after implementation of aCRM:
– the building of models takes only a few days,
– the customer data is always ready,
– the data can be used immediately in the campaign management software, in this case, Vantage from PrimeResponse,
– 70% of the mailings sent out are based on customer models, enabling more targeted mailing,
– the database marketing department learns more about its customers with each direct mail action.

The bank uses Data Distilleries software to distil usable customer knowledge from the enormous volume of available customer data. This results in a shortening of the database marketing process so more campaigns can be implemented on the basis of models by the same number of people. The benefits of this operation are obvious: a higher response, less irritation among customers and a reduction in waste. The large numbers of mailings which the bank sends out in the form of leaflets with statements, further reinforce this effect. If the waste can be reduced by just a few per cent, this will quickly lead to annual savings of many millions.

Real-Time Recommendations

The customer is increasingly taking the helm, specifically in the interactive distribution channels such as call centres, voice response units and the Internet. The customer or prospect is himself taking the initiative to contact the company. That is the ultimate moment for marketing. Real-Time Recommendations is a suitable approach to adopt here.

The concept remains the same whether one is dealing with a call centre or the Internet. At the time when the customer seeks contact with the company, existing knowledge of that customer must immediately lead to a personal, commercial proposition tailored by the company. In itself, not a problem. Selecting the best approach on the basis of a battery of scores would appear to be the solution. But what if the customer comes to make a complaint or order a product, making the proposed approach pointless? The customer is taking the initiative to make contact and will therefore first be asking something of the company. This information is essential to determining any commercial approach towards the customer. Pre-calculated scores on the basis of score models are insufficient. The real-time calculation of score models, with existing customer information along with information that has just been obtained (in real-time) is therefore a must; hence the term Real-Time Recommendations.

Real-Time Recommendations are highly lucrative. (See example) It is a simple matter to flag the best cross-selling and up-selling opportunities for inbound customer contacts and to use them directly, driven by underlying score models. Sales are directly and visibly increased, because the customer is given a product offering that best suits him, and he receives the product offering at the time when he is actually receptive to it. In addition, using Real-Time Recommendations focuses inbound marketing activities on the most profitable and most loyal customers in the database. Marketers can opt to sell the products with the highest margin. They are given the opportunity to give a higher priority to recommendations which will increase the profitability and loyalty of these most valuable customers.

Example: Real-Time Recommendations

A good example of a successful ROI is the case of a financial organisation which does not operate through intermediaries – a direct writer.

Objective: to exploit a percentage of the incoming customer contacts for cross-selling to promote the most profitable product.

Application: Data Distilleries software in the call centre. As soon as an attractive opportunity arises, the software gives a suggestion for a product to be offered by means of real-time recommendations . In the corner of the screen a separate small screen appears. This states the name of the product and the probability of success. With a double-click the employee gains an insight into the structure of the underlying model and a number of product-specific sales arguments, and he sees a possible opening sentence.

What was measured?
Prior to the actual measurement, the ‘autonomous’ response was measured over a certain period. This data made it possible to determine the extra returns of real-time recommendations . After all, the real return is equivalent to the real-time recommendations results minus the autonomous response. In addition, two groups of call centre agents were defined: one group used real-time recommendations , and the other engaged in cross-selling without the software. This structure made it possible to take account of the difference in the selling quality of the agents. Three sets of measurements were then taken:

1) The number of times an opportunity arose; how often the suggestion was presented; how many telephone calls were received for the agents connected to the system and how often the suggestion actually appeared above the threshold value.
2) The number of times the call centre agent took up the suggestion; how often he was presented with a suggestion and did nothing with it, and how often the agent tried to do so, but without success.
3) How often success was actually achieved. This figure was broken down into commercial successes (leads created, brochures sent out, information provided, a product sold other than the proposed product) and genuine success (actually selling the proposed product).

In our example of the direct writer:
– a suggestion was presented in 15% of the telephone calls,
– 30% of these were taken up by the agent in order to make an offer to the customer,
– 50% of these were commercially successful,
– 50% of these were successful sales of the proposed product.

The precise figures were:
– 50,000 telephone calls per month (via the connected agents),
– 7,500 telephone calls were potentially attractive for sales,
– agents made offers to customers on 2,250 occasions,
– 1,125 successes were achieved, and
– 563 contracts were concluded.
With a profit margin per contract of 500 Euro (spread over 20 years), this generated an additional profit contribution of more than 3.3 million Euros per year. And that was just for one product and a small group of approximately six call centre agents.

