For insurance companies that want to utilize AI technology to improve their claims processing, one of the biggest barriers to success is often their reliance on outdated legacy systems. By partnering with an experienced Insurance Software Development company, these carriers can gain access to the knowledge and resources required to overcome these challenges, allowing them to integrate AI technology into their current systems more easily. Working together with an insurance software developer will also allow carriers to realise the full benefits of AI through automation, achieving 50 to 80% faster processing times, 30 to 50% lower costs, and greatly improved customer service, without sacrificing compliance or data integrity.
Legacy Systems in Insurance: A Closer Look
Many insurance companies rely on legacy insurance systems that were developed using COBOL and run on mainframes or early client server architectures as their backbone for policy administration, as well as claims processing. Many of these systems are designed primarily for batch processing and have built in transactional reliability; however, what they lack is the ability to provide application programming interfaces (APIs), real-time access to data and scalability, which are necessary to enable the use of AI. Data is often stored in proprietary formats in isolated databases, which limits the ability to create connections between the different systems. While legacy systems may provide a stable environment for the way that traditional insurance operations have been conducted, these systems also create very significant barriers to AI adoption and make it difficult for insurance companies to use machine learning models that require clean data, streaming data, and low latency for inference.
Key AI Technologies Transforming Claims
While the emergence of artificial intelligence technology creates efficiencies within the entire claims process through automated data extraction, decision making and workflow orchestration, it is also important to have modern data pipeline architectures capable of supporting these innovations since current legacy architecture systems are not conducive to the implementation of such technologies.
ML
Machine Learning enables predictive analytics regarding future claim outcomes, identifying fraudulent activity patterns, optimally directing claims based on a number of criteria and providing at least a 90 per cent level of accuracy for each of these types of predictive analysis models.
NLP
Natural Language Processing provides the capability to extract relevant insight from a variety of claim source documents (.example. .e-mail, .pdf, and/or voice notes); which makes up approximately 98% of the accuracy of data extracted by a Natural Language Processing type model through its summarizing the incident, validating coverage, and processing medical terminology used during integrating AI in underwriting process. As paper documents are stored on legacy systems that do not have a searchable format, these legacy systems cannot be used effectively for training purposes on Natural Language Processing models.
RPA
Robotics Process Automation handles repetitive tasks such as entering data into core systems. By combining Intelligent Robotics Process Automation with Machine Learning, end-to-end process automation can occur. However, because of the limitations of existing legacy user interfaces, Robotics Process Automation will not function as intended and relies on cumbersome screen scraping to provide the needed information required for operation.
Optical Character Recognition
Optical Character Recognition powered by Artificial Intelligence can convert handwritten documentation or documents of poor quality into structured electronic data sets. In addition, approximately 99% of that data can be converted with a high level of accuracy; however, legacy repository systems will have Un-searchable TIFF files, and therefore, these documents cannot be processed through automated means.
Computer Vision models
Computer Vision models allow for the analysis of both photos and video of damaged property to be estimated within an accuracy range of 5%. While drones can be used for remote assessments of property damage, existing systems that support claim processing are unable to process image file streams or interface with Computer Vision API systems due to their tight integration with each of the repair estimating systems currently being used by many different insurance carriers.
Key Challenges Legacy Systems Pose in Implementing AI
Technical, operational and organisational barriers related to legacy technology make it difficult to use AI effectively within an organisation.
Data Fragmentation and Inconsistent Data Quality
The majority of claims data resides on various mainframe systems, excel spreadsheets and departmental systems that have different formats used to store them. Missing lineage tracking means that there is a lack of ability to perform data cleansing for ML training. This is cited as the most significant barrier to AI by 57% of insurance companies.
Outdated Technology Not Suitable for Modern AI
Mainframe systems process claims data in nightly batch processing and do not have the ability to stream real-time information for AI inference. Limited CPU/RAM resources prohibit the use of local models, meaning that all data must be exported to a cloud environment at significant expense.
Integration Has a High Degree of Complexity
There are no REST APIs for connecting systems, thus all connectors created must be custom-built and will ‘break’ with every core update. Middleware development costs can reach millions of dollars and changes resulting from this development frequently ‘cascade’ through very brittle integrations.
