Adtelligence AI Framework
AUTOMATE YOUR DATA SCIENCE
Move from insight to automated action and decision
Move from insight to automated action and decision
Machine Learning is key when it comes to scaling your digital business. Starting with manual one time scores and manual triggers, most use cases need a lot of manual efforts, working hours and costs to be successful. This lack of scalability and automation leads to average results in the long term. To create a pillar for your digital business models, automation, self learning and optimization is key.
Credit card issuers need to leverage all of their data with AI to learn and act on user behavior in real-time and personalize communication automatically.
With any new data set our Machine Learning models update in on the fly to trigger next best action on the preferred channel.
“Machine Learning is the automation of data science. Automated Machine Learning models boost productivity when it comes to personalized customer interactions along the lifecycle and increase scalability. When data is the new oil, payment providers sit on the biggest nearly untouched oil field. Machine Learning will become the keystone of future revenue models of payment providers and card issuers.”“
Michael Altendorf, CEO, Adtelligence GmbH
The model library provides a wide range of capabilities to increase revenue and drive revenue along the customer lifecycle. The Adtelligence AI framework enables connectivity via APIs. Input your data and let our models train you. We provide you with pre-built models, from affinity scores to clustering to optimization with neural networks or Baysean Bandits. You can use your scores and models as well, but also create new models.
|Lookalike Audiences||Lead Score||Conversion Score||Audience Segmentation||Activity Score||Referrer Score|
|Functionality||Segmentierung||Prediction||Prediction and |
|Definition||Lookalike audiences are a group of potential customers who have similar data characteristic such as current accounts||Lead Score ranks prospects according to their potential value as high potential customers||Conversion Score calculates the potential for a prospect to become a customer within a certain time period||Creating customer segments is one of the most common ways to target campaigns more specific to groups based on specific data points which describe the segment such as age, gender, location or product categories||Activity Scoring analyzes customer behavior and ranks it compared to other current accounts||Potential of a customer to invite friends and family to your product|
|Typical use case||Lookalike audiences are customer segments that are similar to current account segments and are used to target new customers across different channels such as Google or Facebook||Lead scoring is used in lead mgmt to optimize potential efforts to win a customer||Conversion scoring is used to predict and optimize website personalization and abandoned cart campaigns||Creating customer segments manually can be a good start to avoid sending the same message to all customers. The amount of customer master data, behavioral and transactional data is increasing to a level that manual segmentation produces a high amount of workload and a trade-off for each campaign manager to which level it is feasible||Activity scoring is used by e.g., retailers, banks, gaming and payment card companies to analyze the health of a customer relationship||Refer-a-friend offers create a simple and often used way to win new customers at a statically calculable cost, in addition to traditional marketing channels|
|Challenges of traditional data science approach||Lookalike audience campaigns do not allow a 100% matching but a rough audience selection, resulting in average targeting and average conversion rates||Static and periodic lead scoring does not allow to leverage current prospect behavior and leads to wrong or incorrect predictions||Static website or campaigns are no longer competitive, as prospects can bounce any second due to irrelevant content and product offerings||The ML based approach automates segmentation, clustering of key segments. It automates manual process steps and allows the marketeer to create audiences in real-time. Audience segments are updated automatically up to real-time with every new data point||Traditional activity scoring is periodic and analyzed by data science or tech savvy growth marketeers. Automated alerting functions are often not in place or based on a small number of data points||Static referrer scoring models overlook the huge potential of "refer-a-friend" campaigns, partner products and other viral invitation methods|
|Advantage of dynamic Machine Learning models||An automated clustering of your current account data improves the audience target segments continuously and automates optimization of lookalike audiences fast and iterative||Automated optimization of lead scores allows to increase the amount of data and to score in real-time depending on a variety of data parameters from marketing channel to context and live behavior on a website or app||Tailoring personalized contents and products||The ML based approach automates segmentation, clustering of key segments. It automates manual process steps and allows the marketeer to create audiences in real-time. Audience segments are updated automatically up to real-time with every new data point||A modern approach increases the level of sensitivity and takes all data points into account in real-time and creates affinity scores per customer segments||Highly personalized offers for friend referrers optimize number of new customers for a low CPA|
|Machine Learning techniques||HDBScan, k-Means (vector quantization method)||Regression, Neural Network, Decision Tree||Reinforcement Learning, Neural Network, Multi-Armed Bandits & Bayesian Optimization||HDBScan, k-Means (vector quantization method)||Statistical correlation, Bayesian classification, Hidden Markov Models||Regression, Neural Network, Decision Tree|
|Business potential||Increase conversion rates in ad campaigns and reduce customer acquisition costs||Increase the number of high potential customers and reduce CLV campaign costs||Higher customer acquisition rates, more new customers, higher revenues||Hyper-personalisierte Kampagnen erhöhen den Warenkorbwert||Improved alerts and targeted campaigns based on customer activity increase lifetime value and customer revenues per month||Higher customer acquisition rates with lower costs per new customer|
