Campaign optimization using Business Intelligence and Data Mining
Download as PPT, PDF2 likes419 views
The document discusses the optimization of marketing campaigns using business intelligence and data mining techniques, emphasizing the importance of customer targeting and performance analysis. It outlines processes for designing effective target groups, analyzing campaign responses, and implementing personalized marketing strategies to improve customer engagement and satisfaction. The document also highlights the significance of data-driven approaches in enhancing campaign management and decision-making within e-business environments.
Campaign optimization using Business Intelligence and Data Mining
- 1. Campaign Optimization Using Business Intelligence and Data Mining George Krasadakis March 2007
- 2. http://www.datamine.gr Outline Key concepts & definitions A common language regarding campaigns, the main dimensions & metrics involved The need for campaign optimization The typical campaign management lifecycle and the need for optimization Designing the Target Group Data-driven approaches for target group definition – use of BI and Data mining techniques Performance Analysis Analyze campaign response data, model customer responses, compile reports Application within E-Business environments Campaign, recommendation, profiling and personalization
- 3. http://www.datamine.gr Key concepts & definitions Campaign A set of systematic promotional activities (multiple offers, scenarios & channels) against a well defined target group (advanced business logic for accurate customer selection) within a controlled environment (infrastructure for response gathering, reporting, analysis and modeling). Campaign Management Infrastructure & processes enabling efficient design (Target group definition - customer selection, eligibility criteria, profile analysis), smooth execution (integration with communication channels) and effective response analysis (response gathering, analysis, reporting and modelling). Data Mining & BI (Business Intelligence) −BI is based on several technologies & scientific areas such as information technology, multidimensional data exploration technologies (OLAP), data mining, statistical modeling, text mining, visualization techniques −BI enables companies to explore, analyze, and model large amounts of complex data −BI can greatly enhance Campaign Management processes from Design (TG definition), Execution (efficient communication planning), to response analysis & modelling (exploratory and/ or with data mining)
- 4. http://www.datamine.gr The need for optimization The ultimate goal Enable the right treatment on the right customer at the right time through the right channel. This further enables customer understanding (needs, preferences, usage & buying patterns) enabling customer response analysis and modeling The roadmap Design, implement and automate solid campaign management processes. This will provide flexibility (in handling customers, products and promotions), reliability (regarding execution, response gathering) and robust measurement & analysis processes - functions. This will enable a systematic monitoring and analysis framework to support decisioning in general The business value − Winning the performance game (On-time Schedule Indicator, Cost Per Activity) − Customer insight - usage patterns, profiles and customer base trends may reveal significant cross-selling or up-selling opportunities − Assessment of marketing actions, special offers or campaigns can be assessed in detail using customer responses and changes in usage patterns: The Closed Loop Marketing − Retain (ensure) or increase Customer Satisfaction levels
- 5. http://www.datamine.gr Campaign Management System Customer database Documents & templates Communication Channels Campaigning: lifecycle Target Group Definition The MKT user interacts with CMS in order to explore the customer base and design the most effective target group 1 Customer Profile Analysis CMS retrieves customer information in order to provide sufficient segmentation capabilities to the MKT user 2 Target Group Release for contact List of customers –Target Group- as defined from the MKT user, and after applying the selected, predefined exclusion logic 3 Customer Communication The offer assigned to the campaign is being communicated to the customer according to the predefined script or template 4 Customer Response Customer responses are being forwarded into the system for campaign assessment, monitoring and optimization 5 Campaign Analytics Campaign performance statistics, customer demographics, campaign lifecycle information, call center performance reports and analytics 6 Campaign performance Assessment Sufficient input for better campaign design, customer behavior modeling. Insight for process monitoring, KPIs for assessment studies 7
- 6. Target Group Design Locate, profile and manage customers according to composite business logic
- 7. http://www.datamine.gr Designing the target group Using Segmentation schemes effective schemes for categorizing and organizing meaningful groups of customers Customer Profiling the process of analyzing the elements (customers) of each segment in order to generalize, describe or name this set of customers based on common characteristics. It is the process of understanding and labeling a set of customers The process − the target group definition process is an iterative procedure aiming in compilation of a well structured set of customers with certain degree of homogeneity regarding a set of attributes. − Involves business knowledge, ideas & creative thinking as well as data-driven concepts, facts and modelling activities − Requires effective exploratory analysis and in-depth understanding of the customer base − Can be optimized using advanced modelling techniques and data mining algorithms
