Innovation is difficult to achieve within insurance firms as evidenced by excessive levels of paperwork and processes customers are required to complete to make a claim or sign up for a new policy. Automation is challenging — innovation will not be an overnight process and rather requires iterative experimentation and a safe environment to test, fail, and learn from experiences.
Traditional budget approval processes for business case-based projects take time and resources to develop — failing fast is not an option, as replanning and adjusting the scope can be an expensive (and time-consuming) outcome. As a result, projects with a higher certainty of success are prioritized for funding and high-risk projects often remain stagnant, stifling innovation from the top down.
Leveraging a platform like Dataiku and starting with a carefully selected use case that has defined business value allows insurance organizations to form a foundation for future innovation. Although the first project should always be carefully selected based on internal capabilities, future projects can then proceed with a higher risk/reward payoff supporting widespread cultural change and innovation.
Key Capabilities for Market Disruption
While the insurance industry is undeniably highly regulated, organizations in this space will likely continue to embrace and undergo digital transformation, especially in the age of Generative AI, to improve operations and customer service, reduce fraudulent claims, and optimize sales and marketing efforts.
In order to truly harness AI and reap the benefits that come with it, insurance companies need a well-thought-out alignment of people (both organizational and attitudinal), processes (breaking down silos and ensuring reuse and collaboration), and technology. While the latter will be the core focus of this blog, it is by no means a magic bullet and rather one element of the bigger picture. Below are a few key capabilities that insurance firms should aim to master in order to successfully navigate market disruption.
1. Agility
The future is unpredictable, so teams need flexibility to change and adapt with the environment — the speed at which they can adapt to meet changing customer needs or environmental situations is key. Having agility reduces the impact of making informed decisions that have unforeseen consequences, each decision has therefore less effect which in itself reduces hesitance within the decision making process.
Agility is generally hampered by multiple technologies and the separation of production environments, being able to move quickly between development and production is essential. Dataiku is designed from inception to underpin the entire model workflow from design to production in a single, collaborative platform. This is based on the belief that a deployment workflow is not static, and effective collaboration does not always follow linear handovers between teams and individuals. It can, in practice, involve many touchpoints in all directions.
Enabling each of the key stakeholders (e.g. different analysts across pricing, claims, fraud, reserving, underwriting, and other business units, as well as actuaries, data scientists, data engineers, IT specialists and team leaders) to collaborate seamlessly drastically minimizes information bottlenecks and handovers. Less redevelopment increases accuracy, quality, and speed to market.
Insurance policy restrictions or artificially high quotes shouldn’t be required if models can be developed, approved, and promoted to production quickly. Allowing more dynamic underlying risk factors to be accurately priced facilitates access to a greater portion of the market, increasing gross written premium within a single market.
2. Scalability and Elasticity
When something truly innovative is created, market demand can take you by surprise. Having the ability to seamlessly scale with demand is essential to avoid restricting supply (unless you subscribe to the theory that restricted supply and the news flow that follows creates a marketing opportunity that increases demand — in which case this could be simulated!).
In Dataiku, models in production are managed by the API Node, which is designed to be horizontally and elastically scalable. By using containerized deployment of models to elastic clusters, resources can be auto-scaled up or down dynamically to handle unexpected traffic surges and maintain response times. Batch processes can also be scaled elastically with external compute resources made available for specific high compute processes. Wherever possible, data processes are offloaded into specific hardware specifically designed for data manipulation.
With more complex models and processes to provide increased accuracy, additional resources may be required. Resource constraints can either be addressed by provisioning elasticity requirements for a single model ahead of time or configuring multiple models behind a load balancer to distribute requests evenly between identical models.
3. Support and Community
Support and community are both internal and external. A center of excellence generally provides internal support and additional capabilities to an insurance company's varied user base. Support can enable other business functions to become self-sufficient by providing best practices on usage and internal consultancy for more complex developments. Frequently internal plugins, applications, or templated projects are developed centrally, with specialist resources rolling out the extended capabilities to ensure non-coders have full access to the best open source packages in a governed and controlled framework.
Foundations for Success
Insurance companies need to combat the lingering issue of legacy tools and systems, a process that can become infinitely more challenging to upskill staff if tools are difficult to use, aimed solely at the coder population, or even if the user experience involves traditional operating models focused on historical reporting.