|CLV Score||Next Best Action||Next Best Product||Next Best Category||Risk Score||Churn Score|
|Functionality||Prediction and Classification||Association and Correlation||Association and Correlation||Association and Correlation||Prediction and Classification||Prediction and Classification|
|Definition||Customer Lifetime Value Score predicts the potential monetary value of a customer during his lifetime||Next Best Offer scores the affinity to a special offer||The affinity score predicts the products which fit best as next purchase for a specific customer data set or behavior||Next Best Category ranks a customer against a variety of product categories and its affinity for an up- or cross-sell in that category||A risk score predicts the potential risk of a customer not paying according to terms or risk of behavior for potential loss caused by his behavior||Churn Scoring predicts potential customer attrition or reduction in spend|
|Typical use case||Customer lifetime values are used to classify a specific customer type based on its past purchasing behavior compared to other customers and calculate a score to benchmark its value||Creating targeted marketing campaigns via email, app or onsite||Creating upsell or cross-sell opportunities for customers. Recommendation engines, "You may also like" are common on any shop to show further products in the same category||Typically used for cross-sell campaigns, onsite, across mail or email campaigns as well as in-app offers||Analyzing the risk of fraud, payment defaults and other business risks||Churn can occur in a variety of ways: revenue churn, usage reduction - predicts churn, customer churn|
|Challenges of traditional data science approach||Static periodic CLV scoring may trigger wrong campaigns, wrong timing and overlooking of high-potential customers during a certain period||For creating affinities, purchase data is of much greater interest than purely demographic data. In addition, active behavior on shops and higher churn potential if neither transaction data nor behavioral changes or marketing channel preferences are considered||Traditional recommendation engines can not take real-time customer behavior into account and are based on past purchases of other customers||Next best category campaigns are often created manually by category managers or marketeers for bundle campaigns and need high amounts of efforts to be precise||Traditional decision-tree or statistical risk scoring do not take real-time behavior into account and lead to incorrect scores and negative scoring of potential customers according to static rules||Static churn scoring reacts episodically and potentially too late in a customer’s lifecycle leading in declining results and higher churn rates. Today, churn scores are often calculated based on CRM data without leveraging online touchpoints|
|Advantage of dynamic Machine Learning models||Dynamic CLV scoring allows highly targeted and personalized customer experiences in real-time which leads to higher conversion rates and purchase volume||Highly targeted offers triggered by AI based behavior and context analysis algorithms improve purchase rates and customer engagement||Dynamic recommendations and real-time customer behavior data improve product recommendations up to 1to1 personalized offers and allow long tail campaigns||Real-time optimization of scores allow using customer behavior to increase cross-selling opportunities and incentive usage and purchases||Dynamic approaches score on a multidimensional data signal from installed font types to historic purchase history and allow a much higher granularity||Machine Learning powered RFM modelling - Scoring recency, frequency and monetary values in real-time including online customer behavior and card usage data. This allows churn prevention campaigns to be segmented and tailored on time on the right channel. Optimizing anti churn campaigns is key for a long-term revenue stream|
|Machine Learning techniques||Statistics, Recurrent Neural Networks (RNN)||Statistical correlation, Bayesian classification, Reinforcement Learning, Q-Learning, Association Rule Learning||Statistical correlation, Bayesian classification, Reinforcement Learning, Q-Learning, Association Rule Learning||Statistical correlation, Bayesian classification, Reinforcement Learning, Q-Learning, Association Rule Learning||Regression, Neural Network, Decision Tree||Regression, Neural Network, Decision Tree|
|Business potential||Higher focus on profitable customer segments and higher retention rates||Higher revenues, higher value per purchase, lower manual efforts||Higher revenues, higher value per purchase, lower manual efforts||Higher revenues, higher value per purchase, lower manual efforts||Lower payment defaults, lower fraud and improved customer acquisition||Reduced churn risks, improved customer retention and higher revenues|
Credit card issuers have excellent data sources to leverage through data-driven customer lifecycle management
Available data sources range from static customer relationship management (CRM) data – such as age and gender – to more sophisticated information, such as creditworthiness. Product information, loyalty data provide insights about the products to which a customer is currently subscribed, and opportunities to cross-sell and upsell other products that the customer is not yet using. Credit card transaction data provide live information about the customer’s spending behavior. These insights can be used to make marketing decisions in real-time.
Utilizing first party data with Machine Learning can overcome the challenges of the cookieless future
Most browsers no longer support tracking by third-party cookies. Google has announced that the Chrome browser will stop accepting third-party cookies in 2023. Marketers will have less information about online customer behavior, and brands will be less able to reach customers with targeted messages – unless they can make customers accept first-party cookies and use artificial intelligence to exploit available data as effectively as possible.
Examples of the variety of data points
Move from insight to automated action and decision