Personalizing promotions and product recommendations in retail with AI-powered models
Project Overview
Our client is the largest multi-channel retailer in its country. We developed a PoC for causal AI-driven pricing and promotions, as well as generative AI-led personalized product recommendations tailored to customers.
There were two parts to the project, with the following objectives:
#1 Causal AI for personalized pricing
- Defining the structure and parameters of a synthetic dataset ensuring it captures relevant customer behavior and price/promotion sensitivity patterns.
- Designing a synthetic data generation framework that simulates randomized controlled trials (RCTs) with various promotional discounts.
- Creating an uplift modeling and baseline evaluation – perform RCT analysis, extract data-driven insights on the effectiveness of personalized pricing, and develop the baseline causal AI model from RCT data.
#2 Generative AI for outbound comms recommendations
Synthetic data generation:
- Data analysis: Analyzing existing customer and product data to identify key attributes, relationships, and behavioral patterns.
- Synthetic data model: Developing a synthetic data model that can generate realistic customer profiles (demographics, preferences, purchase history) and product information (specifications, categories, pricing).
Conversational AI development:
- Dialogue design: Creating conversation flows for customer acquisition, retention, upselling, and cross-selling scenarios.
- Integration with synthetic data: Connecting the conversational AI to the synthetic dataset to provide personalized recommendations and information during interactions.
- Natural Language Understanding (NLU): Implementing NLU capabilities to interpret customer queries and intents accurately.
Challenges
Business challenges included:
- Resource constraints: Our client wanted to ensure that promotions would provide a strong ROI on a limited budget
- Manual workload: Manually calculating and comparing different promotion scenarios was time-consuming and subject to avoidable errors
- Lack of insights: There was uncertainty around which discounts are most effective for different customer segments (tech enthusiasts, budget shoppers, etc.)
Solution overview
Neurons Lab created a RAG system combined with a vast knowledge graph (KG) to retrieve better contextual information, generate more accurate responses, and reduce the frequency of AI hallucinations.
The tech stack was provided by AWS Cloud to extract, process, and feed GitHub data and other documents into the model and build an AI orchestration. The user query and the initial response are run against a large language model (LLM) to generate a relevant discount coupon code.
The LLM evaluates the retrieval context and also the prediction, comparing the results against a KG – providing a better understanding of the reasoning behind the answers, with accurate data parsed from the retailer’s website.
User interface
The customer segmentation dashboard includes:
- A visual display of key customer segments based on purchase history, browsing behavior, demographics, etc.
- Each segment is labeled with its estimated size, average spending, and potential responsiveness to promotions.
- This helps the Campaign Manager understand their target audience and tailor promotions accordingly.
A campaign overview portal provides a vast range of data. It displays current active campaigns, showing the overall return on investment (ROI) and uplift for each one. It also highlights the campaigns nearing ROI thresholds.
There is also a display for the total promotional budget and data for the allocation of budget across different campaigns.
Promotion allocation optimizer
Capabilities in the promotion allocation optimizer overview include:
- Interface to select products for promotion
- Input field for promotion details – discount percentage, free gifts, etc.
- Setting start and end dates for promotions
- A tool that leverages uplift modeling to predict the increase in sales for different discount levels and customer segments
The optimizer considers budget constraints and ROI targets for each campaign, recommending the optimal allocation of discounts to maximize overall sales and profitability.
The tool can simulate different scenarios and visualize the predicted impact on sales and ROI. For example: “What would happen if we offer a 20% discount on TVs to tech enthusiasts?”
.
Results
Neurons Lab created a robust causal AI model that analyzes customer behavior and promotion sensitivity patterns based on synthetic data, ready for uplift modeling.
We also delivered a conversational generative AI prototype capable of answering user queries with accurate reasoning and recommendations based on a KG from product specifications.
#1 Causal AI for personalized pricing
Results included:
- A fully working demo that successfully covers the following steps: generating customer data, running a RCT Experiment, analyzing the RCT results, exploratory data analysis, building an uplift model and running a discount targeting policy.
- An initial model for promotion targeting which we built and evaluated.
#2 GenAI for outbound comms recommendations
The solution is capable of extracting relevant product specifications and constructing a KG from the product specification. It answers user queries with reasoning from the KG.
Key deliverables included:
- Synthetic dataset generation model
- Conversational AI prototype
- POC report detailing the results and recommendations
About Neurons Lab
Neurons Lab is an AI consultancy that provides end-to-end services – from identifying high-impact AI applications to integrating and scaling the technology. We empower companies to capitalize on AI’s capabilities.
As an AWS Advanced Partner, our global team comprises data scientists, subject matter experts, and cloud specialists supported by an extensive talent pool of 500 experts. We solve the most complex AI challenges, mobilizing and delivering with outstanding speed to support urgent priorities and strategic long-term needs.
Ready to leverage AI for your business? Get in touch with the team here.
