Automated customer personalization at scale, powered by the world’s most intelligent recommender system
Galahad’s Scalable Recommender System Solution
Helps Clients Realize the Value of Personalization
We work with clients in these industries:
Banking
Insurance
Wealth Management
Healthcare
Retail
CPG
Travel
Entertainment
Non-profit
What We Deliver
• Acceleration Plan
– Based on discovery with key stakeholders, Galahad will document the business requirements for the personalization strategy including priority use cases, data, predictive analytics, channels and measurement
– Best practice recommendations will be detailed in an action-oriented “Blueprint” for the areas of customer experience, marketing strategy, digital, data, analytics/data science, channel and technology.
Deliverable: Personalization Strategy & Best Practice Blueprint
• Implementation Roadmap
– Based on client’s current capabilities and investment budget, a prioritized implementation plan will be created to set up, test and rollout personalization tactics and supporting data, insights and technology.
– The roadmap will be created with the objective of creating a self-funding process that maximizes investment returns. The customized Recommender System can be expanded beyond the initial version to incorporate additional features.
– A detailed Test and Measurement Plan will also be provided for potential use in a proof-of-concept campaign prior to the full solution setup.
Deliverable: Roadmap Timeline, Test & Measurement Plan
• Rapid Start Predictive Customer Intelligence
– We assess each client’s inventory of predictive intelligence including customer segmentation and models. These can be quickly integrated into the recommender system solution. In cases where clients do not have the existing intel needed, Galahad’s Customer Cube insights for customer growth, retention and price elasticity can be used as the quick start intelligence to deploy the recommender system.
– The system uses a robust set of offline and online data including purchases, profitability, RFM, demo/life stage, email engagement and digital activity.
Deliverable: Ready-to-use intelligence data including segment, profitability, growth potential and attrition risk
• Next Best Action Recommendations
– The automated, scalable recommender system is ‘The Brain’ that orchestrates data-driven personalization, determining the next best action to be taken for each customer and the optimal channel to interact with that customer at any point in time. It is the engine under the hood that powers a great personalization program.
– Next best actions can include product offers, transactional updates, reminders, informational and educational communications. Channels include both inbound and outbound interactions.
– Minimum viable product of the initial version recommender system can be tested using a few key actions and channel examples prior to deployment.
– Hands-on setup help of the initial system configurations and running test simulations to validate the appropriate message is delivered at the right time to the right customer through the right channel.
Next Best Action Recommender Engine
Overview of Recommender Engine
Features include:
• “Brain’ that determines next best action for each customer
• Dynamic offer engine for real-time decisioning
• Orchestrates & personalizes interactions across all channels
• Tailored to your business requirements and includes roadmap
– Next best actions can include product offers, transactional updates, reminders, informational and educational communications. Channels include both inbound and outbound interactions.
– Minimum viable product of the initial version recommender system can be tested using a few key actions and channel examples prior to deployment.
– Hands-on setup help of the initial system configurations and running test simulations to validate the appropriate message is delivered at the right time to the right customer through the right channel.
Testing and Measurement Framework
• The recommender system enables advanced A/B testing experiments so that personalization parameters such as audience segments, channels, offers, creative content can be evaluated with the measurement framework.
• The measurement framework also enables multi-touch attribution by quantifying the incremental value generated by each customer interaction in a series of contacts. This information can be used to update the configuration parameters of the system to better optimize marketing spend.
• Testing of the system is performed in several phases to ensure all aspects are operating as expected prior to going live with customers:
– Phase 1: testing with synthetic data in the development environment. The goal of this phase is to verify the operational functionality of the data flows including APIs and that the arbitration decisioning is functioning as expected.
– Phase 2: testing with actual customer data to ensure the efficacy of the recommendations being made. This phase typically entails close examination of hundreds or thousands of customer-level recommendations to evaluate if they are the optimal next best action given the specific profile of each customer.
– Phase 3: involves testing with the actual customer-facing delivery systems, typically at a channel level such as mobile app, online, call center or store.
• In conjunction with measuring incremental effects achieved from the personalization solution, the best practice includes using the results data to bolster the underlying algorithms that power the recommender system. With a fully deployed, multi-channel personalization program, the sheer volume of actions/offers by the various channels makes it challenging (if not impossible) to keep up using traditional data scientist-trained models.
Continuous Deep Learning
– Our Continuous Learning model-based framework can be implemented so that audience algorithms can be updated autonomously as sufficient results data becomes available. Our framework is typically implemented in two stages starting with baseline self-learning capabilities and evolving to more advanced self-learning powered by Reinforcement Learning.
• Continuous Learning Stage 1 – Effect Matrices by Micro-segment & Channel – As shown in the example below, incremental ‘effects’ for each action/offer recommended can be quantified using a combination of customer micro-segments and channel. – The ‘Effect Matrix’ quantify the lift over average of each action/offer to each micro-segment/channel combination. These can then be added to the recommender system’s arbitration process as a weighting to boost the most effective actions/offer/channel combinations for the right audiences. This method is a great starting point for continuous learning, as it is based on actual results data. It also does not require high-cost data mobilization. • In conjunction with measuring incremental effects achieved from the personalization solution, the best practice includes using the results data to bolster the underlying algorithms that power the recommender system. With a fully deployed, multi-channel personalization program, the sheer volume of actions/offers by the various channels makes it challenging (if not impossible) to keep up using traditional data scientist-trained models. • Continuous Learning Stage 2 – Reinforcement Learning – Deep Learning is a more exhaustive modeling approach that can identify more complex non-linear patterns in data to ensure the system is learning from even the most complex patterns of customer response. More importantly, it can be automated when the number of actions in the recommender system becomes so large rebuilding individual models manually becomes unfeasible. – The type of Deep Learning that fits best within the recommender system decisioning process is Reinforcement Learning (RL), which optimizes expected value of actions. One of the many benefits of the Test Design Approach referenced in the previous section is that the results data can be used to automate updates of the predictive models to better identify the next best opportunities for each customer. – As summarized in the diagram below, RL is a recursive modeling system that recommends Actions for each State (or snapshot instance) based upon maximizing rewards. Because RL algorithms learn from each action, they can be expected to perform better over time. – We have in-depth knowledge of RL and experience putting into practice for the application of customer personalization. We believe having this capability of autonomously, self-learning models, will create a very responsive dynamic of your personalization system’s ability to adjust to new customer information very quickly. The ability to create a distinctive, more personalized customer experience becomes a competitive advantage“Effect Matrix” – Quantifies lift over average for each Action/Offer using these combinations
Channel 1
Channel 2
Channel 3
Micro-segment 1
Response Effect (1,1)
Response Effect (1,2)
Response Effect (1,3)
Micro-segment 2
Response Effect (2,1)
Response Effect (2,2)
Response Effect (2,3)
Micro-segment 3
Response Effect (3,1)
Response Effect (3,2)
Response Effect (3,3)
Please contact us to learn more about our Recommender System Solution.
For qualified prospects, we will share a live demo of our Recommender System using synthetic data.