The Center for Data Innovation spoke with Abdel Mahmoud, founder and CEO of Anterior, a company based in New York City that provides an AI platform to streamline administrative tasks in healthcare. Mahmoud discussed how Anterior’s AI agent, Florence, automates the process of issuing medical authorizations, how synthetic medical data helped train the models powering Florence, and how advances in generative AI will shape the future of Anterior.
Martin Makaryan: What does Anterior do?
Abdel Mahmoud: Healthcare insurers in the United States hire tens of thousands of clinicians like nurses in full-time administrative positions who often perform administrative tasks that automation could help streamline. For example, a nurse will spend hours sending and receiving faxed medical records to authorize a certain medical procedure with the patient’s health plan, but a smart assistant that can do this task frees up significant time for the nurse to focus on patient care, instead of mundane tasks. Our mission is to provide clinicians with the tools to automate and streamline these administrative tasks, such as preparing medical records, finding pertinent patient information, and even performing preliminary clinical reasoning. We help healthcare organizations boost productivity as nurses can focus on reviewing the smart agent’s work and approving or overruling it, reducing workflows that take hours today to minutes.
Florence is our solution to these problems in healthcare. Florence is a smart assistant that uses generative AI to streamline administrative operations with and within health insurance organizations representing over 50 million people. Florence is extremely generalizable and works across the full scope of health insurance workflows, such as prior authorization, which are approvals that patients need from a health plan before they can receive a service or a treatment. Millions of Americans face long wait times as prior authorizations take significant time, leading to issues like patient dissatisfaction or sometimes even preventable deterioration in health. Florence was able to instantly and accurately approve medical care for 76 percent of medical prior authorizations. These authorizations would have otherwise taken lengthy human review, which could have delayed treatment by days or even weeks for some patients.
Makaryan: How did you train the AI models that power Florence?
Mahmoud: For many health insurance organizations, training models on patient data is a red line. Instead of training our AI models on patient data, Anterior fine-tuned existing large language models (LLMs) using high-quality synthetic clinical data that an internal team of clinicians and machine learning engineers created. We ensure data quality with a combination of clinical experts performing manual reviews and AI-assisted evaluations.
Makaryan: What sets Anterior apart in the industry?
Mahmoud: What distinguishes Anterior most in the industry is Florence’s ability to reason through the most complex clinical cases while integrating easily into the existing digital infrastructure of our clients. In healthcare, performing various functions can be slower and less efficient despite the abundance of available technologies or software. Instead of positioning Florence as another block in their tech stack, Florence integrates easily across many other industry-standard software. Thanks to this integration, Florence can go beyond just performing discrete workflows and instead, suggest the next best action at every step. For example, after reviewing a patient’s prior authorization request, Florence uses the same medical data to flag high-risk patients for additional care management. Florence also has a very user-friendly interface, unlike many other software options in the market, and we continuously use input from nurses and experts to refine Florence’s interface.
Makaryan: What is a challenge you have faced as a data innovator?
Mahmoud: A significant challenge is determining ground truth in healthcare. Two experienced clinicians can disagree on the right answer in a given medical situation, which makes training and evaluation of an AI tool for this field difficult. In designing a tool for such a high-stakes application, we have had to quantify uncertainty and create safeguards. Another challenge has been thoughtfulness about where to deploy cutting-edge AI technologies. There is a ton of low-hanging fruit on the back office and operational side of healthcare, which is where we are focusing to transform how clinicians perform their duties.
Makaryan: Do you plan to expand Anterior’s services globally?
Mahmoud: Today, Anterior focuses on helping U.S. healthcare insurers save time and cost, but we see many opportunities in the future to help the global healthcare ecosystem beyond the United States. Many healthcare challenges are similar across the world, and while every market has different incentives and structures, many of the underlying problems are the same. We believe that patients need clinicians to spend time where they excel—focusing on patients, not paperwork. AI agents like Florence can handle administrative and mundane tasks, helping boost healthcare systems everywhere in the world, which indicates a potential for Anterior to grow and expand into new markets.
We are also seeing continued and impressive progress in generative AI, which has expanded our vision for its immediate and long-term impact. Trends like increased multi-modality, advanced agentic systems, and models that can process more data while executing sophisticated reasoning have opened new doors for us. Generative AI bridges the gap between human reasoning and computational strengths. While we cannot precisely forecast what new advancements in AI will happen in the next few years, we expect to integrate them into our product for healthcare organizations in a responsible manner.
