Machine Learning at Alan
At Alan, we decided to invest early in Data expertise. After 3 years building the foundations of our infrastructure, our team, our data and our practice, we are in a solid place to reap the rewards of this investment by building more and more data-driven products & tools.
With more than 300,000 insured members, we have reached a critical mass that unlocks clear use cases for Machine Learning (ML), both because of the amount of data available and because of the potential impact ML can have.
Our approach: Ambitious & Pragmatic
Machine Learning is a means to an end. Like any piece of technology, we believe in using off-the-shelf, tried-and-tested tools, and relying on the rich open-source ecosystem when possible. When we realize that going to the next level requires us building our own tech, we invest in doing so.
We have developed the general conviction that full-stack approaches maximize impact, and we believe it applies to ML. Beyond iterating on the model or pushing the state of the art, what leads to great outcomes is building an end-to-end product that solves the problem as well as envisioning the lifecycle of the system in the long-term.
We don’t have “POCs”, we prefer MVPs. “Proofs of concept” usually aim at demonstrating technologies and feasibility to stakeholders. They too often gather dust, waiting for “implementation in prod” at a later time. Instead, we believe in making bold investment decisions for the long-term and shipping solutions that create value right away. We tackle the full spectrum of problems, aiming for a minimal viable product that we can iterate on. The ML operators are embedded in product teams with people from the Engineering, Product Management and Ops communities.
Like the rest of what we build, we ask ourselves what an amazing outcome looks like, and we build brick after brick to get there. We want to share two use cases of ML that will unlock clear impact for Alan.
Automating Document Processing
Document Processing is a big part of Alan’s operations. For instance, we allow our members to send pictures of documents directly from their mobile app in order to get reimbursed — Alan’s main job as an insurer.
Being able to reliably extract information from those documents brings two advantages: efficiency and delight. By processing these documents without human intervention thanks to ML, we reduce errors and costs for Alan so that we propose better reimbursements at a lower price to our members. In addition, our members can receive the money in their bank account in seconds instead of days!
We’re only at the start of the journey, with less than 35% of documents received that are parsed automatically. This problem is difficult: we have several stages (transcription, classification, parsing) that all are imperfect, and currently dependent on the OCR technologies we use. We want to double this rate in 2022!
Increasing the productivity of our Care Team
Anyone covered by Alan that had to ask a question or solve a problem with us knows that our Care Team is one of our key differentiators. We need to keep the top quality while improving the productivity of our 80 User Care Experts who receive more than 1,000 conversations per day.
We took the unconventional approach of avoiding phone conversations as much as possible, instead sticking to our “written” culture using chat and email conversations with our members. This unlocks an incredible opportunity for ML applications.
We integrated a triaging product into the workflow of our User Care Experts, leading to a radical improvement in productivity: User Care Experts now receive only questions relevant to their expertise.
In the domain of applying ML to our care conversations, the sky is the limit. While our approach is always to enhance the human, we believe we can go much further than our triaging tool, and we want to suggest full answers to the care expert. We could even answer the question _before_ the member asks using contextual information to display some smart help directly from the product!
The future: Machine Learning is disrupting Health
Looking forward, we know that data-powered applications will radically change our relationship with our health and with healthcare. It’s already the case, with clear, distinct use cases for ML like drug discovery, image processing, protein research, patient triaging, etc.
Still, we believe the future will be made of integrated, holistic products that revolve around the patient and her data — leveraging ML technologies at every step. We want to make health more friendly, comprehensive, proactive and personalized. We’re only at the beginning of the journey, but we know that ML has a big part in it.
Beyond the well-identified high impact use cases of today, we want to build a world-class ML expertise in order to prepare for the opportunities that we see on the horizon.
That’s why we’re looking for seasoned Machine Learning operators and leaders to help us scale our ML practice. If this sounds exciting to you, join our team!