VC funding for AI tools in healthcare was The projected $11 billion figure for last year is a sign of the widespread belief that artificial intelligence can transform a sector.
Many healthcare startups are using AI to improve efficiency by automating the administrative tasks that surround and enable patient care. Hamburg-based Elea fits this mold in general, but it is starting with a niche that’s been overlooked and underserved — pathology laboratories, whose work involves analyzing patient samples to detect disease. From there, it believes it will be able scale up the voice-based AI agent-powered system it has developed to boost labs productivity, to achieve global impact. It can also be used to accelerate the output of other departments in healthcare by transferring its workflow-focused approach. Elea’s first AI tool,
is designed to revolutionize the way lab staff and clinicians work. It replaces legacy systems and other established ways of working, such as using Microsoft Office to type reports. Instead, it uses an “AI operating-system” that deploys speech to text transcription and other forms automation to “substantially shrink” the time taken to output a diagnose. Elea claims that after a half-year of using its system with its first users, it has been able reduce the time taken by the lab to produce half their reports to just two days.
Dr. Christoph Schroder, CEO and cofounder of Elea, said that the step-bystep, often manual workflow in pathology laboratories could be improved by using AI. “We turn this around — and all the steps are much, much more automated,” he says. ” [Doctors] Speak to Elea. [medical technical assistants] Speak to Elea. Tell them what they see and what they want to use it.”
Elea is the agent. She performs all tasks in the system, prints things, prepares slides, for instance, the staining, and all that stuff — so [tasks] can go much, faster, much, smoother. The AI OS will orchestrate all tasks.
The company is fine-tuning large language models with specialist data and information to enable core capabilities within the pathology lab context. The platform integrates speech-to text to transcribe voice notes of staff — and “text-to structure”; this means the system can convert these transcribed notes into active directions that power the AI agent’s action, which can include sending lab kits instructions to keep the workflow moving. Schroder reports that Elea plans to create its own foundational model of slide image analysis as it works towards developing diagnostic capabilities. For now, Elea is focused on scaling up its initial offering.
According to the startup’s pitch, what would take labs two to three weeks to complete using conventional processes could be completed in just a few hours or days. This is because the integrated system can compound productivity gains and replace things like the tedious back and forth that can occur when manually typing up reports. Human error and other workflow quirks are also a major source of friction. The system is accessible by lab staff via an iPad app, Mac application, or web application — offering a variety touch-points for different types of users.
According to Schroder, the business was founded early in 2024, and its first lab was launched in October after spending some time in stealth developing their idea in 2023. Schroder has a background applying AI for autonomous-driving projects at Bosch, Luminar and Mercedes.
Dr. Sebastian Casu, the startup’s CMO, has a clinical background. He spent more than 10 years working in emergency departments, intensive care, and anesthesiology. He was also a medical director at a large hospital group. Elea has signed a partnership (it won’t say which hospital group) with a major German healthcare group that processes 70,000 cases per year. So far, the system has hundreds users.
Schroder says that more customers will be launched “soon”and it is also looking to expand internationally, with an eye towards the U.S.
Seed funding
This startup has revealed for the first time that it raised EUR4 million in seed funding last year, led by Fly Ventures. The money was used to build its engineering team and deliver the product to the first labs.
Compared to the billions of dollars in funding that are being thrown around in the industry, this figure is quite small. Schroder says that AI startups do not need hundreds of millions of dollars and armies of engineers to succeed. Instead, they should use the resources available wisely. In the healthcare context, this means taking a departmental approach and maturing the targeted use case before moving onto the next application area.
Elea will shift gears to actively market to get more labs to buy in, instead of relying on word-of mouth marketing they started with. He tells us that their approach is different from the competition in the healthcare industry. “It’s a spot-solution versus vertically integrated.”
We built it into our own pathology operating system, or laboratory information system. This means that users don’t need to use a new UI. It just speaks to Elea and tells it what it sees, what it wants to accomplish, and what Elea should do in the system.
He also says that you don’t need as many engineers as before. You only need a dozen or two dozen really good ones. “We have about two dozen engineers on the team… and they are able to do amazing things.”
The fastest-growing companies you see today don’t have hundreds or even dozens of engineers. They have just one, two dozen experts who can build amazing things. “That’s our philosophy, too, and it’s why — at least initially — we don’t need to raise hundreds of millions,” he says.
It is definitely a paradigm change… in the way you build companies.
Scaling a workflow mentality
Elea chose to start with pathology laboratories because the market is “extremely international” and the global lab companies are enhancing scalability of its software as a services play. This is in contrast to the fragmented situation surrounding the supply of hospitals.
For us, it is super interesting, because you can build an application and scale that already — from Germany to the U.K. or the U.S.,” says Schroder. Everyone is using the same workflow, thinking and acting the same. If you solve the problem in German, which is great with the current LLMs [and other languages like Spanish] then you can also solve it in English[and other languages like Spanish]… So it opens a lot of new opportunities.”
He praises pathology laboratories as “one the fastest growing areas of medicine” — pointing to developments in medical science such as the growth in molecular pathsology and DNA sequences that are creating a demand for more types and frequencies of analyses. All of this means that labs will have more work to do, and they will be under more pressure to be more productive.
Elea may move to other areas of AI application in healthcare once they have matured their lab use case. For example, they could support hospital doctors to capture interactions with patients. However, any other applications that they develop will also be focused on workflow.
He says that they want to bring a workflow mindset where everything is treated as a workflow task and at the end there is a final report that needs to be sent. In a hospital setting, he adds, “we would not want to go into diagnostics, but we would really focus on operationalizing workflow.”
Challenges
How about accuracy? Healthcare is a sensitive use case, so any errors in AI transcriptions could have serious consequences. For example, if Elea’s reports are not in line with what a doctor says or what Elea hears when checking for cancerous tissues in a biopsy. Schroder said that they are currently evaluating accuracy based on things like the number of characters that users alter in the reports generated by AI. He says that between 5% and 10% of the time, manual interventions are made in automated reports. This could indicate an error. He also says that doctors may have to make changes due to other reasons, but they are working hard to “drive down the percentage” of manual interventions.
He argues that, ultimately, the responsibility for the AI outputs rests with the doctors and staff who are required to review and approve them. This suggests that Elea’s workflow doesn’t differ much from the legacy processes it was designed to replace (where, say, a doctor would type up a voice note by a person and such transcriptions may also contain errors) whereas “now, the initial creation is made by Elea AI and not by a typing assistant”
Automation may lead to higher throughput volumes, but this could put pressure on human staff to perform such checks, as they are now dealing with more data and reports. Schroder acknowledges that there are risks. He says they’ve built in a safety net feature where the AI will try to spot any potential issues – using prompts to encourage doctors to look again. “We call this a second set of eyes,” he says, adding that “we evaluate previous findings reports and compare them to what [the doctor] is saying right now and provide him with comments and suggestions.”
Concerns about patient confidentiality could be another issue for agentic AI which relies on cloud processing (as Elea) rather than data remaining in-house and under lab control. Schroder says that the startup has addressed “data privacy” issues by separating diagnostic outputs from patient identities.
Schroder says, “It is always anonymous throughout the process — each step only does one thing — we combine the data where the doctor can see them.” “We use pseudo IDs in all our processing steps. They are temporary and are deleted afterwards. But when the doctor sees the patient, the data is combined on the device.”
“Our main customer is a publicly-owned hospital chain in Germany, called critical infrastructure. We had to make sure that everything was secure from the perspective of data privacy. They gave us the thumbs-up.”
āUltimately, we probably achieved more than was needed. It’s always better to be safe, especially if you are handling medical data.
