Case Study: AI Integration for Precise Retailer and Transaction Recognition

Take a detailed look at how the Cloud Combinator team used our AI & ML Accelerator framework and advanced AI systems to overcome a problem faced by a loyalty platform used by 3 million people and some of the biggest brands in the UK. 

About The Customer

Airtime Rewards, a UK-based fintech company, has revolutionised retail loyalty and mobile marketing by creating a unique currency that smartphone users can apply towards their mobile phone bills. As a result, the Airtime Rewards platform connects consumers, retailers, and mobile networks - streamlining transactions, enhancing user convenience, and encouraging custom at partner retailers.

Customers simply register with their phone number, allowing the system to automatically apply earned credits from purchases as discounts on their phone bills, bypassing the usual complexities of loyalty schemes. This simplicity boosts consumer participation and satisfaction - but also requires no system integration at the retailer’s end either, keeping things simple across the board. 

The Airtime Rewards platform has excelled at helping retailers acquire and retain customers through intelligent data analytics - offering targeted segmentation that provides deep insights into consumer behaviours. This insight allows retailers to tailor their offerings - significantly increasing the chance of customer engagement and loyalty.  

To date, this enhanced consumer participation has seen Airtime Rewards help to create £1bn+ in incentivised spend for over 150+ retail partners - including Tesco, Greggs, Burger King, Nandos, Halfords, Boohoo, Argos, New Look, Harvey Nichols, and a range of other household-name brands. By combining user-friendly technology with effective marketing strategies, Airtime Rewards has become an invaluable tool for retailers looking to navigate the competitive - but often complex - fintech landscape. 

Customer Challenge

As is often the case, Airtime Rewards’ simple user interface requires complex behind-the-scenes processes - not least the mechanism that identifies retailers a customer has used from transaction descriptions that are logged. In some cases, Airtime Rewards’ current system was unable to discern the retailer from this transaction description - so points that should have been allocated to users for their purchases were not being correctly assigned, leading to inaccuracies in reward distribution.

If this problem went unaddressed, Airtime Rewards risked a series of issues. Firstly, unallocated points could easily lead to customer dissatisfaction - potentially leading to decreased customer engagement and retention. Additionally, the inability to accurately track and attribute transactions could potentially also limit the company’s ability to provide personalised offers - which could ultimately reduce the value offered to retailers and lose the company’s competitive edge in the fintech market.

As such, the Cloud Combinator and Airtime Rewards teams agreed that the solution should aim to address three main issues:

• Airtime Rewards' system had difficulty identifying retailers from transaction descriptions, resulting in inaccuracies in reward distribution.
• Unallocated points due to misidentified transactions could lead to customer dissatisfaction, decreasing customer engagement and retention.
• The uncertainty around accurately tracking and attributing transactions could limit the ability to provide personalised offers, potentially reducing the value offered to retailers.

Partner Solution

To tackle the issue of identifying retailers from transaction descriptions, Cloud Combinator engineered a solution centered around the use of a Large Language Model (LLM) hosted within Amazon Bedrock. The LLM was selected for its capability to analyse and extract retailer information from transaction references. The in-house prompts designed for the LLM ensured accurate parsing and extraction of relevant data.

Key Components of the Cloud Combinator Solution:

1. Amazon Bedrock: The LLM within Bedrock played a crucial role in extracting company names from complex transaction codes and checking if they were parent companies or subsidiaries. This ensured accurate retailer identification and enhanced the precision of points allocation. Cloud Combinator’s expertise around generative AI allowed for the creation and rigorous testing of prompts within the LLM. 
2. AWS Lambda: This serverless computing service handled the entire data processing workflow. Lambda functions were deployed to control the flow of JSON files containing transaction data, ensuring efficient handling and processing without the need for dedicated server management.
3. Amazon RDS: Results of the pipeline process were stored in RDS - and compared to list of company names provided by Airtime Rewards.
4. Secure Data Storage and Management: All processed outputs and retailer mappings were securely stored in AWS databases, ensuring data integrity and confidentiality. The LLM's output was directly saved in a secure environment, accessible only to authorised systems and personnel.
5. Training and Handover: Post-implementation, Cloud Combinator provided comprehensive training to Airtime Rewards' project lead, ensuring that their team was fully equipped to utilise and manage the new system. This training covered operational aspects of the solution, including how to handle new transaction types and troubleshoot common issues.

When the LLM had extracted the company details from transaction descriptions, it was compared to a real-time list of Airtime Rewards partner companies – ensuring an accurate match.

Beyond the matching task, using Amazon’s Relational Database Service (RDS) to store all identified transactions also provided added value for Airtime Rewards. With countless database management possibilities within Amazon RDS, the team at Airtime Rewards were then also able to interrogate this wealth of now clear data, without having to seek further external support.   

This robust, serverless infrastructure allowed Airtime Rewards to significantly enhance their loyalty program's reliability and customer satisfaction by ensuring accurate and fair points distribution based on a deep understanding of transactional data. In turn, these system enhancements also ensured maximum effectiveness when applying customer behaviour data, safeguarding the benefits that retailers expected when partnering with Airtime Rewards.

Results and Benefits

The implementation of Cloud Combinator's solution using AWS services successfully addressed Airtime Rewards' challenge of accurately identifying retailers from transaction descriptions. 

The integration of the Large Language Model, supported by AWS Lambda and DynamoRDS, has ensured that the loyalty program now reliably recognises both direct and subsidiary retailer transactions. This precision in identification has led to correct points allocation to users, significantly enhancing customer confidence and retailer satisfaction and trust in the Airtime Rewards platform.

Across the board, retailer feedback has been positive. From the outset, the Airtime Rewards team understood and explained how important robust adherence to offers and reporting was to their retail partners. As such, the improved strength of the company’s data has ensured that big-name partners have been retained - an important part of delivering an excellent product for end-users and partners alike. 

In Numbers:

Prior to the roll-out of Airtime Rewards' transaction categorisation solution:

• Misallocation of points occurred in around 0.5% of transactions.
• With over 78 million transactions, this equated to approximately 300,000 transactions with mismatch issues.

After implementing Cloud Combinator's solution:

• The misallocation rate was reduced to less than 0.1%.
• This equates to fewer than 4 transaction issues for new partner companies.

The new transaction recognition system:

• Reduces intervention time from 2-3 minutes to around 30 seconds.
• Represents a 75% time-saving for resolving unforeseen issues.

Why Cloud Combinator?

Cloud Combinator’s expertise in the AI/ML field made the company a natural contender for Airtime Rewards’ transaction categorisation project - but it was demonstration of the team’s proven AI & ML Project Accelerator program that established Cloud Combinator team as the go-to partner for the task. The AI & ML Project Accelerator process mapped out clear timescales and delivery dates too - crucial for a team looking at addressing a time-sensitive issue.  

Cloud Combinator’s position as an Advanced AWS partner also made it possible to leverage funding unavailable to many other service providers. This allowed for a clear demonstration of business value and success long before the board at Airtime Rewards were expected to start exploring funding channels for the project - further adding to the speed of delivery.

Client Satisfaction

“"Cloud Combinator has enabled us to improve our customer & member experience by introducing a first for us - an AI-driven transaction matching service, this has reduced toil for our member experience teams and led to greater retention of our critical retail partners. Throughout the engagement Cloud Combinator were professional, open, and delivered a high-quality solution, I appreciated the regular check-ins and demos and the deployment guides provided."