Engineering challenges at Alan in 2024
Co-authored by Hugo Venturini
We regularly get questions from candidates about what’s cooking at Alan. They want to know more about the journey they’re about to embark on — the tech challenges we face and where they would eventually contribute if we hire them.
Alan’s mission is to help people get extra years of fulfilling life, and to help companies by providing the best Return on People investment. Alan is the health partner who prevents, insures, and cares daily. We want to continue streamlining the health insurance experience for our members and expand our insurance & healthcare offering to new markets and countries.
This raises many technical challenges from an engineering perspective. We’ll present some of them in this article. There are too many to fit in only one article! The remainder of the article focuses on three of our main initiatives in 2024:
- Expand: As we’re considering expanding our health insurance services to new markets, we are working on using global reusable components from one country to another. We‘re also scaling our automatic document processing capabilities and boosting the productivity of our support team with AI-enabled internal tools.
- Explore: We’re currently exploring new health partner services, including payment solutions for health spending and automating the support to our members using state-of-the-art chatbots.
- Run: We’re building internal tools to empower Alaners with AI, scaling our infrastructure to support a growing number of members, and - as always - improving our security and privacy.
Expand — Globalisation
In the early stages of Alan’s expansion, the decision was made to create separate codebases for each country we ventured into. This approach allowed us to quickly adapt and tailor our applications to meet the unique healthcare and regulatory requirements of France, Belgium, and Spain. However, as we continue our growth and are considering expanding to new markets, the time has come to reassess this strategy.
The initial choice was to enable quick iterations by starting from scratch…or almost. We followed an SDK-Driven-Development, the need for rapid iteration, and sharing common libraries, e.g. HTTP requests and base queries for SQLAlchemy. While this strategy served its purpose, it led to increased maintenance efforts, duplicated work, and challenges in sharing innovations across different markets at a business level.
Alan has now reached a pivotal moment where merging these separate codebases into global and configurable components has become imperative. This consolidation will enable us to streamline our development processes and reduce redundancy. By creating a shared foundation, we can more efficiently introduce new features, ensure consistency across all markets, and better leverage our collective knowledge and resources.
The transition to a single, modular codebase will not only enhance our operational efficiency but also strengthen our ability to deliver a cohesive and seamless experience to our users, regardless of their location. As we embark on this journey, we are committed to maintaining the flexibility to cater to local needs while capitalising on the benefits of a unified global platform. This strategic shift is a significant step towards achieving our vision of making healthcare seamless and accessible to everyone, everywhere.
Expand — Scaling Processes with LLM
Alan has implemented several AI-driven processes to scale operations and enhance efficiency across various domains.
A pivotal area of focus has been the enhancement of document parsing capabilities, a cornerstone for automating and streamlining insurance claim processes. We originally started with classic regex parsers or parsers based on Named Entity Recognition approaches. Alan has since focused on leveraging Large Language Models (LLMs). We have already leveraged them to reach 50% automation at the end of 2023. In 2024, we want to explore the potential to fine-tune these models to our specific needs as well as the possibility to leverage advanced OCR techniques for table and layout extraction from insurance documents. Our goal is ambitious yet clear: to achieve 75% automation by the end of 2024. By testing and integrating technologies like GPT-4, and eventually multimodal LLMs, we’re not just keeping pace with innovation; we’re aiming to set new standards in the insurance industry.
We have developed a platform called Smart Support to enhance the productivity of customer support resources by providing context-aware, smart and extensible tools for ticket resolution. This platform aims to accelerate the resolution of conversations with members by providing accurate context instantly and guiding agents through resolution autonomously, thereby improving their productivity. In its initial phases, the focus has been on delivering numerous tools to correctly assign the conversation to the best support team, protecting our SLA, as well as adding internal tooling to aid in the resolution step. Future priorities include consolidating these tools and deep diving into specific usages during different phases of answering a conversation. We are also exploring new opportunities for AI and non-AI tools to accelerate resolution and continue the migration of existing dependencies to newer versions.
Expand — Onboarding larger companies and more configuration
In 2024, Alan faces several technical challenges to tailor products and services to meet diverse and specific demands of large companies. There is a growing need to interface with more third-party solutions such as HR and payroll systems (HRIS). This includes the ability to handle mass onboarding and registration. This is why it’s important to have processes that scale and run smoothly.
We also need to adapt to the growing need to provide more configuration and parameterisation of our services. For instance, some companies require flexible billing and pricing options with new rules for employer contribution calculations or price structures.
On the coverage side, the needs of our members are also addressed with more possible options to make their coverage more configurable.
The tricky part here is to meet these new challenges and make our system more configurable. The key to success is to find the right trade-offs between generic design, without adding too much abstraction complexity, and custom specifics that are hard to articulate and hence to maintain.