Regulatory and Compliance Risk
Training AI on biased legacy data creates a risk for discrimination lawsuits. Companies that maintain inadequate audit trails are in violation of NAIC “explainable AI” regulations. The training of AI Models for use in Underwriting requires a central repository of data provenance that is present across all siloed sources.
Change Management and Workforce Resistance
Adjusters are worried about the automation of their jobs and approximately 40% of them represent resistance to the pilots of AI technology, slowing down the efforts to establish pilot testing. Additionally, the legacy user interface systems frustrate employees who were trained to use modern dashboards.
Strategies to Overcome Legacy System Challenges
Here’s a list of ways proven modernization scenarios can allow incremental AI adoption and remove the “big bang replacement” requirement:
– API-Led Modernisation:
The rest methods expose legacy components via API Gateways (cf. REST) as opposed to removing and replacing a full legacy platform as it would take many months or years to transition from a closed and restricted access model to a completely open model.
– Data Lake Approach for Keeping Legacy Data:
A data lake will provide consolidated access to the vast majority of legacy data for ML/AI. Data profiling via ML is used to auto-generate quality checks and balances. Several vendors such as Snowflake and Power BI help to create AI-compatible datasets when transitioning away from a traditional system.
– Middleware and Integration Platforms for Legacy Systems:
An iPaaS (Integration Platform as a Service) such as MuleSoft and Boomi have over 200 Mainframe/Cloud Connectors; (Apache) Kafka allows for real-time event streaming (from batch jobs) into AI services regardless of where the event is created.
– Phased-in Modernization (Gradual Upgrades):
The former (site) functionality of Front Surety/Fraud Claim Automation has been consistently producing near-immediate wins and justifying initial investment (ROI); Phased-in Approach allows for the next phase of implementation being concentrated around fraud and triage and replacing a minimum of 20% of old infrastructure each 3-5 years.
– Hybrid Cloud Strategies to Leverage AI Capabilities:
Hybrid Cloud “Lift and Shift” Strategy to Move Workloads into the Cloud While Retaining Data on the Mainframe and Train ML Models on Legacy Extractions with Amazon’s (SageMaker).
– Improve Security and Compliance
Data Governance Platforms such as DataRobot allow for the management and tracking of Bias/Lineage; Federated Learning addresses data security vulnerabilities related to moving records from a user location to a centralized location.
– Providing Training and Managing Change:
An Augmented Workforce Academy (e.g., Google) provides information to employees about continuing to use the legacy systems, but also provides champions from the pilot sites for promoting the use of AI.
How A3Logics Helps Insurers Modernize and Automate Claims?
A3Logics provides full support for the transition from legacy to intelligent systems:
Claims AI Accelerator: Out of the box Machine Learning, Optical Character Recognition and Natural Language Processing using the Guidewire/Duck Creek platforms through Application Programming Interfaces with an expectation of installation within 90 days
Legacy Data Factory: Automated extract, transform and load function that cleans mainframe data to be used for training purposes;
Low-Code Middleware: No-code connector that connects core systems to Cloud Artificial Intelligence;
Phased Roadmap: Projected to achieve 30% automation within the first six months and 80% by year two;
Compliance AI: Ability to comply with NAIC Regulations with a focus on explainability and identifying bias; and
The projected results are 65% cycle time deduced, resulting in $5.2 million saved.
Conclusion
The advancement of using AI in claims automation does not have to be compromised by the presence of outdated Legacy Systems. API Modernizations and Data Lake implementations, as well as Middleware Solutions and Phased Cloud Strategies, can provide 70% Straight-Through Processing; 40% Less Fraud. Predictive Risk Models using Unified Data Lakes provide additional help for Insurers in Claims Automation through AI With A3Logics, a partner of the Insurance Software Development Industry to speed up the transition from legacy systems to AI capabilities, allowing Insurance Companies to take advantage of the latest innovations while remaining compliant and focusing on their customers’ needs. Companies that are investing in Modernising their Systems today will be able to benefit from the new IA driven Business Model and lead the way in Insurance Industry Innovation in the Future.
