Multi-channel Adaptive Context-sensitive Intermediation: Reconciling Marketing goals and Users needs Vincent Chevrin, José Rouillard, Alain Derycke, Sébastien Sockeel Trigone Laboratory Bâtiment B6 – CUEEP – 59655 Villeneuve d’Ascq FRANCE [email protected]; {alain.derycke,jose.rouillard,sebastien.sockeel}@univ-lille1.fr

Abstract In this paper, we show how we built a bridge between Human Computer Interaction (HCI) and Direct Marketing (DM). Here we address some major issues in the area of the Multichannel Adaptive Context-sensitive Intermediation and tackle them in our software architecture. We present a prototype based on this architecture. But our main objective, here, is to show a major issue in the field of Multi-channel Adaptive Context-sensitive systems, which is the management of different –and sometimes conflicting– contexts (e.g., from the user i.e. context-aware for HCI community and from the organisation, its policies, rules, etc.).

1

Introduction

In collaboration with 3 Suisses International1, a large group of e-Commerce, we have been studying multi-channel area from the perspectives of both Human-Computer Interaction (HCI) and Direct Marketing (DM). We started working in this domain four years ago, and now we have some results we believe can be of use for the two communities. First, we must say that multi-channel is a new concept in the HCI community. But it is not the case in DM area. Indeed, Multi-Channel interaction with customers is an old technique used by remote sales organisations to address the largest number of people possible. However, the rapid growth of new mobile devices with new modes of communication2 brings further opportunities for these professionals. In fact, this would permit to develop other means to address more customers and/or prospects. There are three main communication channels in marketing: (1) Direct sale: the organisation has a physical contact with the customer - via a vendor for example; (2) Indirect sale: the organisation sells its products and/or services via an intermediary; (3) The remote sale: customers are not in physical contact with the organisation, and they use a means such as phone, mail, e-Mail, etc. to order the products and/or services. Obviously, this definition is too vague to be useful. So, a bridge must be built between HCI and DM. This paper deals with the presentation of several issues related to the management of multi-channel interactions from our experience in direct marketing. 1 2

In this paper we will write “our partner”. For example: SMS (Short Message Service).

The term “channel” being ambiguous even inside the DM area, after reading available literature and discussing with our partner, we decided to propose a definition for the term channel that fits our purpose. The articulation of our study will be as follows: - Firstly, we will see how we have built a bridge between HCI and DM; - Secondly, we will present the main issues of multi-channel interaction from our DM experience. We will begin to present a multi-channel scenario in the DM area; - Thirdly, we will present our solution to solve the issues related to the multi-channel interaction management and discuss some related work; - Lastly, we will conclude this subject by presenting our perspectives.

2

A Bridge between Human-Computer Interaction and Direct Marketing

To build a bridge between HCI and DM, we had to work on several domains. For more details on this, see [Chevrin 2006]. Our first observation is that DM cannot be reduced to a simple exchange of information, goods or services between a customer and an organisation. Most people forget on one side all the customer knowledge methodology and on the other side all the mechanisms for treating the customers claims. So, it is important to stay user-centred. We can say that Direct Marketing must be Relational3, where the notion of personalisation is highly concerned (for examples of one-to-one strategies, see [Peppers & Rogers 1997] or [Peppers & Rogers 1999]). From this point of view, the knowledge management (work on the information from the customers, via data mining process, etc.) is important and must be taken into account. In our architecture (Figure 1), we can see that, for us, context influences the interaction. We will see in the section 3.1 that the context can be seen from two points of view: - The organisation side (organisation rules, customers knowledge management, etc.); - The user side (context-aware4 in HCI); These two points of view need a “cross-fertilisation”. We will speak about that later in this paper. We cannot use the term multi-channel without having a clear definition of what a channel is. To explain this point, we constructed a taxonomy for our problem space [Chevrin and al. 2005b]. Furthermore, we defined “channel” from a global point of view as well as from the HCI community and DM professionals perspectives. Such definition can also be easily transported to other domains, such as e-Learning [Chevrin and al. 2006]. Thus for us an interaction channel with an organisation is the union of: -