Explore — Payment solutions
The payment solutions represent a significant engineering challenge for Alan in 2024. This innovative initiative aims to provide members with a seamless healthcare payment experience, allowing them to access necessary healthcare without worrying about upfront costs, which can cause problems for low-income salaries. Members can use this payment solution at healthcare providers, with Alan advancing the costs and later charging the member’s account for any out-of-pocket expenses.
Currently, a visit to the doctor typically involves paying out of pocket initially, then seeking reimbursements from the national social security (“sécurité sociale”) and then from personal health insurance, such as Alan. It’s often automatic, but it’s the member’s responsibility to make sure the national and personal health insurance receive everything to get reimbursed. This adds a mental load to members. Furthermore, things could go wrong, for example forgetting to send the “feuille de soins” to the social security or “télétransmission” issues between the social security and Alan.
From an engineering standpoint, several challenges are being tackled. These include reconciling care events in the claim engine with bank transactions to compute out-of-pocket costs accurately. There is also a focus on ensuring that the solution is used exclusively for legitimate healthcare purposes. Additionally, the implementation of secure communication with payment providers is crucial. Moreover, it’s essential to guarantee that transactions can be processed in real-time to ensure efficiency and user satisfaction.
Explore — Automated resolution
We continue exploring how to use AI to safely answer members’ questions in real-time. We are building specialized AI engines that are trained to solve the most frequent questions such as Coverage, Quotes, or how to transfer documents, as well as a generic AI engine that leverages the Care Knowledge Base to answer the long tail of simple questions. A conversational experience (the Assistant) uses the most appropriate AI engine to answer a member’s question.
As with any exploration phase, especially with no human in the loop, we are also investing in a scalable Review platform to evaluate the quality of the AI answers to verify answer acceptability a posteriori, improve our guardrails, and iterate on improving the overall quality of our answers.
Run — Scaling
Alan is growing fast in terms of team and number of members. We ship new services and features to our members regularly. But we also need to keep the lights on. We are making sure that production runs seamlessly while improving its reliability, robustness and member user experience.
We also have to adapt existing systems to new needs as they arise. For example, we are dealing with new user lifecycle management issues: members transferring from one company to another or members who were beneficiaries (e.g. partners) before joining a company and becoming a main member. The initial system was not designed for these cases and we need to adapt it iteratively.
At the same time, we need to continue scaling our infrastructure, improving its performance and making it more reliable. Reliability has become crucial with more real-time services such as the Alan Clinic chat and online medical consultation. This is why we also want to invest in our observability tools. Our objective is to improve visibility into the availability and performance of our services and products.
Run — Empowering Alaners with AI
In 2024, every bit counts towards our ambitious path to profitability. Every euro, every minute not spent on a repetitive/low-value task, gets us closer. This is why Alan invests in internal tooling to boost the efficiency and productivity of Alaners and become a fully enabled AI company.
As you may know, everything is written, shared and open at Alan. We rely on this knowledge database that spans across different tools such as Notion, GitHub, and Slack to build knowledge tools. These tools leverage LLMs to increase productivity and answer company-related questions with context-aware responses such as Dust.
We are also investing in new self-hosted tools for automation - like Zapier, but with access to our complete data model to build more customizable automation flows and lower the technological barrier to entry for all Alaners.
Run — Security
Our security team aims at obtaining the ISO 27001 certification, showcasing our commitment to Governance, Risk Management, and Compliance (GRC). For that, we are structuring and organising our security work much more formally, and making it threefold:
- GRC is addressing the ISO 27001 certification, which means creating an entire organisation to deterministically and reliably deliver on continuous improvement of our security posture and of our management of security risks.
- Security engineering is securing our interactions with members, in the context of intense cyber threats (cf Almerys/Viamedis/France Travail data breaches). This translates into multiple technical initiatives such as improvements to user authentication (including a push towards more 2FA adoption) as well as new tools for secure file sharing.
- SOC is timely addressing vulnerabilities.
Wrap-up
Building on the success of 2023, Alan is aiming to grow its member base by over 35% in 2024. It makes the technical landscape at Alan very dynamic, with initiatives spanning across the Explore, Expand, and Run framework. In Explore, Alan is innovating with payment services to simplify healthcare transactions for members and developing a health benefits platform. Expand focuses on globalizing health insurance services, enhancing document processing capabilities with AI, and automating member support with advanced chatbots. Run is dedicated to internal tool empowerment through AI, scaling the infrastructure, and bolstering security and privacy measures.
As Alan moves forward, it remains focused on its mission to streamline the health insurance experience, making healthcare accessible and seamless for everyone, everywhere. This journey is not just about building a company; it’s about shaping the future of health through innovation, efficiency, and expansion.