The software has a very favourable ROI which is relatively easy to achieve. Good results can be achieved in a very short period even with a limited number of products. In order to gain an even better picture of the ROI, the measurement can be regularly repeated for different products and across several channels.

Real-Time Analytical Interaction Management

The logical step after implementing Real-Time Recommendations is to use aCRM for real-time control of all marketing activities. Companies wish to get a better grip on the dynamic process of interaction with the individual customer (including the interactions initiated by the marketer himself). They want to be able to control and adjust the process rapidly and be better able to predict the results of the interactions so as to bring the objectives at customer level in line with corporate objectives. The aim is to actually maximise total yield and profit over the long term for each individual customer (Customer Life Time Value).

Real-Time Analytical Interaction Management allows a marketer to pursue a wide range of objectives, and define them in various personal, real-time interaction strategies with the customer. Targeted propositions aimed at the individual customer are developed for this purpose. These propositions cover a specific action in which a product or service is offered, expressed in a particular message. Various propositions are related to different corporate objectives (such as cross-selling, up-selling, cost reduction, retention, penetration and profitability). The interaction strategies are complementary.

Interaction strategies are defined, or may be defined covering a wide range of distribution channels (= multi-channel), contact moments in time (= multi-step), for various customer segments, in a combination of inbound and outbound contacts. Predictive models are used to do so. The models are easy to use and provide a stable prediction of customer behaviour, evolving over time, among the various distribution channels and in a rapidly changing environment.

Interaction strategies are implemented automatically. The situation is constantly being analysed to see whether there are opportunities and/or threats. On the basis of the complete (including the most recent) information, changes in the customer situation are flagged that provide good grounds for communicating a specific proposition. This information may emanate from the customer himself (if for example he calls, visits the website or responds to an offer). It is information stored in the databases that generate interesting events (such as a birthday, wedding day, Valentine’s day, etc.). It may also be information which is initiated by the company (defined interactions for direct mail, or a phone call to the customer).

In addition, companies are prepared to exploit opportunities/threats that may arise at any time and for various propositions and company objectives. This requires pro-active preparation for interaction with the customer. The best proposition for this specific customer is generated at the time of interaction. This means a dynamic weighing-up between various customer objectives such as cross-selling, up-selling, cost savings, loyalty and customer value. This is based on a consistent customer picture which changes (in real-time) to the new situations. Only then is it possible to exploit opportunities in time.

Furthermore, the propositions are constantly being disseminated by central control of the various distribution channels by automatically providing the best propositions, at the right time. Customer knowledge is collected and disseminated in turn among the various distribution channels from this central position. This calls for customer knowledge to be integrated from/in this central position. A clear understanding of all activities targeted at the customer enables every customer to be approached to best effect and consistently via each distribution channel required and at each desired moment. Only then is it possible to flag opportunities and threats in a timely manner from analysis and to make pro-active preparations at every interaction with the customer.

If supplier wishes to make the central position of CRM a reality, he must be able to collaborate. Apart from the functionality listed above, the application will have to integrate with any application, database or infrastructure. The source of the supplier determines the technical possibilities that can be achieved in this central position. The fact is that the technology could soon become outdated because the wrong architecture was adopted in the past. Technical developments are proceeding so quickly that incorrect choices at the heart of things are virtually beyond rectification. The technology for the CRM engine must in any event be easy to expand across all distribution channels, and easy to integrate with the existing infrastructure in line with generally accepted standards.

To achieve the objectives set, the results of the strategies are evaluated continuously in real-time, and the marketer is given the opportunity to adapt the strategies “on-the-fly”. How does one work towards achieving customer objectives?

Real-Time Monitoring and Real-Time Alerts are deployed to remain abreast of the result. Performance being achieved with the strategies is monitored by a number of pre-defined key variables. These mainly focus on the results during the period in which the strategy is active. The information is presented to the marketer in real-time, i.e. in (milli-)seconds by means of a monitor/dashboard. Various types of monitors may also be developed (for example for marketing and sales directors, marketing analysts, call centre managers, etc.) to monitor their specific area. To be able to flag (significant) deviations, the results are compared directly with the forecasts, expectations and/or objectives previously made.