- 8. http://www.datamine.gr Designing the target group The Physical Customer Structure Physical Customer Identification is a critical point in customer segmentation & insight: A physical customer may have several accounts with contradictive behavior regarding usage or payment. The physical customer (a) must be correctly identified and (b) must be efficiently scored in the top level Physical Customer Usage History Usage metadata Customer Care & Contact History Application, ordering & payment History Time Related Patterns Statistical & Data Mining Modeling Analytics, segmentation & profiling Benefits − A complete picture of the customer, in all dimensions (profitability, risk, loyalty, satisfaction etc) − Elimination of contradictive communication attempts (bonus due to product A ‘performance’ while in collections procedure due to product B payment habits)
- 9. http://www.datamine.gr Dimensions & Filters Customer -Risk Class -Revenue Class -Socio -Economic data -Demographics -Location data (GI) -Tenure (CLS) -Traffic Patterns -Contact Patterns -Prior Classifications Product - Services -Accounts, status & types -Services & Tariffs -other properties Designing the target group: input Target Group Design Involves all the important aspects of each customer: risk, tenure, profitability, or Customer value must be combined in order to explain or optimize a set of metrics or specific behaviors Measures -total revenue -Balance by type (source) -frequencies -’recent’ statistics -’lifetime’ statistics -AMOU / average duration -ARPU / average revenue -Specific Traffic metrics (services usage – destination analysis, incoming vs outgoing etc) -Churn Behavior -Campaign Responses -Customer Satisfaction metrics Metadata − Macro segmentation for management & decision support and performance evaluation purposes − Micro segmentation schemes, campaign specific, for product development, up selling or cross- selling program design, for loyalty – churn management, marketing actions
- 10. http://www.datamine.gr Designing the target group: CBE Customer & Products Attributes enabling the dynamic target group definition 1 Dimensions & Measures Enabling custom views of your customer base 2 Customer Sample Random sample of Customers for verification reasons 3 Customer Profiling Analysis of the resulting customer set versus any combination of attributes 4
- 11. Performance Analysis Browse, report and model customer responses
- 12. http://www.datamine.gr Campaign response analysis A Measurement Environment A set of metrics, KPIs and predefined reports, enabling an instant picture of each specific campaign. Reports also include suitable comparisons with ‘global constants’ such as group averages, baselines and predefined limits thus enabling comparative performance analysis of a campaign. Customer Contact History Customer campaign memberships and response history (memberships, contacts, feedback, offers & promotions attempted) should be maintained and further processed in order to generate related customer metadata. This ‘customer communication history’ should also be available to other systems as well, thus extending the knowledge regarding customers, their needs and preferences. Detailed Campaign History Campaign History & Reporting provide rich history of the full lifecycle of each specific campaign. Information on campaign execution events can be used as markers against the evolution of the customer base (reporting before and prior the campaign) for trends, indirect results or pattern identification. Formal evaluation ROI models, comparisons of expected results against actual, analysis versus initial statistical profiles of the target group, all packed in standardized, well define reports
- 13. http://www.datamine.gr Campaign response analysis Campaign Analysis Cube Analyze campaign response data in any meaningful way. Start with exploratory analysis, browsing the results in order to see the shape of the response set. A powerful, high-performance environment for browsing customer response data. Basic dimensions: 1. Customer segment: enables the projection of the target group of your campaign (and any subset as well) against the available segmentation schemes 2. Customer Profile type: similarly the customer set can be analyzed in terms of well-known & understood customer profiles 3. Channel: the channels available/ selected for the specific campaign. Enables analysis of performance (for instance response rate against channel used and in combination with other dimensions) 4. Offer: the actual promotion, offering to the customer 5. Contact Time: the time zone (day and time – according to schemes in use) 6. Timing: the time positioning of the communication event in terms of customer critical dates (e.g. forthcoming contract expiration or renewal process) 7. Script: the actual communication ‘dialogue’ – how the offering has been proposed to the customer 8. Agent profile: Characteristics of the agent involved (demographics, experience, seniority, specialization)
- 14. http://www.datamine.gr Campaign response analysis Campaigns – working list Quick or composite campaign search functionality and the resulting campaigns list. To be used as navigation tool for exploring and managing campaigns 1 Campaign Viewer A set of different views against the selected campaign (from sophisticated analytics to execution oriented reports) provide instant & accurate information on the aspect of interest 2 Cohort Analysis Specialized computations & Charts provide direct insight to campaign performance factors. Quick tabulation along with export utilities in a standardized output ensures optimum results with minimum effort 3
- 15. http://www.datamine.gr Campaign response analysis Customer base mapping according to generated profiles 100 75 50 25 0 RevenueRank Tenure Rank 0 25 50 75 100 Customer Profiles projected against by revenue & tenure Response A Response B Response C Response D Response E Response categories Categorized customer responses Customer projection Projected on a two dimensional space (revenue-tenure) ranks, and colored by response category for the selected profile
