Medical billing is a $280 billion industry, but almost none of it has been touched by real innovation since the software running it was built. Providers lose money to denials they can't predict, drown in manual rework and wait months to get paid for work they've already done. Nick Perry and Doug Proctor looked at this space and saw the same fragmentation and manual-workflow bottlenecks that they had spent years untangling in their previous roles at Palantir. Today, they and their team are building Candid, a revenue cycle management platform that pairs deterministic rules with agentic AI to automate the end-to-end claim lifecycle.
This foundational approach has triggered explosive market adoption. Candid is currently the top-billing infrastructure player for venture-backed digital health leaders – powering household names like Talkiatry, Tia, and Nourish – while increasingly white-labeling its tech for large, traditional outpatient networks and Electronic Medical Record (EMR) companies. Candid has driven Touchless Claim Rates as high as 97% for its customers, and the company has just closed a $120 million Series D, valuing it at nearly $1 billion. The startup has scaled rapidly with consecutive year-over-year revenue growth of 6x, 3x, and 2.5x. This momentum is anchored by an industry-defying retention rate, with the team recording just a single regrettable customer churn over their entire seven-year history.
In this conversation, Nick and Doug discuss why traditional billing systems are a black box, how they engineered a "Touchless Claim Rate" of up to 97%, and what it takes to build adaptive automation that stays one step ahead of changing insurance algorithms.
How did you start the company?
Nick: At a high level, medical billing is an extremely manual industry. It’s also massive – about $280 billion, which is roughly 1% of US GDP. The reason it requires so many people is that almost everyone is using legacy software built in the 2000s. It’s virtually impossible to modernize or automate that old software. Coming out of Palantir, it was obvious to us how to build a much more automated system that automates the end-to-end medical billing process for healthcare providers.
However, automation can be a “garbage in, garbage out” situation. If you don't integrate all the data and get the full context, the automation won't work correctly. You need that context, whether you're using deterministic rules or AI agents. Our core emphasis is on fixing things upstream so providers can submit claims correctly the first time, preventing denials entirely.
Doug: Nick had direct exposure at Palantir to how broken the payer-side claim adjudication systems were. That gave us unique insight into how frustrating, complex, and expensive this is for providers, not to mention the poor impact it has on the patient experience. My focus at Palantir was more on national security and defense. But, when you look at the foundational parts of US infrastructure – whether it's healthcare, national security, logistics, or energy – you see the same issues with fragmentation of core systems, fragmented workflows, and a proliferation of manual work. If you just try to drop AI on top of that, it fails. What makes AI impactful is the data integration and context graph underpinning it. At Candid, we rebuilt the fundamental infrastructure from the ground up. We built a data integration platform with a comprehensive ontology of all objects in the system. This allows us to empower human operators with efficient deterministic rules for highly granular scenarios, while deploying large language models and voice models in specific areas to maximize automation.
What implementations are delivering the most ROI for your customers right now?
Nick: In medical billing, value is created for healthcare providers in three distinct ways: collection rate, the percentage of dollars you actually collect from insurance. When you automatically submit claims correctly the first time, your collection rate goes up. The cost to collect is how much a provider spends to collect their revenue. When automation reduces the need for manual intervention, this cost goes down, and accounts receivable (AR) time. If claims aren't getting denied, providers get paid much faster. Whether it's a deterministic rule ensuring a claim is correct out of the gate, or an AI agent suggesting new rules to avoid future denials and making phone calls, it all drives those three metrics.
Because this process has historically been entirely manual labor, it must be tough to convince healthcare leaders to trust software over a human team. How difficult is it to get past that barrier? Are they becoming more open to it lately?
Nick: We’ve grown explosively over the last five years because the value proposition is incredibly compelling. For example, one of our largest and fastest-growing customers reduced its billing team by 40% while at the same time doubling their claim volume – that’s an effective 80% reduction of transactional work – that’s a game changing impact. When the reward at the end of the journey is that high, leaders are willing to manage change. Now that we’ve built the underlying infrastructure, we are starting to package our automation and AI agents for independent sale. Providers can integrate our agents into their existing legacy billing systems to automate manual work without ripping out their core software. That makes the transition much less significant. The proof is in our retention. We have had only one regrettable churned customer in the last seven years, and we're working hard to win them back.
That’s incredibly impressive for a seven-year track record. Who is your typical customer? Are they mostly large healthcare companies, smaller startups, or a mix?
Doug: Over the last seven years, we have discovered that the hard problems we solved for fast-growing digital health companies are the same core issues that the largest traditional players face. Today, we deploy a full end-to-end revenue cycle management (RCM) automation solution that replaces 20- to 30-year-old legacy systems. Because we built it from the ground up, our solution is entirely specialty-agnostic. We work across oncology, physical therapy, behavioral health, pediatrics, primary care, and a wide range of other specialties.
When a customer makes the leap and switches to Candid, what is the most significant operational metric that changes for them in the near term?
Doug: The most distinctive metric we track is the Touchless Claim Rate (TCR). The industry typically relies on vanity metrics like the "clean claim rate," which only tells you if a claim successfully reached the insurance company – clean claim rate is a bit of a vanity metric – it’s not hard to get close to 100%, so it can be misleading and give folks a sense of accomplishment, but in practice, it measure neither the revenue nor the cost part of the RCM equation. It doesn't account for how much manual labor it took to get it there or how it was ultimately adjudicated. Contrast that to TCR, which measures both in a single metric. We have customers consistently achieving 95%+ TCR. Before working with us, many of these providers had a 0% TCR because humans had to touch every single claim. Legacy systems can't even measure this because they don't track manual human activity. We natively track where human operators spend time, aggregate that data through machine learning models, and automatically generate suggestions for new rules to prevent future manual work.
What is a concrete example of how the AI and technology work behind the scenes to drive that touchless claim rate?
Nick: A great example involves multi-state healthcare providers. Most legacy billing systems were built decades ago for single-office private practices, so they don't handle multiple addresses correctly and default to sending every claim out under a single corporate address. This is a major problem because if a claim lists the wrong office address, the insurance company will adjudicate it as out-of-network rather than in-network. Before Candid, our customers literally had entire teams whose sole job was to log in, review a queue of hundreds of claims, and manually update the addresses one by one. Our platform automates this entirely. It dynamically overwrites and applies the correct address based on the specific tax ID and situation, so humans don't have to touch it.
Another example is time-based medical coding, such as for a therapy visit. Most legacy billing systems don't have a data field to store appointment duration because time-based coding rules were introduced more recently. Workers have to manually read appointment notes to code the claim. Because we built a modern data foundation, our system automatically pulls the appointment time directly from the integration layer and instantly codes the claim. Traditional billing systems fail here because they simply lack the data integration required to see the complete picture.
How does your system stay ahead of the changes from payors without your internal team having to constantly fix things manually behind the scenes?
Nick: At Palantir, we called these “adaptive adversaries.” Payers change their rules annually or quarterly, and they frequently introduce bugs that break existing workflows. The name of the game is building a system with a continuous feedback loop that identifies these anomalies immediately. When a batch of claims that used to go through is denied, our system detects it and determines how to introduce a new rule upstream. You have to design for constant evolution.
When you have customers running thousands of concurrent rules, you need deep visibility, versioning, and automated testing to safely introduce a new rule without breaking everything else. If your automation lacks observability, it becomes a massive bowl of spaghetti, making it impossible to trace where an individual noodle went wrong. We brought that Palantir-level data engineering and orchestration to ensure everything coexists seamlessly, even alongside human operators.