Concrete modifications to ensure that the objectives are attained are proposed on the basis of historical data/results. An organisation can learn much from customer contacts. One condition is that the communication with the customer is logged as far as possible via a wide range of distribution channels, contact moments, for various customer segments and propositions. Analysis of the contact data shows which desired changes can be implemented by the marketer. A number of levels are available here: strategy, contact and model level. In the first instance, the strategy is modified. For example, the respective priorities between the strategies are changed, the influence of general business rules weakened/strengthened, or strategies halted or started. At contact level, distribution channels are removed or added, response options changed, more specific contacts deferred or advanced. At model level the definitions of the predictive models are altered. Generally by producing a new predictive model which forecasts who responds, with which offering, via which distribution channel and when. After implementing the changes, a new balance is achieved in communication with the customer, which takes account of previous propositions.

Conclusion

Analytical CRM is gaining a place for itself as the brains in the centre of any CRM business strategy. It is an essential link in establishing, developing and maintaining customer relationships by the real-time flagging of opportunities and threats, and directly exploiting these moments. Mainly by generating the best propositions automatically and distributing them by a central control of the distribution channels. Interactions with the client are initiated dynamically, dealt with and promoted, everywhere and at all times.

The central position is needed to achieve real competitive advantage with Customer Life Time Marketing. ACRM is needed to identify who the (potentially) profitable and loyal customers are. Furthermore, activities can be initiated with aCRM which are designed to maximise the customer’s lifetime value. This is achieved by anticipating expected customer behaviour by analysing historical customer behaviour and predicting future behaviour. But what the new generation of aCRM most achieves is that it puts the marketer (back) in control.

page 1:
page 2:
page 3:
page 4:

Online Scoring; missing link in Customer Relationship Management

Gert Haanstra en Melbert van Emmerik, Journal of Database Marketing, Henry Stewart Publications, Londen, Vol 7, number 3, 1999, pages 275 – 279

printversie: 2 Online scoring; a missing link in customer relationship management

In this paper we describe a new kind of decision support technique for cross selling clients in interactive media. This so called ‘online scoring’ predicts client behavior real-time as opposed to pre-scoring, and suggests the call centre agent to sell a specific product. In addition, we evaluate an implementation in a customer relationship management solution and provide a framework for implementing online score models.

Introduction
The combination of data mining and campaign management has rapidly emerged as a powerful technique in customer relationship management. However, one-to-one marketing is not possible/feasible yet because of the lack of interactivity based on information in the database. In a already advanced setting clients who contact a call centre or the internet get an offer based on profile information. Different customer profiles fit into different campaigns. When a client is identified the contact management system will match his profile to the different campaigns. The customer gets an offer associated with the campaign the best suits his profile. Up to this moment the offer is based on past interactions and knowledge available in the marketing database. But people search contact for a reason. He wants to withdraw money, files a complain or requests information. This very relevant information is not available at the time of the profile scoring process. So the in advance suggested campaign might not be the best move anymore. With online scoring not only the pre-available information will be the basis for profiling, but also the newly acquired information during the contact at present. Therefore online scoring is indispensable for optimising the effectivity of different client contacts and increasing the efficiency of using interactive distribution channels.

In essence, using online scoring means that the communication with clients will move from a monologue to a dialogue. If necessary information about a client is available, an effective dialogue can be established. Such a dialogue means that the client will get insight in one’s production process, and will get the feeling he can influence it directly. Result will be an increase of customer satisfaction, en strongly related to that, a long time relationship with your clients. Therefore using online scoring gives one the opportunity to acquire customer effectively, and building successful relationships with most profitable customers, and to maintain their loyalty and profitability.

On-line scoring and CRM

Generally, customer relationship management is about building close relationships with current and potential customers. Companies try to know their (potential) customers, and want to understand what their customers want. As a result companies can offer products or services that best match the needs of their clients. Data mining is used for understanding customer behaviour by analysing data, identifying ways to serve customers better by testing and evaluating a range of possible (personalised) campaigns. As best they can, companies try to satisfy their needs through personalised service. As a result, they expect to earn loyalty and a large share of their customers’ share of wallet. Campaign management is used to develop, execute, monitor and evaluate customer campaigns in order to encourage customers to develop a better relationship with the company.

Online scoring has the potential to immediately identify and use new cross sell opportunities based on predefined business goals and model definitions. It has therefore additional advantages compared to existing customer relation management solutions. Advantages of this method include the ability to (a) implement predictive modelling in interactive media (e.g., call centre, internet, voice-response media), (b) increase convergence rate by presenting most prioritised product offer, (c) focus on business objectives evoked by relevant score models, and (d) directly relate decision outcomes to business objectives.