- 16. http://www.datamine.gr Applying Data Mining Data Mining refers to statistical and machine learning algorithms, applied in large amounts of data, aiming in identifying hidden relations and patterns. Popular data mining models include decision trees, clustering & association rules. − Association rules can identify correlations between pages/content not directly or obviously connected. May lead to previously unknown – not obvious- associations between sets of users with specific interests thus enabling more efficient treatment of customer − Clustering is a set of statistical algorithms aiming in grouping together items (customers) that present at least a certain degree of homogeneity relevant to specific measures. In contrast, the ‘distance’ between groups is maximized, thus forming a physical ‘segmentation scheme’ for further processing or event direct use. − Classification refers to a family of algorithms that ‘learn’ to assign items to pre-defined (existing) groups. − Sequential Analysis is a methodology for unveiling patterns of co-occurrence
- 17. http://www.datamine.gr Campaign response modeling Sample rules as derived from Decision trees: CreditLimit >= 15150,007 and ProfessionClass = 'Medical staff' > (positive=91%, negative=9%) CreditLimit >= 15150,007 and ProfessionClass not = 'Medical staff' and Residence not = 'ΘΕΣΣΑΛΟΝΙΚΗ - ΠΡΟΑΣΤΙΑ' > (positive=82%, negative=18%)
- 18. Web Analytics Campaigns, recommendation and personalization for the e-business
- 19. http://www.datamine.gr Personalization: Definitions, Needs & Business Value Personalization − consists of mechanisms used to adapt a web-site in terms of information / content served or services/ functionality enabled, based on user navigational patterns, their profiles and their preferences. − improves customer experience, resulting in more efficient actions through an ‘intelligent web site’ able to adapt according to user’s profile. May dramatically improve customer (user) satisfaction & Loyalty, usage boost, cross-selling & up-selling opportunities Personalization within typical e-commerce environments can take the following forms: − Recommendation. Determine suitable material (content, links, listings etc) for the specific user and the specific session. The ‘suitability’ of the material is computed from data mining algorithms which process large volumes of data and identify ‘hidden’ relationships. − Localization. User’s physical geography (as registered), or retrieved (connection based) can be used and ‘appropriate’ content is displayed − Targeted Advertising. ads that are expected to interest the user most (based on data mining – profiling & segmentation models) − Email Campaigns. Personalized messages to highly targeted users (according to their profiles/interests & segmentation schemes)
- 20. http://www.datamine.gr Personalization: An overview Website I.T. Infrastructure CMS DOC Billing User Interaction Session data that describe typical user interaction with the portal/ web site. Includes requests, user registration and preference data, navigational information 1 2 3 User Request/ data submission registration and preference data, navigational information Web Analytics Infrastructure Data mining models ETL Data gathering, Cleansing, preparation & standardization, data mining specific transformations Analytics Database Customer profiles, content structure & Metadata, processed traffic information Recommendations Engine Reporting Engine Personalized Output Personalized content (links, documents), controlled functionality 4 5 Systematic Raw Data Feed Raw data describing key portal entities, traffic data, content. Gathered systematically from the ETL components for further processing, analysis and modeling Portal Personalization transaction Portal submits visitor's identification data. RE retrieves metadata, compiles a Recommendation’s List and forwards it to the portal Personalized Data Recommendations List as served from RE
- 21. http://www.datamine.gr Personalization: Data Requirements User data includes information that can be used to define profiles of the physical user (individual and/or company) such as: − Demographics: gender, age, socioeconomic data, profession, education level, company attributes etc − Interests & preferences: communication settings, interests against specific content categories or functionality offered (as submitted by the user through registration process) − User experience: experience in the domains of interest, roles etc Usage data consists of the set of data that describe in detail every single user-portal interaction. A usually complex, large volume data set including log file information, session specific data, content structure. Environmental data refers to information describing the technological infrastructure enabling each user to access services and content offered (hardware, software, operating system) ‘Portal data’ refers to information providing structural representation, content definitions, relations, actions, processes (registration, applications, service activation, inquiries etc)
- 22. datamine ltd Decision Support Systems 22 Ethnikis Antistasis ave 15232 Chalandri Athens, Greece T: (+30) 210 6899960 F: (+30) 210 6899968 info@datamine.gr http://www.datamine.gr George Krasadakis Head of engineering g.krasadakis@datamine.gr Contact information
Editor's Notes
- #6: the typical lifecycle of a campaign, brief description. Must a.explain the lifecycle and b. show the directions of optimization, i.e. where is the waste of effort, money or negative results
- #14: show a hypothetical campaign response analysis using a cube (1-2 slides). The point is to demonstrate that there are patterns regarding customer response. These patterns can be spotted using BI – exploratory analysis and trigger analysts to new target group definition processes. A ‘campaign analysis cube’ may have dimensions such as customer cluster, segment, profile, key-demographics, channel used, agent profile, product offered, recency figures our advice will be that there is a strong need for powerful multidimensional (exploratory in nature) and also that further more advance modeling can be used (data mining).