How financial institutions avoid GenAI data-quality headaches

How would you feel if an airline service agent told you your fare was refundable, and the airline subsequently refused to refund your money? How about if a car dealer rep offered you a new vehicle for $1?

These examples aren’t theoretical or imagined blunders: they actually happened, thanks to GenAI.
In the case of the airline, regulators sided with the passenger, awarding him a refund despite the airline's official policy.

Control your LLM with data governance

Financial Services Institutions (FSIs) dread the prospect of suffering similar errors with GenAI. Yet when GenAI applications lack sufficient guardrails and use anything less than top quality data, all possibilities remain open.To ensure our customers’ GenAI success – and protection – here are three guidelines you should follow:

• Securely expose the large language models (LLMs) that power GenAI responses to high-quality, proprietary data.
• Train the models to respond, in accordance with your policies and as regulations for your specific use case.
• Set up your use case in a way that prevents sky-high infrastructure costs.

Our GenAI deployments have seen us partner with leaders in highly regulated industries, such as financial services and telecommunications. We have encountered FSIs that have already started. Typically, this involves setting up a retrieval augmented generation (RAG) database model to help train an LLM on their data and provide continuous access to relevant information.

Still, many FSIs stumble along the way: It isn’t easy to build a RAG database that meets FSI data governance standards without taking too long or driving up infrastructure costs. In contrast, we've also seen that those who take proven paths go further faster. This raises an important question for FSIs considering GenAI implementation…

Data platform for GenAI: DIY or buy?

To manage your data for GenAI using a RAG-based strategy, you could take a do-it-yourself (DIY) approach. That involves assembling a data platform for GenAI that applies the right RAG options to your use case. You need suitable filters, data frames and data governance tools. Watch out for misconfigurations can cause infrastructure costs to surge.

Beyond that, you must train the LLM to respond in ways that are consistent with the data, eliminating false answers and hallucinations. For instance, if there’s a 15-minute lag between an event and data availability, you’ll need to account for that in training, and ensure your LLM consumers are informed of these types of limitations.

With the complexity of each of these areas, it is likely to take an organization pivoting to use of advanced data capabilities, such as LLMs, a long time to refactor their data platform and implement a cohesive approach and strategy. This means either sacrificing training time for your models or dramatically slowing down time to market.

In contrast, established data platform vendors, such as one of our key partners Databricks, offer the controls and frameworks that you would otherwise need to assemble yourself. You can start with a ready-to-use data platform which accelerates and simplifies deployment By gaining speed and efficiency, you can devote more time and resources to training your LLM and using your data effectively.

GenAI data platform must-haves

Whether you choose a vendor solution or build your own, understanding the essential components of an effective GenAI data platform is crucial. Look for data platform technology that helps you assemble, define and monitor your RAG model without the time drain of asking data developers to start from scratch. That means the platform should deliver:

• Fast time to market: The right platform will reduce RAG database deployment time by up to 90% compared to building it yourself.
• Reduced complexity: Most FSIs do not have the resources trained and available to successfully navigate the complexity of creating a high-quality RAG model quickly. Only the largest FSIs have the in-house data development resources to do so and typically those teams are already stretched thin by other modernization and data-quality initiatives.
• Reliability: FSIs need water-tight data governance. The platform should offer proven fine-grained data governance for AI to deliver the standards required.
• Higher ROI: It is crucial to accelerate the data-quality part of your GenAI journey to gain more time for model training which can lead to exceptional AI-powered experiences for customers and business users. The right platform can help expedite that process.

Master the cloud and GenAI data details for speed and cost-savings

At Amdocs, our expertise helps customers deploy leading data platforms, such as Databricks, rapidly and cost-effectively. We've helped our customers achieve up to 50% faster time to market and up to 90% lower costs. We work with our clients’ internal teams to implement data development practices that accelerate and automate the use of LLMs and GenAI to scale at the speed of business use cases. Just as importantly, we optimize public cloud resources to avoid the misconfigurations that drive up platform and infrastructure costs. Here are some examples:

• A North American bank started their Databricks deployment in-house using highly manual development, testing, and production practices. When timetables started slipped, they turned to Amdocs. We're now migrating their critical data to Databricks and upskilling their teams to ensure long-term effectiveness and success.
• A major Asian FSI engaged us when their Databricks project costs began to spiral. Our team reduced costs by nearly 90% through various strategies such as workload tagging, rationalization and cluster size optimization.
• Beyond financial services, a communications service provider worked with us to cut Databricks costs by 55% while achieving end-to-end data visibility and improved data quality. Read the detailed case study to learn more.

Let’s make GenAI amazing

At Amdocs, we’re helping FSI leaders transform AI concepts into tangible ROI results, using RAG models as the data foundation. Talk to us about accelerating your GenAI training and implementation while maintaining the data governance standards your business demands.