An example
If the client uses the internet, he will receive a suggestion on what to buy. If he contacts a call centre, the agent will get a signal that suggests him what to sell. The suggestion for the agent consists of the product name (of the most likely product the client will buy), an indication of the expected success ranging from 0% (low) to 100% (high). In addition, the agent will be shown the reasons why the client would be interested in the suggested product (insight in the model). An agent decides what to do with the suggestion and also can use a predefined script and a description of the key features of the product.

More specific, a client contacts his bank. The moment the agent identifies the client, a first suggestion is calculated and presented immediately based on the current situation of the client. Then the new information acquired during the conversation is incorporated. If somebody withdraws his money from his saving account, obviously the score model should provide a different score compared to the case where the same person deposits money into his account. Therefore, during the conversation with the client, the models should be recalculated and the most appropriate action should be presented online within several seconds. So, by using on-line score models even the latest changes in the clients (financial) situation, his behaviour and preferences are taken into account when offering a new product to the client.

Of course not everybody is evenly suited for action. The best campaign for a specific customer might not have a very high success expectation. “Bad leads” for call-center agents should be avoided. In the end they might sabotage your system. Even worse, you don’t want to bother your customers with all sorts of offers all the time when you don’t expect them to respond. So in addition to scoring, you will want to set thresholds to scores below which no suggestions will be made. An experienced agent will get suggestions for offers with an expected change of success above 20%, whereas less experience agents only get suggestions for the most potential offers.

Development of online scoring

Online scoring is a result of the way data mining has developed itself in the marketing & sales process over the years. The ‘Exploring phase’ is used to find data mining opportunities, to learn about the processes involved, and to overcome the initial scepticism in using data mining by showing very impressive results. Companies become aware of the profits it can generate. They do some ad hoc research in order to find out about costs, profits, and expected organisational changes. Based on the results they will decide whether data mining can increase the effectivity of marketing and, in addition, if data mining is useful for finding new opportunities/segments in the database.

The next step is to bring continuity in using data mining; continuity in developing, using and evaluating score models. As a result, structural improvements of campaigns are established which provide companies a maximum of gains. This will often lead to building data warehouses (DWH) to offer quick and consistent input for data mining. In this stage all effort will be put into optimising the process of using data mining.

On the other hand, companies can prefer to enlarge the use of data mining for application to other fields instead of bringing continuity in their processes first. In this enlargement phase, on more fields results are gathered, which will prove that data mining has many opportunities.

As the process of data preparations is very time consuming, organisations choose ideally first the stage of optimising before they think about enlarging the possibilities of data mining. But sometimes the stage of enlargement will have to be preferred to get broader attention and hence more resources. All relevant information has to be gathered from operational databases, then data will have to be cleaned and merged. After that, new information will be derived through data manipulation and finally many fields will have to be recoded. Then the analysis file is ready for use. After the data mining process you will have to translate the outcomes into business rules, score models or just plain scores. This will then be used for different kinds of campaigns.

Either way, the next phase is to integrate data mining into the daily marketing & sales process (’Integration phase’). Therefore, (on-line) score models are brought to places where the cross selling actually takes place. E.g. marketers will use them in selecting prospects or selecting campaigns that best fit their customer’s needs. Call centre agents will use them in their interaction with customers. So, decisions about what to offer to whom, via what channel and when, will be primarily based on data mining results.

Implementing online scoring in a CRM environment

Most companies have the operational data stored on a platform separated from their data warehouse. The reason lays in the fact the data warehouses are primarily used to offer raw, derived and summary data for analysis and reporting. By offering people this information on a separate platform, you don’t have to concern yourself about the performance of operational databases. Beside the information did not have to be up-to-moment and therefore updates could easily be run through batch processes. When implementing online scoring this construction leaves us a challenge. A predictor in a certain model might be the ‘number of deposits a customers has made during the last three months’. This predictor has been calculated by counting the number of deposits in that time frame from operational data. A change in operational data during an interaction could not be input for online scoring during that interaction with the regular up-date routines. An option could be that you don’t use aggregated operational data that could change during the interaction session. This would largely reduce the amount of predictors that you could use in the score models. Another solution is to directly tap the newly acquired data into the data mining system so you could use it for recalculating the predictors and the models.

For building a score model, the following functionalities are needed in a data mining tool:
• ‘data access’ for flexible and dynamically updating tables from the data warehouse or/and transactional databases,
• ‘data preparation’ for aggregating data tables and creating user defined tables and variables,
• ‘behavioural scoring’ for creating and using score models for prospect selections,

Score models are transferred back into the DWH, and used in the campaign management for prospect selections.

