On September 14, 2026, Stockholm-based Tandem Health closed a $100 million Series B led by EQT’s Scaleup Europe Fund — the fund’s first healthcare investment — with Kinnevik, Northzone, Amino Collective, and Visionaries Club returning from its 2025 Series A. That brings Tandem’s total funding to $160 million. The headline number is not what should catch a US coding compliance team’s attention. The detail buried in the announcement is: Tandem’s AI scribe, coding assistant, and clinical decision support tool all carry Class IIa medical device certification under the EU’s Medical Device Regulation (MDR).
Tandem’s software is already deployed across roughly 10,000 care organizations in 14 European countries, with about 175 staff including more than 75 clinicians. Its coding assistant generates billing code suggestions alongside the AI scribe’s note drafting, and the company says it plans to expand into a broader “AI Clinic Operating System” that adds patient-flow management to documentation and coding.
Why medical device certification is the real story
Most AI coding tools on the market today — autoscribe add-ons, CPT/E&M suggestion engines, HCC capture assistants — ship as software products, not as regulated medical devices. That distinction matters because a device certification comes with mandatory risk management documentation, post-market surveillance obligations, and a formal claims process that most AI coding vendors simply haven’t pursued. Tandem’s decision to certify three separate AI components, rather than treat coding suggestions as a low-stakes convenience feature, is a bet that regulators and buyers will increasingly expect it.
What Class IIa actually requires
Under EU MDR, Class IIa is a mid-tier risk classification: it applies to devices with moderate potential for patient or organizational harm and requires a notified body to review technical documentation, clinical evaluation, and quality management processes before the product can carry a CE mark. For an AI tool that suggests CPT codes and E&M levels, that means the vendor has to formally document failure modes — a missed diagnosis code, an upcoded E&M level, an incorrect CCI edit — and show a process for monitoring and correcting them after deployment, not just at launch.
How this differs from the US regulatory posture
The US has no direct equivalent requirement for AI coding-suggestion software today; FDA’s device framework has focused on clinical diagnosis and treatment tools, not billing code recommendation engines. That gap is exactly why Tandem’s move is worth watching: it previews a possible template for how AI coding tools could be evaluated if US regulators — or increasingly risk-averse health system compliance committees — start asking for the same kind of documented risk controls informally, through procurement requirements, even without a formal mandate.
What this means for RCM and compliance teams evaluating AI coding vendors
Whether or not a US vendor pursues formal device certification, Tandem’s approach hands compliance and revenue cycle leaders a concrete checklist for AI coding vendor due diligence:
- Does the vendor document known failure modes for its coding suggestions, including upcoding and undercoding risk, not just accuracy claims?
- Is there a formal post-deployment monitoring process, or does oversight stop at initial validation?
- Can the vendor show how code suggestions are audited against CMS Correct Coding Initiative edits and payer-specific rules before they reach a biller?
- Who is accountable when an AI-suggested code is wrong — the vendor’s documentation should say so explicitly, not leave it implied.
- Does the vendor’s roadmap treat coding as a standalone, auditable function, or bundle it invisibly inside a broader documentation product?
The broader pattern
Tandem’s certification push lands amid a string of 2026 developments that all point the same direction: agentic AI in coding is moving from experimental add-on to infrastructure that regulators and payers expect to be governed. CMS’s WISeR prior authorization pilot and the AMA’s recent moves to formalize CPT codes for AI-assisted documentation both reflect the same underlying shift — the tooling is maturing faster than the oversight framework built around it, and vendors who get ahead of that gap are positioning themselves as the safer procurement choice for risk-averse health systems.
Where this leaves US buyers
US health systems and physician groups evaluating AI coding tools today are mostly relying on vendor accuracy claims and reference calls, because there is no formal certification pathway that maps cleanly onto billing-code suggestion software the way MDR does in Europe. That leaves compliance teams doing, informally, what a notified body would otherwise do: asking vendors to produce their own failure-mode analysis, their own audit trail design, their own answer to “what happens when the model is wrong.” Tandem’s approach suggests that answer is becoming table stakes rather than a differentiator reserved for the most cautious vendors.
It also raises the bar for what “validated” should mean in an RFP. A vendor that can point to a documented risk-management file and a defined post-market monitoring process — even without a formal device certification — is making a fundamentally different claim than one offering only an aggregate accuracy percentage. Compliance officers reviewing new AI coding tools this year have a reason to ask which kind of claim they’re actually getting.
For US coders, RCM leaders, and compliance officers, the practical takeaway isn’t that every AI coding vendor needs a CE mark. It’s that “how do you prove this tool is safe and auditable” is becoming a standard procurement question, not an edge case — and vendors without a good answer will increasingly lose deals to ones that have thought it through.
Read more on the raise from Tech Funding News.
Evaluating an AI coding vendor’s accuracy and audit trail is exactly the kind of due diligence Medikode’s automated medical coding platform is built to hold up under — every suggested code comes with a documented, auditable rationale designed for compliance review, not just throughput.