-

3 4

A means for providing information, documents, tangible products, services, etc. to a specific location (fixed, mobile), having specific features (routing delay, temporality, Quality of Service (QoS) such as availability, fidelity, costs, etc.); and An interaction device with the user (customer, citizen, learner, etc.) and having “modals” properties (using of natural language or not, direct action, etc.), temporal properties, and some “affordance” (semiotic and metaphoric aspects, use schema or guidelines).

In comparison with the Relational Marketing. We consider the definition of [Dey and al. 1999].

Then, our definition can be used to formalize a particular cellular phone with its network, a traditional (physical) market, etc. This is one of the elements constituting the bridge between HCI and DM.

3

The main issues of multi-channel interaction from our Direct Marketing experience

In order to illustrate what a multi-channel interaction is, in the area of remote sale, we can take a simple scenario: a customer searches some goods via an e-Commerce web site. She/He fills her/his virtual cart with simple mouse clicks. At the moment to pay, a power failure interrupts the interaction between the customer and the organisation. For the organisation, this rupture inhibits a potential sale. Retrieving from the previous coherent state of the transaction is an important challenge. Any available channel can be used to achieve this goal. That will avoid customer's frustration. In our scenario, the customer telephones to the organisation call centre to complete her/his order. An employee picks up this call and with the organisation’s Computer Telephony Integration (CTI), he/she knows who the customer is. Here, the system must also feed the employee with information like last operations of the customer, in the present case the contents in the virtual cart. Thus, the employee can propose the customer to pay her/his order. This scenario allows pointing three main issues to multi-channel interaction: (1) The first one, relative to a poor integration of the different channels into a common “infostructure”, an engineering problem in order to provide a good articulation of the channels inside a same transaction [Derycke et al. 2003]; (2) The second one, relative to the quality of the relationship with the customer, considered from the marketing viewpoint. Especially it appears that, during a business transaction which can last a long time, the commutation between the different channels (i.e. from Web to phone) can lead to a lost of the business opportunity for the seller and to a frustration for the client. In HCI words, the system is not “Seamless” [Derycke and al. 2003]. (3) The last one, which translates the problem of the dependence between the channel used and the services composition (we will explain that in the section 3.2) [Chevrin and al. 2004]. The section 3.1 of this paper deals with the two first points and the section 3.2 deals with the last point. 3.1 Integration of any channels for seamless interaction We have seen that the use of multiple channels is natural and well adapted for the remote sale. The appearance of new devices only accentuates that. Besides, channels do not appear at the same time and organisations implement them incrementally without a deep concern for integration issues. Thus, there is no communication between the different channels. Moreover, data must be well treated to become a kind of knowledge useful for the organisation (this is the area of Customer Relationship Management (CRM), etc). This aspect is not going to be exploited in this paper (for more on this subject, see [Holsapple & Singh 2000]). So, it is necessary to redesign information systems and integrate communication channels seamlessly.

Moreover, this integration must take into account the issue of the rupture in the interaction, as for example, due to a network failure, etc. As we said in the last section (point 2), this can do harm to the organisation. With our prototype –to be presented in section 4–, and based on our software architecture –shown in Figure 1–, we tackle this issue. For the moment, the only way of re-establishing the lost interaction depends on the customer’s initiative of re-contacting the organisation. We sustain that this is a task that should be best done by the system itself, by detecting the rupture and calling automatically the customer, using the best adapted channel taking into account the contexts. Here, there can be, as we said, incompatibility between the context of the user (HCI community), and the context of the customer (cf. organisation). Indeed, if the rupture arrived during the payment phase, the user would try to repair the interaction from where it has crashed, so after having filled her/his cart. However, in a real case, the organisation can not always guarantee the stock of the different goods taken into the cart by the customer.