Additionally, a campaign manager should have the following functionalities:
• ‘event triggering’ for detecting client life-cycle events,
• ‘targeting system’ for choosing / evaluating the best way to contact a client,
• ‘scheduling’ for atomising prospect selection,
• ‘channel management’ for distributing and co-ordinating several, multi-channel campaigns,
• ‘decision support’ for prioritising the use of score models and the use of distribution channels,
• ‘reporting’ for determining (financial) consequences of several campaigns and its influence on customer value.

Results of the campaigns are distributed to different channels. In a multi-channel situation clients can be contacted:
• Directly (e.g. mailings, door-drops, coupons),
• Differently (e.g. campaign A for client X and campaign B for client Z)
• Personally (e.g. franchise, brokers, retail),
• Massively (e.g. radio/tv-commercials, newspapers),
• Electronically (e.g. telephone, voice-response, sms),
• Interactively (e.g. e-mail, internet).

The client scores and campaigns for outbound contacts (mail, leads, calls) are calculated in batch processing. In batch, a scheduler is running jobs on historical data from the data warehouse. Automatically, the system will allocate customers to different campaigns.

For inbound contacts we use online data processing and online scoring, which means the data is up-to-moment and the scoring is done real time. In order to get the best prediction, actual information from the conversation taken place should be incorporated in the scoring. Therefore, if somebody withdraws money from his saving account, data should be transferred immediately to environment where the scoring takes place. And the scoring is done, starting the moment a client is identified.

Opportunities for on-line scoring

Expanding the possibilities of online scoring will contribute to building close relationships with your best customers. Key feature to success remains identifying and immediately using new opportunities.
• Acquiring new business: Instead of identifying a current client by his client-number stored in the DWH, a potential customer can be identified by his postal-coding. One can predict what product to suggest, based on geographical information.
• Customer selection: For example, models for credit scoring can be implemented for selecting customers that suit the product characteristics, instead of offering a product to everyone.
• Customer extension: In order to increase the loyalty and the profitability of your customers, the most profitable customers can be identified. For example; these customers could get a VIP-treatment from your call centre agents.
• Customer retention: Optimising payment collection, solving complaints, or tracing possible quitters are part of monitoring subscriber loyalty for keeping the customers for as long as possible.

Conclusion

Knowing what to sell the moment a (potential) client contacts you. In interactive media, the time between knowing what to do and the actual selling has to reduce from several days, or even weeks, to seconds. Online scoring provides such a solution, and is therefore indispensable for optimising the effectivity of different client contacts and increasing the efficiency of using interactive distribution channels. In addition, it has the opportunity in supporting building successful relationships with profitable customers, and to maintain their loyalty and profitability.

pro actively.

 

page 1:
page 2:
page 3:
page 4:
page 5:

‘Database Marketing bij Spaarbeleg zal niet meer hetzelfde zijn.’

Interview met Gert Haanstra, InterAegon, Den Haag, jaargang 2000, nummer 2

printversie: 4 Database Marketing bij Spaarbeleg zal niet meer hetzelfde zijn

Het afgelopen jaar is er bij Spaarbeleg hard gewerkt aan het project Octopus. Een uniek project waarmee een belangrijke en noodzakelijke stap is gezet in de verdere ontwikkeling van (database) marketing bij Spaarbeleg. ‘We hebben moderne marketing technieken als datamining en campagne management, volledig geïntegreerd in ons marketing- en verkooproces. Een behoorlijke klus die maar weinig bedrijven zo goed kunnen doen.’, zegt Gert Haanstra projectleider van Octopus. Inclusief voorbereiding heeft het project anderhalf jaar geduurd. In die periode werden de plannen gemaakt, werden de twee leveranciers geselecteerd en de systemen geïnstalleerd. Vervolgens moesten de mensen ermee leren werken, en konden de eerste resultaten worden behaald. ‘De eerste resultaten zijn boven verwachting en beloven veel voor de toekomst’, aldus Haanstra.

