Evaluation of cost efficiency and speed in processing data of claims with Amazon Nova Micro and Amazon Nova Lite
Amazon operations include the globe, affecting the lives of millions of clients, employees and sellers daily. From the wide logistics network to the front technology infrastructure, this scale is a testimony of the company’s ability to innovate and serve its customers. With this degree comes a responsibility to manage risks and address claims-if they include employee compensation, transport incidents or other insurance-related issues. Risk managers supervise claims against Amazon throughout the cycle of their lives. Documents of claims from different sources increase while claims are baked, with a single claim consisting of 75 documents on average. Risk managers are required to strictly follow the standard standard procedure of standard functioning (SOP) and to review the evolution of dozens of aspects of requirements to evaluate severity and take appropriate actions, reviewing and addressing each request fairly and efficiently. But as Amazon continues to grow, how are the risk managers strengthened to continue with the increasing number of demands?
In December 2024, an Amazon internal technology team built and implemented an empowered solution by it, as applied to data related to claims against the company. This solution generates structured summaries of requirements below 500 words in different categories, improving efficiency while maintaining the accuracy of the requirement review process. However, the team faced challenges with high conclusion costs and processing time (3-5 minutes for demand), especially when new documents are added. Because the team plans to expand this technology into other business lines, they researched the Amazon Nova Foundation models as possible alternatives to addressing cost and latency concerns.
The following graphs show performance compared to latency and performance compared to the cost for different foundation models in requirements data.
Assessment of the case of using the claims summary proved that the models of the Amazon Nova Foundation (FMS) are a strong alternative to other large border language models (LLM), achieving a significant performance with significantly lower costs and higher general speeds. Amazon Nova Lite model demonstrates strong compilations in the context of long, different and messy documents.
Settlement
The summary pipeline begins by processing raw requirements data using AWS GLUE jobs. It stores data in the intermediate buckets of Simple Storage (Amazon S3), and uses Amazon’s Simple Queue (Amazon SQS) service to manage summary work. Request compilations are generated by AWS Lambda using foundation models organized in Amazon Bedrock. We first filter out the trivial data data using a LLM -based classification model based on NOVA Lite and summarize only the relevant requests of the claims to reduce the context window. Considering the importance and summary requires different levels of intelligence, we choose the right models to optimize the cost while maintaining performance. Because the claims are summarized after the arrival of new information, we also make the results and intermediate summaries using Amazon Dynamodb to reduce the copied conclusion and reduce the cost. The following image shows a high -level architecture of the case -consuming case use.
Although Amazon Nova’s team has published performance standards in several different categories, the summary of claims is a unique case of use given its variety of inputs and long context windows. This prompted the technology team that possessed the resolution of claims to further investigate their comparison study. To evaluate the performance, speed and cost of Amazon Nova models for their specific use case, the team cured a standard database consisting of 95 pairs of requests and verified summaries of aspects. Request documents range from 1,000 to 60,000 words, with a majority of about 13,000 words (average 10,100). Verified summaries of these documents are usually short, containing less than 100 words. Inputs in models include different types of documents and summaries that cover a variety of aspects in production.
According to Benchmark tests, the team observed that Amazon Nova Lite is twice faster and costs 98% less than their current model. Amazon Nova Micro is even more efficient, running four times faster and costs 99% less. Significant improvements in cost and latency offers more flexibility for designing a sophisticated model and scaling the test calculation to improve summary quality. Moreover, the team also observed that the latency gap between Amazon Nova’s models and the other best model expanded to long Windows Contekst and long production, making Amazon Nova a stronger alternative in the case of long documents while optimizing the delay. Moreover, the team conducted this comparison study using the same quickly as the current solution in fast transportation production. Despite this, the models of Amazon Nova successfully followed the instructions and generated the desired format for processing after. Based on comparison and evaluation results, the team used Amazon Nova Lite for the use of classification and summary.
cONcluSiON
In this post, we have shared how an Amazon internal technology team praised Amazon Nova’s models, resulting in significant improvements in the speed of conclusion and cost efficiency. Looking at the initiative, the team identified several critical factors that provide key advantages:
- Access to a varied model portfolio – The availability of a wide group of models, including compact but powerful options, such as Amazon Nova Micro and Amazon Nova Lite, enabled the team to experiment quickly and integrate the most suitable models for their needs.
- Scaling and flexibility – Cost and latency improvements of Amazon Nova models allow more flexibility in designing sophisticated models and scaling the test calculation to improve summary quality. This scale is especially valuable for organizations that handle large volumes of data or complex work flows.
- Ease of integration and migration -The capability of models to follow the instructions and generate results in the desired format simplifies after processing and integration into existing systems.
If your organization has a similar use of large processing of documents that is costly and takes time, the above evaluation exercise indicates that Amazon Nova Lite and Amazon Nova Micro can change the game. These models excel in the treatment of large volumes of different documents and long context windows – perfect for complex data processing environments. What makes this particularly convincing is the ability of models to maintain high performance, while significantly reduces operational costs. It is important to repeat new models for all three pillars – quality, cost and speed. Order these models with your use and data data.
You can start with Amazon Nova on the Amazon Bedrock keyboard. Learn more on the Amazon Nova product site.
About