Figure 1: Synthetic view of our software architecture

Today, large DM groups expect to develop in the future this kind of architecture (Figure 1). We can refer to [Siebel], which offers multi-channel CRM solutions to the organisations. There are also other solutions such as multi-channel portals, etc. We can see in Figure 1 that we take account of two kinds of context, as we said in section 2: - The context provided (directly or not) by the user and used by the organisation in order to adapt the different services; - The “policies” of the organisation, provided by the organisation. These two kinds of context are not always compatible. For example, if a customer has the habit of using services via her/his cellular phone, and the organisation does not deliver this service via this channel, the two contexts are not compatible. So, our main interrogation is: how to manage these two contexts at the same time and in a coherent way?

3.2 The inter-dependence between service composition and channels We will focus, now, more precisely on the nature of the dependence of service composition in relation to the channels used during the interaction. The collaboration with our partner was carried out through discussions with DM and eCommerce professionals as well as through empirical studies of several areas [Chevrin 2006]. A surprising report of this work then appeared: the execution process of the different eServices of the organisation is closely related of the channel used during the interaction. For example, processes are different on a vocal server and on the web. In [Chevrin and al. 2006] we have shown several scenarios where the composition of different e-services is conditioned by the nature of the channel and their interaction styles (direct manipulation versus speech dialogue). Indeed from our previous evaluation of first prototypes [Derycke and al. 2003], [Chevrin and al. 2005b], combining different interaction channels, we have discovered that the concurrent accesses to several e-services conducted to different interleaving of the e-Services primitives, depending of the channel in use. For example, reading and choosing from a table of items using a voice-guided interaction is different from a web interaction. In the former case we cannot force the user to remember all the choices made and on the latter case it is possible to see the whole table on the screen. This is, the fragmentation of the source document (for example a form) must be adapted to the channel characteristics and to the interaction elements (such as widgets, vocal forms, etc.). Moreover, we agree in affirming that the customer should be able to choose the channel that she/he wants and to execute the service that she/he wants. Obviously, this is only possible if the service is available for the selected channel. In the field of software engineering, with the Service-Oriented Architecture (SOA) and in the business field [Rust & Kannan 2003], we can show that the Service-Oriented Architecture is also a good framework to solve this issue. We introduced the Interactive e-Service (e-SI) concept in [Chevrin and al. 2005b]. For two years we have gone a little bit further in our research and a PhD have been started on the subject in Tunisia5 under our joint management. These work deal with the formal specification of e-SI and their composition [Hadjouni et al. 2005]. Moreover, the recent literature is quite rich in this field (e.g. [Benatallah et al. 2002]). So, our model-driven architecture builds, so as to say, a bridge between three domains: - Web-Services (the e-SI side we can see in Figure 1); - Relational Direct Marketing (the policies, rules, in Figure 1); - HCI, for context-aware aspect (see Figure 1); Our architecture takes into account the adaptation of the different services according to the context (and we point again the problem of the “cross-fertilisation” of our two kinds of context).

5

INSAT (Institut National des Sciences Appliquées et de Technologie)

4

The Technological implementation of our solution in e-Commerce area

As we said, we worked the conceptual basis before implementing a solution. We have built a theoretical framework [Chevrin and al. 2003], proposed a software architecture [Chevrin and al. 2005b], [Chevrin and al. 2006], and proposed the bridge between HCI and DM. We claim that our prototype asserts the technological feasibility of the functionalities of our software architecture. We simulate a system of remote sales via Web, WAP or/and phone channels. For the channel adaptation (presentation), we use PlasticML [Rouillard 2003]. This software is plugged into the Interface Agent. Figure 2 shows the global view of our prototype.