Met de nieuwe technieken wordt optimaal gebruik gemaakt van de mogelijkheden die Spaarbeleg heeft. Bijvoorbeeld, een grote database waarin een hoop gegevens over de klant liggen opgeslagen. En natuurlijk de verschillende distributiekanalen waarover Spaarbeleg beschikt. Spaarbeleg benadert de markt zowel via massa communicatie, zoals met de radio en tv commercials, als ook rechtstreeks. Dit laatste gebeurt veelal met gepersonaliseerde mailings als ook met de bekende huis-aan-huis brochures. Bovendien werkt Spaarbeleg met een eigen franchiseformule (de Spaarbeleggers), en doet zij zaken met het onafhankelijke intermediair. Verder is er een eigen call centre in huis met een aparte verkoopafdeling, en is er een druk bezochte internet-site.

De filosofie achter Octopus sluit prima aan op het streven van Spaarbeleg om enerzijds meer te verkopen aan bestaande klanten en, anderzijds, om de beschikbare distributiekanalen nog beter te benutten. Een belangrijk argument hierbij zijn de verkoopkosten. Een product verkopen aan een nieuwe klant is veelal duurder dan een product verkopen aan een bestaande klant. Bovendien kunnen we veel makkelijker leren van onze bestaande klanten. Door goed te kijken naar wie welke producten koopt, is het beter te achterhalen welk product we de klant moeten aanbieden. Dit geldt ook voor de distributiekanalen. Sommige klanten doen liever zaken via de telefoon, anderen reageren sneller op een advertentie. Door hiermee rekening te houden, worden de distributiekanalen beter benut. En dus worden bepaalde groepen klanten nagebeld, en krijgen andere groepen klanten een aanbod te worden bezocht door een intermediair of franchiser. ‘Uiteindelijk wordt de klant hier alleen maar beter van. We leren veel van onze klanten. En die informatie gebruiken we om betere aanbiedingen te kunnen doen. Deze tools stellen ons in staat om dat op een zo goed mogelijke wijze te doen.’

Maar wat gaat er veranderen op gebied van database marketing?
‘Alles draait om het effectiever maken van onze verkoopinstrumenten, en het beter op elkaar aan laten sluiten van de verschillende distributiekanalen. Een mooi voorbeeld is het call center. Vroeger belde een klant voor een overboeking en moest de call center medewerker zelf achterhalen of deze klant interesse had in een bepaald product. Nu voorspelt de database welke aanbieding we de klant moeten doen. Op moment dat de klant belt, wordt direct berekend welk aanbod geschikt voor hem/haar is. Hierbij wordt ook rekening gehouden met recente aanbiedingen richting de klant. De naam van het product wordt aan de call center medewerker gepresenteerd. De medewerker neemt het mee in het verkoopgesprek. Voorzover mij bekend is dit nooit eerder gedaan. En daarmee hebben we met Octopus opnieuw een mooie voorsprong op de concurrentie.’

Hoe zijn jullie tot de conclusie gekomen dat het anders moest?
‘Het is een verdere ontwikkeling in databasemarketing. Spaarbeleg heeft zeer frequent en op allerlei manieren contact met haar klanten. Als je wilt dat iedereen die namens Spaarbeleg in contact komt met de klant kan profiteren van de informatie in de database, is dit een logische stap. In het verleden hadden we veel tijd nodig om de informatie uit de database te ‘vertalen’ naar concrete campagnes. Door er een automatisch proces van te maken, en er zoveel mogelijk distributiekanalen op aan te sluiten, hebben we er maximaal profijt van. Los daarvan krijgen we de mogelijkheid om de relatie met de klant te verdiepen. Duidelijk is dat databasemarketing bij Spaarbeleg nooit meer hetzelfde zal zijn.‘

Maar wat betekent dit concreet, bijvoorbeeld voor het onafhankelijke intermediair?
‘Op korte termijn zal er weinig veranderen. Op lange termijn is het de bedoeling dat het intermediair toegang krijgt tot deze verkoopsuggesties. Zodat ook zij direct kunnen profiteren van de Spaarbeleg database. De komende periode werken we dit uit. Overigens zijn er nu al intermediairs die vergelijkbare technieken kopen. En dus op eenzelfde manier hun klanten willen gaan bedienen als Spaarbeleg.’

En wat brengt de toekomst?
‘Zo’n investering is natuurlijk bedoeld om meer geld te verdienen. Het systeem biedt daarvoor vele mogelijkheden. Het komend jaar willen we bijvoorbeeld mensen direct inzicht geven in de verwachte kosten en verwachte opbrengsten van hun beslissingen. Zodat we drie duidelijke doelstellingen kunnen realiseren; meer omzet, meer winst en minder kosten.‘

"“Worldly wisdom teaches that is it is better for reputation to fail conventionally than to succeed unconventionally.” (Richard Thaler)