10 Points to consider before rolling out a Fraud Management RFP
Telecom fraud is increasing at an alarming rate globally and is one of the most significant sources of revenue erosion for every telecom operator. With bottom lines under tremendous pressure, it is now time for CSPs to invest in advanced technologies to strengthen their fraud management systems. This would enable them to stay ahead of fraudsters and prevent frauds before they occur, and reduce the impact on revenues. In its report on “Market Trends: Maximizing the Value of Analytics, Artificial Intelligence and Automation in CSP Fraud Management,” Gartner estimated that by 2022, new implementations of ML within CSP fraud management would reduce fraud losses by 10%.
If a Fraud Management team is planning to integrate AI/ML capabilities into their FMS and is considering rolling out an RFP, here are the 10 key points they need to consider while drafting an RFP for AI/ML-based fraud detection service provider.
1. Assessment of Maturity Level
Approach to fraud detection using AI/ML technologies could significantly transform from the current controls that CSP might be currently doing. Hence, it would be good to conduct a maturity assessment to understand the maturity of their existing fraud management function and how bringing AI/ML capabilities can enhance this maturity level and lead to more automation.
2. In-house or Outsource
This might be a critical decision that CSP’s might have to take early in the game to go with in-house data scientists or external platforms and vendors. It’s a fact that many of the AI/ML techniques are available in open source worlds, and internal teams can be used to bootstrap and try out various techniques. However, various aspects need to be taken into account before a decision could be made; some of the critical question would be
- What is the right use case?
- How will it scale to meet the increasing data load requirements?
- What algorithms and hyperparameters to be used for solving the business problem?
- What data sets to be used to build the model?
- What should be the best practices in deploying the model in production?
It is a known fact that acquiring clean data engineering itself might take considerable effort in model building. CSPs should explore data engineering and feature store platforms that can cohesively stitch different facets to enable the organization to bring out the required AI transformation.
3. Auto adaptive learning for AI/ML Models
CSPs live in a world of changing data patterns. A data pattern that was learned last month might not be valid this month. Hence it is essential to keep in mind how these changing data patterns can be quickly fed to ML algorithms. Auto adaptive learning systems are a type of online or feedback learning system that can adjust to changing patterns and provide better predictions in the long term. It is essential to have such a framework to re-engineer the data features or re-learn the data trends as and when the business scenarios change.
4. Desired Architecture of AI/ML Model
While deciding upon the AI/ML model, it is recommended to choose a flexible architecture that is scalable horizontally and vertically to adapt to the changing business load scenarios. Increasingly CSPs are adopting cloud architectures with Kubernetes backed nodes to do processing. Such architecture will ensure planet-scale reliability as well as on-demand elastic scalability if business use cases demand.
5. Metrics for measuring success
CSPs must have clear metrics to measure the success of the AI/ML models combating fraud. Some of the key metrics that need to be measured are:
- The accuracy of the AI/ML model
- The efficiency of the AI/ML model
- Coverage provided by the AI/ML model
- Finding the unknown unknowns
To achieve the above, fraud management teams need to ensure that the models are trained and tuned accordingly.
6. Pilot use case
CSPs need to prioritize which fraud types they need to tackle first. First, it would be ideal to choose a use case that is highly prevalent in the existing business scenario rather than the one with a very low probability of occurrence. It is also important to note that while considering this use case, the model should have enough past historical data to learn from to generate accurate results. Considering such a use case would help assess the efficiency of the model compared to how a traditional FMS model would respond to a similar scenario.
7. Open-source community integration
With a rise in the number of open source communities, various open-source algorithms are readily available along with their libraries, which can be adopted for the model that is being built. It is critical to then fine-tune the micro characteristics of these algorithms to meet the business requirements. Readily adapting the models can help CSPs reduce the time that needs to be invested in building the model and be upscaled easily.
8. Support of different models
One of the key aspects to cover while drafting the RFP is to look at what kind of machine learning models can be supported. To start with, CSPs can consider a supervised machine learning model. Further down the line, they can move into exploring the unsupervised, semi-supervised, and deep learning models. Deep learning models can effectively deal with image processing business requirements, such as Contracts monitoring or P.O.S fraud detection.
9. Latency and scalability
Considering a large amount of data CSPs have access to, it is recommended to have low latency and highly scalable models to deal with the large datasets much faster without consuming ever-growing amounts of resources like memory. Also, it is essential to be mindful that the impact of latency on the business scenario can be different for different use cases.
10. Roadmap & Vision
It is suggested that CSPs should have a roadmap and vision in place for achieving the next 3-5 use cases aligned with their business goals. This needs to be done with a specific time frame in mind. The automation procedures would also free up critical analyst resources that are currently stuck up in mundane tasks, which could increase the coverage of the fraud management function to newer digital products.
The above factors are an indicative list, highlighting the key considerations while drafting an RFP. If CSPs wish to augment their fraud management strategy and are looking to roll out an RFP that aligns with the modern-day needs of a Proactive Fraud Management Solution that can optimize and accelerate the digitalization path, now is the time to act. This will enable CSPs to fight against fraud by providing adequate protection that is needed to tackle issues upfront in a preventative manner to avoid loss impacts.
Planning to roll out an RFP any time soon? What are your thoughts about the above points?