Legend: PA: Proxy Agent; DA: Display Agent; LA: Link Agent; SA: Session Agent; UA: User Agent; CRMA: CRM Agent; RA: Rules Agent; GCA: Garbage Collector Agent; GCAa: Garbage Collector Agent awake; SAx: Service X; IA: Interface Agent. Figure 2 : Global view (not formal) of the prototype structure

According to Figure 2: - There are multiple agents DA, SA and UA during an interaction; - UM means UbiMessage, it is a message type we have created to enable the communication between agents. This kind of message inherits from the basic properties of JADE [JADE] to send messages. 4.1 The multi-agents technology selected, JADE, for our implementation We can now present and justify the use of a Multi-Agents system (MAS), JADE [JADE], as the underlying layer for the implementation of our prototype. JADE is provided by a consortium of companies especially from the telecommunication field, and it has been used in the design and deployment of large projects implying the use of mobiles devices. It is possible even to push an agent inside a Smartphone. It has already been successfully used for the dynamic orchestration of web services and for the processing of the Human Computer Interaction for both multimodality and dynamic adaptation of contents to the user devices. Another argument for the use of JADE, as our MAS, is the underlying technologies like Java programming and communication mechanisms that are compliant with the FIPA standard proposal. This means that in the future, the communication will be easier between our infrastructure and other FIPA MAS developed systems, for example dedicated to the building by inference of some context knowledge (useful if we connect our system to a CRM). This is one of the reasons, combined with an e-Services approach, of the openness of our proposal. More explanation on this subject is available in [Chevrin and al. 2006]. Our prototype has a large number of different agents. Each agent has a particular functionality. For example, the Rule Agent manage the composition of the different e-SI; the Proxy Agent manages the junction between the middleware MAS and a servlet (then the Tomcat application server), etc. With this prototype, we support some functionality about multi-channel interaction and present two of them in the next sections. Moreover, the problem of the knowledge and the two kinds of context management is open. So the MAS solution enables us to have a stronger basis to implement a generic solution. Then, every implementation that we produce is upgradeable relatively easily, with the large possibilities given by the MAS technology. 4.2 The e-SI composition management It is the Rules Agent that manages e-SI composition. This agent requests the e-SI to run according to the e-SI already run. The Interface Agent transforms a XML flow, sent by the different Service Agents into an understandable language for the target channels (XHTML, VoiceXML, and WML). Each e-SI is represented by a software agent. When the e-SI is called for the first time, it sends a XML page to the Interface Agent. When the e-SI is called again and it detects the end of this e-SI, it warns the Session Agent (then, an e-SI already called won’t be called again if there is a rupture in the interaction). The Figure 3 shows an example of workflow introduced in our Rules Agent. This kind of composition is dynamic at runtime. If the user changes the channel, the system changes automatically the workflow of e-SI execution.

Figure 3 : Activity diagram (UML) of the different e-SI

4.3 Integrating management and avoiding rupture of the interaction This issue is managed via three main points: - The intrinsic properties of agents (the persistence of data, etc.); - The Session Agent that we have implemented; this agent allows the management of each interaction between a customer and an organisation. Then if a customer interacts with an organisation using two different channels, this user will be associated to two Session Agents, one for each channel; - The User Agent; this agent represents a user. There is a User Agent for each user connected to the system. When there is a rupture in the interaction, for example, a network failure, the User Agent and the Session Agent keep the useful information to repair the interaction. When the customer logins on the system, this repairs the interaction and routes directly the customer on the last eSI not completed.