Although these challenges can’t be solved with a tools-only approach, identifying a platform that can help insurance firms execute on their AI strategy is a critical step in the right direction and one that will take effort, discipline, and time. From a technology perspective, Dataiku can help organizations generate increased value from data and jumpstart the transformation to becoming data driven, as it:
- Provides a safe, collaborative development environment with the key tools to implement advanced enterprise capabilities in a governed and controlled way
- Includes capabilities for the entire development life cycle, from data wrangling to model production and process automation
- Facilitates collaboration externally with the best of open source communities, so teams can incorporate ever-evolving industry best practices quickly, mitigate the risk of technology-deterring progress, and avoid replicating functionalities that are readily available via open source projects
- Offers a controlled and isolated framework that ensures separation from existing production and development projects
Cultural change that looks first at what is possible and then how to integrate best practices (as opposed to what is internally achievable) helps push boundaries and redefines the definition of success. To achieve this, multiple personas need to be involved, as this enables teams to work efficiently, increasing quality and encouraging innovation (through agile developments). From a people perspective, Dataiku:
- Empowers both domain and data experts alike to access and develop a deeper understanding of data
- Enables business users to develop and productionize projects with automated scenarios through an end user computing environment (with limited IT support)
- Lets advanced users develop plugins that provide complex functionality (in a simple user interface) so less technical users can perform advanced functions, therefore extending the platform capabilities and/or supplementing missing capabilities from core operational systems
To enable innovation, users need the headspace to be forward thinking. They must have some free capacity away from fighting fires during business-as-usual activities. The solution is to free experts from mundane tasks which can quickly become boring and decrease enthusiasm. Automation is the key to reducing manual processes, creating the capacity to spend more time on strategic, stimulating projects that — with the right frame of mind — are the genesis for transformative ideas.
Adding value also isn’t limited to developing models. Powerful process automation tools can be leveraged to automate repetitive, low-value tasks, freeing up resources to perform higher-value activities and simultaneously reduce operational risk by eliminating manual processes.
From a process perspective, here’s how Dataiku can help:
- Custom integration and ETL jobs can be developed and monitored in the automation node, pushing down computation into the data infrastructure leveraging existing architecture and security
- Actions can be logged and data pipelines can be viewed in a transparent way, enabling regulatory compliance to be monitored more real-time in order to ensure no surprises if an audit arises (in which case, AI can help automate manual work such as validation of customer data and customer data security operations)
- Data exploration and visualization facilitate the analysis of large datasets that can’t be handled effectively in spreadsheets (i.e., the removal of error-prone, user-written macros and incorporation of automated workflows reduces risk and increases efficiency)
- Processes are automated on central servers using time-based triggers or are linked to changes in the underlying data, accelerating execution times and removing key person risk associated with end user computing solutions
- By utilizing a visual interface and allowing code development, it provides the capabilities for all users to make sense of large, complex datasets, build and productionize models, and design and automate complex workflows
- Enterprise-grade security and production-ready configurations enable IT to safely empower business users to work in an agile way, further benefiting efficiency and reducing unnecessary controls that stifle innovation
Within Dataiku, models are an important part of the development and implementation processes, but they do not have to be deployed to generate knowledge. In the insurance sector, machine learning capabilities can be used for alternative purposes. Through the use of model interpretability, a deeper, non-linear understanding of data can be developed, highlighting the external impacts on insurance products so teams can develop a deeper understanding of their customers. Increased customer understanding has the potential to:
- Spark ideas for new concepts or market opportunities that, if leveraged quickly, are the springboard of innovation
- Enable internal teams to develop new marketing campaigns based on feedback on support calls and social media posts
- Identify customer segments for personalization and targeted marketing
- Save time and costs by automating parts of the customer service value chai
Traditionally, machine learning models do not include insights into how or why they arrived at a certain outcome, making it hard to objectively explain the decisions made and actions taken based on these models. Prediction explanations in Dataiku open the black box by describing which characteristics, or features, have the greatest impact on a model’s outcome.
By providing the capabilities users need in a graphically-driven environment required to make sense of large complex datasets, teams can seamlessly build and productionize models and design and automate complex workflows. Enterprise-grade security and production-ready configurations enable IT to safely empower business users to work in an agile way, further benefiting innovation and reducing manual, inefficient processes and, ultimately, costs.
Successful Implementation
As with all large change projects, implementation contains risk. Most insurance companies like to minimize risk and therefore a phased approach to implementation (as opposed to a large-scale change) is preferable.
On a global scale, more and more insurance companies are beginning to augment their technological capabilities so they can enhance productivity, drive down costs, and improve security and governance efforts. In order for AI tools to truly make a lasting, holistic impact, they should be accessible to all (not dependent on the ability to code) and they should benefit the whole organization, not just a specific community. Transparency, which aids collaboration, provides the business with an enhanced understanding of how data and analytics projects are collectively developed, accelerating adoption.
We are entering a new normal for the insurance industry — one that even involves Generative AI use cases. Not investing in AI capabilities now could be a costly mistake, reducing organizations’ ability to understand and adapt in line with market needs.