A medical note can say that a patient needs a referral, another appointment and a follow-up call. Writing those instructions does not mean anyone has completed them.
That gap helps explain Tandem Health’s next chapter.
The Stockholm-based company announced a $100 million Series B on September 14, led by the Scaleup Europe Fund, managed by EQT, with continued backing from Kinnevik, Northzone, Amino Collective and Visionaries. The financing brings its total funding to $160 million and supports European expansion alongside a broader ambition: moving from an AI medical assistant toward software that coordinates more of a clinic’s work.
Founded in 2023 by Lukas Saari, Oliver Åstrand and Oscar Boldt-Christmas, Tandem now reports use across more than 10,000 care organizations in 14 European markets. Its next challenge extends beyond producing a better note: keeping the work connected to that note from fragmenting across people and systems.
The workflow an AI scribe had to rebuild
Tandem’s starting point was straightforward. Capture the consultation, generate a structured medical note, let the clinician review and edit it, then transfer it into the electronic health record. The same encounter can support documents such as referrals and patient letters.
But automating documentation exposed a less obvious problem.
As the founders explain in their product discussion, dictation had served two purposes. Doctors recorded what happened during the appointment, but also told secretaries what needed to happen afterward. Removing the dictation step could remove that handoff too.
Tandem’s own Secretary release describes the resulting friction: clinicians had to switch tools to delegate work, losing context along the way. Tandem Secretary reconnects the note with the task, its owner and its status. A secretary can complete the assigned work, add an update and mark it done.
The human role remains. What changes is the connection between the clinical record and the person responsible for the next action.
A faster note solves a writing problem. A reliable handoff solves a coordination problem.
A broader assistant, with different stages of availability
The product expansion follows the consultation rather than treating documentation as an isolated task.
Tandem’s roadmap spans preparation before the visit, documentation during it, coding and guidance, and the work that follows. Its March roadmap identified the AI Scribe, Coding Assistant and Secretary interface as live, while listing Visit Preparer and broader Task Management as forthcoming capabilities.
Clinical Decision Support subsequently entered a selected-customer rollout in June. Within Ask Tandem, clinicians can ask questions using the context of the encounter and receive answers linked to supporting sources, including clinical guidelines and their organization’s uploaded protocols. The company states that nothing is applied without clinician approval.
These distinctions matter. Drafting a referral, suggesting a code, retrieving guidance and completing a follow-up are different actions. Evidence that one works does not establish that the entire process can run autonomously.
The reviewed releases also do not establish universal availability of every capability described in the founders’ roadmap.
What Tandem’s clinical evidence actually measures
Tandem has published evidence beyond testimonials, but the findings need to be read at the right level.
A peer-reviewed study published in July examined editing metadata from 236,153 notes generated by 1,295 clinicians within Sweden’s Capio Ramsay Santé. Separately, 177 fully onboarded clinicians supplied retrospective survey responses about their experience.
Those respondents estimated that documentation time fell from 6.69 to 4.72 minutes per note, a reduction of approximately 29%. That is a finding about self-reported documentation time per note, not proof that every clinician recovered 29% of the working day.
The study was observational and focused on sustained adopters. Its authors caution that the findings may not generalize to clinicians who stopped using the product or never fully adopted it. It therefore supports an association with improved documentation experience, rather than a universal or causal claim about better patient care.
A separate customer example addresses a different problem: incomplete coding.
In Tandem’s published account of a two-month pilot at Aleris Ultragyn, 20 doctors actively used the system. Among 841 examination visits with confirmed coding suggestions, none required a subsequent change by the coding controller. Doctors reviewed and approved the suggestions before they proceeded.
That is useful evidence about documentation completeness in one clinic. It is not a claim of perfect diagnosis, nor a substitute for independent evaluation across other settings.
The important product details make AI easier to check
Some of Tandem’s most consequential changes are less prominent than the promise of a medical assistant.
Its March release notes introduced links from sections of generated notes back to the transcript passages supporting them. Clinicians can inspect the source rather than accept a polished summary without context. The same update added note-transfer integrations with Epic and Oracle Health Millennium.
These features address two practical questions: Where did this statement come from? And how does the approved result reach the existing patient record?
Tandem also lists its AI Scribe, Coding Assistant and Clinical Decision Support as CE-marked Class IIa medical devices under the EU Medical Device Regulation. Its security documentation states that patient data is processed and stored in Europe, consultation audio is deleted after transcription, and patient or personal data is not used to train its AI models. Those are company disclosures, not an independent security assessment.
The company says it now has more than 75 in-house clinicians helping develop and implement its products. Its clinical-team directory places medical professionals in operational, implementation and compliance roles, bringing clinical knowledge into the work of deploying the software.
That combination is central to the wider ambition: software that assists with more clinical work also needs to make its outputs easier to inspect, approve and correct.
The next test is completed work, not generated text
Tandem’s funding gives it room to pursue a larger role in AI-supported clinical workflows. The harder task is demonstrating that the broader system delivers value beyond the documentation gains already studied.
For providers, useful measures would include whether approved tasks reach the right person, whether follow-ups remain visible until completion, and how much checking and correction the system itself requires.
Those questions follow directly from the problem the founders encountered. Automating one step can expose the informal coordination that step previously carried.
Tandem began by helping clinicians finish their notes. Its larger opportunity is helping care teams finish the work those notes describe.
The next milestone is not simply a consultation recorded. It is a clinical plan that remains connected to the people responsible for carrying it out.