5

Related Works

There are some researches with some relations to our approach. The MAIS (Multichannel Adaptative Information System) project [MAIS] is contemporary with ours (end in November 2005). This project is particularly oriented to networks. It studies adaptivity at all levels in the information systems, from the application to the networks and devices. Several levels of adaptivity are considered and particularly the needs of multi-channel information systems and the systems for the handicapped users are addressed. The Human-System Interaction Container (HIC) proposed by Thales [THALES]. This group is a leader in aeronautics, security and defence systems. This approach is also contemporary with ours. Through the adoption of a user centred approach for the management of the interaction, HCI is, according to Célestin Sedogbo [Sedogbo 2004] the future of the HCI systems, where the users will collaborate and interact via a heterogenic whole of devices and modalities. According to this author, the HIC facilitates the designers’ work: “In terms of software development and management, the HIC as an interaction middleware offers the HCI designers an open framework enabling them to design an interaction system that meets business requirements such as enhanced user support and easy integration of new devices, without implying important changes to the application, which is a major requirement of THALES system designers and developers.” Lastly, recent projects, such as AMACONT [Hinz & Zoltan 2004], in the Web Application area are also relatively close to our approach from the e-SI point of view. In their projects, they propose a step by step electronic document generation via a pipeline such as the Cocoon framework [Cocoon] in the Apache suite. Nevertheless, their notion of device does not take into account nature of the different networks.

6

Conclusion and perspectives

During the last four years, we have worked in collaboration with a large group of eCommerce. Naturally, we acquired knowledge and competence in this area. Our goals were in a first time to build a bridge between HCI and Direct Marketing. We began to identify common problems [Derycke and al. 2003]. Then we have elaborated a common vocabulary, by building a theoretical framework [Chevrin and al. 2003], and a taxonomy of the problem space [Chevrin and al. 2005b] (we have also given a definition of a channel that we can see in section 2 of this paper). Finally, we proposed a software architecture [Chevrin and al 2005a], [Chevrin and al 2005b] to address the main issues for a multi-channel context-sensitive system, three of which we detailed in this paper: (1) To build a bridge between HCI and Relational Direct Marketing (our architecture can manage other domains, such as e-Learning [Chevrin and al. 2006]) and the management of context from the organisation (rules, policies, law, etc.); (2) To use (adaptive) Web-Services in an intermediation with the multi-channel management (we said that channels and services are inter-dependent, and this is really a central point for us); (3) The management of context (context-aware for the HCI community). The two contexts allow adapting the different services for a good presentation to the user. However, we have seen that a large problem emerges from the management of these two

kinds of context at the same time. Indeed, there is often incompatibility between the two contexts. So, our main question is: how to bring a “cross-fertilisation” between context-aware (3) and policies of the organisation (1)? It is an open and important issue, but to which we do not have solution for the moment.

7

Acknowledgments

Authors want to thank Yves Bayart, e-Commerce Director of 3 Suisses International group, the Cité Numérique for its support in these research works. They also want to thank the MIAOU program (Nord Pas-de-Calais Region) and the UE funds (FEDER) for their partial financial support, and also Cesar Moura for his help.

8

References

[Benatallah et al. 2002] Benatallah, B., Dumas, M., Fauvet, M.C., Rabhi, F. Towards Patterns of Web Services Composition, In Patterns and Skeletons for Parallel and Distributed Computing, Editors: F.A. Rabhi and S. Gorlatch, Springer, 2002. [Benford and al. 1993] Benford, S. D., Prinz, W., Mariani, J., Rodden, T. R., Navarro, L., Bignoli, E., Brown, C. G. and Naslund, T., MOCCA - A Distributed Environment for Collaboration, in Proc. Telepresence'93, Lille, France, March 1993. [Chevrin 2006] Chevrin, V. Etudes Techniques des Diverses Approches pour le e-Commerce Multicanal : Premières Recommandations en vue d'établir un Système d'Intermédiation Généralisé. Working paper, 3 Suisses International. 2006. [Chevrin and al. 2006] Chevrin, V., Derycke, A., Rouillard, J. Project UBI-Learn: an Intermediation Infrastructure Multi-channel Access to Future LMS. Conference IEEE, AICT/ICIW 2006, 19-24 February 2006, La Guadeloupe, France, 6 pages. [Chevrin and al. 2005a] Chevrin, V., Derycke, A., Rouillard, J. Some issues for the Modelling of Interactive E-Services from the Customer Multi-Channel Interaction Perspectives. EEE 05 International. IEEE international conference on e-Technology, e-Commerce and e-Service, IEEE Press, Hong Kong, pp. 256-259, 2005. [Chevrin and al. 2005b] Chevrin, V., Rouillard, J., Derycke, A., Multi-channel and multimodal interactions in E-marketing: Toward a generic architecture for integration and experimentation, HCI International conference, Las Vegas, 22- 227 June 2005, Lawrence Erlbaum editors, 10 pages. [Chevrin and al. 2003] Chevrin, V., Derycke, A., Rouillard, J., Un Cadre Théorique pour la Caractérisation des Interactions Multicanal en E-Marketing, IHM 2003, ACM Press, Caen, pp. 97-104, 2003. [Clark & Brennan 1991] Clark, H. and Brennan, S. Grounding in Communication. In Perspectives on Socially Shared Cognition. Edited by L.B. Resnick, J. M. Levine and S. D. Teasley. 1991.

[Cocoon] http://cocoon.apache.org/fr/ [Derycke and al. 2003] Derycke, A., Rouillard, J., Chevrin, V., Bayart, Y. When Marketing meets HCI: Multi-channel customer relationships and multimodality in the personalization perspective. HCI International 2003. Heraklion, Crete, Greece, Volume 2, pp. 626-630, Lawrence Erlbaum Associates. 2003. [Dey and al. 1999] Dey, A., Salbert, D., Abowd, G, Futakawa, M. The conference Assistant: Combining Context Awareness with wearable Computing. 3rd International Symposium on Wearable Computers, October 19-19, San Fransisco, California, pp. 21. 1999. [Greenhalgh & Benford 1997] Greenhalgh, C., and Benford, S. Boundaries, awareness and interaction in collaborative virtual environments. In Sixth IEEE Workshops on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE), Cambridge, June 1997. IEEE Computer Society. [Hadjouni and al. 2005] Hadjouni, M., Tirellil, I., and Laroussi, M. Collaborative Platform for Mobile Learning Based on e-Services. Conference IADIS international “Mobile Learning 2005”, Isafas, Borg, Kommers, Bonamo (eds) 28-30 June 2005, Qawra, Malta, pp 226-230. [Holsapple & Singh 2000] Holsapple, C. Singh, M., Electronic Commerce:from a Definitional Taxinomy Towards a Knowledge-Management View. In journal of Organizational Computing and Electronic Commerce, 10(3), pp 149-170. 2000. [Hinz & Zoltan 2004] Hinz, M., Zoltan, F. AMACONT: A System Architecture for Adaptive Multimedia Web Applications. Berliner XML Tage 2004: 65-74. [JADE] http://jade.tilab.com. [Rust & Kannan 2003] Rust, R. Kannan, P.K. E-service: a New Paradigm for Business in the Electronic Environment. Communication of the ACM, June 2003, vol. 46, N°6, pp 37-42. [MAIS] http://black.elet.polimi.it/mais/ [Peppers & Rogers 1997] Peppers, D., Rogers, M. The one to one future: Building relationships one customer at a time. New York: Doubleday. 1997. [Peppers & Rogers 1999] Peppers, D. Rogers, M. Dorf, B. The One-to-One fieldbook: the Complete Toolkit of a 1to1 Marketing Program. Doubleday, Random House, NY. 1999. [Rouillard 2003] Rouillard, J. HCI International 2003, Plastic ML and its toolkit, Heraklion, Crete, Greece, Volume 4, pp. 612-616, Lawrence Erlbaum Associates 2003. [Sedogbo 2004] Sedogbo, C. Human-system Interaction Container Paradigm - Sedogbo, Grisvard, Praud – Thales. http://www.w3.org/2004/02/mmi-workshop/. 2004. [Siebel] http://Siebel.com [THALES] http://www.thalesgroup.com/home/home/