Healthcare’s Most Overlooked Bottleneck: The Front Desk

With Annalisa Merelli and Meredith Nolan

What’s old is new again. Companies in health-tech are finding support from the current administration for initiatives that reduce healthcare expenditures, particularly if that involves using cutting-edge technologies like artificial intelligence. Companies that are driving value-based care models are finding new sources of optimism, as well as those that provide digital access.

But the question becomes: How much can you cut costs, while also ensuring there are sufficient incentives to truly transform the system we have? Sadly, money moves our system better than any mission ever can. The ACCESS model is a good example of that, which aims to change the way we treat patients with chronic conditions. The government’s thesis: Rather than pay by procedure, let’s reward providers based on outcomes we can measure.

With the government releasing pricing for the ACCESS model earlier this week, companies across the health-tech industry are wondering whether the juice is worth the squeeze. As I wrote in an analysis, those that win in this new environment will have access to patients, whether through health system partnerships or another channel, and are already heavily favoring technology over heavy services.

Without naming names, I’ve been seeing some knee-jerk reactions in our community against venture-backed healthcare, as if the sheer fact of being backed by venture capital is inherently bad. To that I say: venture-backed healthcare can be some of the highest-quality out there. A pool of capital is fueling a company to unprofitably deliver highly efficient care overseen by some of the most expensive doctors — all in the hopes of achieving efficiencies through scale. Of course, there are bad apples out there that are pursuing a growth at all costs mentality. But there’s also plenty of good. Conditions are getting de-stigmatized, patients are accessing care at home, and doctors are getting paid competitive salaries to do more entrepreneurial work. Let’s not forget that venture is also fueling some of the most impactful in value-based care, which most of us believe to our core is the future.

One last thing I’ll say here: We shouldn’t act like every entrepreneur in healthcare is some clueless tech bro. That might have been more true 15 or 20 years ago, but it’s no longer the case today. Let’s be open to smart, thoughtful people trying to improve the system we have. Is the status quo perfect? I think not.

With that, the news of the week brought to you with our HBS intern, Meredith Nolan and freelance writer Annalisa Merelli.

The news: As part of the Prevent Diabetes Act, Congress extended the ability for digital providers to participate in the Medicare Diabetes Prevention Program (MDPP) through the end of 2029. 

What is it: MDPP consists of 22 sessions that aim at promoting lifestyle changes that could help prevent diabetes.

Who can use it: Patients at risk of Type 2 diabetes who meet specific thresholds in weight and other measurements can join for free and keep track of their weight remotely. 

Why it matters: This is good and long-awaited news from companies like Omada Health, which brought diabetes prevention programs to the commercial market. Read Chief Privacy and Regulatory Officer Lucia Savage’s full statement here. 

The news: Realfood.gov features a chatbot that is supposed to answer questions about nutrition guidelines.

The tech: The bot is actually Grok, the chatbot part of X (formerly Twitter), which belongs to former government advisor Elon Musk.

The guidance: Journalists at STAT engaged with the bot and found that his answers were often confusing and at times contradicted the most recent guidelines. For instance, it doesn’t like the new pyramid much. 

Why it matters: AI is still an imperfect solution when it comes to nutrition recommendations and weight loss guidelines. It’s worth reading the full piece for a snapshot of what nutritionists think about these AI tools, including their concerns about chatbots and agents further stigmatizing obesity. 

The news: Novo Nordisk is going to start selling its GLP-1 product Wegovy in vials, following Lilly’s lead. The product is otherwise available in injector pens.

The price point: Novo hasn’t shared its pricing for the vial, but the move is done with the intent of offering a less expensive product. The company has also been slashing the prices of existing products in an attempt to make up for some of its market losses.

The other front: Novo is also suing Hims & Hers over its compounded version of Wegovy.

Why it matters: Hold the popcorn. We can barely keep up with the GLP-1 market, so Second Opinion is looking for a volunteer to write up an explainer of what is really going on, how companies are reacting, and where there are risks and opportunities. If that’s you, please do reach out. 

A reminder on upcoming webinars:

Hinge Health's shares were up 13% in after-hours trading on Tuesday after the digital health company reported a strong fourth quarter and a bright outlook for 2026.

Oscar expects to hit profitability in 2026 

CVS beat expectations on revenue and profit, even as the insurance sector is in turmoil 

The digital health M&A wave is here: AI and the need for scale are fueling the push for consolidation

AI will drive growth, direct-to-consumer will continue to thrive, and ROI will be crucial

What happened to Carbon Health? Stuart Miller breaks it down on Substack, showing why it points to structural failure beyond the individual company. 

A VC-backed AI and behavioral health darling is shuttering after a promising start. Endpoints reached out to former CEO Grace Chang for an interview.

$130 million for Solace: The patient advocacy platform closed a Series C round led by IVP. The funding brings the company to unicorn status, with a valuation of $1 billion.

Garner Health gets $118 million: The digital care navigators provide employers with a doctor ranking platform. It closed a Series D funding round led by Kleiner Perkins, which places the startup at a valuation of $1.35 billion. 

$100 million for Loyal: age1 led the Series C round to advance the first canine longevity drug, bringing the total investment in the company to $250 million. 

$19.2 million for Somethings: The digital mental health platform with the goal of expanding peer support for teenagers, closed a Series A round led by Catalio Capital.

$210 million for Talkiatry: The telepsychiatry provider closed a Series D round led by Perceptive Advisors. The company has raised $400 million so far. 

1) A close family member told me recently that they decided to seek out a different specialist for a procedure because they had such a challenging experience with the front desk. Why is that such a big problem in the outpatient space? 

CR: Provider practices are chronically understaffed for a few reasons. In many localities, people working at Starbucks can get paid more. The reason why that happens is often down to the marginal structures these practices have. There’s just not a lot of money to be spent at the front desk. We are also finding that hiring, recruiting, and retaining people who can speak multiple languages is difficult in many regions. It’s a niche skill. Then there are challenges related to managing these hires. When it comes to onboarding, training the staff, and so on, it’s outside of the comfort zone of many of the physicians starting these practices. Before starting the company, I shadowed hundreds of front desks at provider offices, and these challenges came up repeatedly. On average, we saw 30% attrition across all these practices.

2) So how do you see voice AI being useful here? There’s been a lot of talk about its potential when it comes to the front office. And what would ROI look like in practice?

CR:  I think of it as primarily being useful for patient access. Not everyone starts there, as many startups in the space will pitch practices on converting patients into revenue opportunities. We think patient access matters most. Imagine every time you call a specialty care provider, like your family member, that there are no appointment slots for weeks, or sometimes no one even picks up. In some of the groups we audited, that meant 10,000+ inbound calls a month, with thousands going unanswered and average hold times stretching beyond 10 minutes. There are huge downstream consequences related to that.

The second piece is capacity expansion. Most practices, within 4 to 6 weeks of going live, see measurable impact. With AI, we effectively add the bandwidth of two to three full-time employees per location, without increasing headcount. No one is getting replaced, but we expand labor capacity. In several deployments, that has meant 60%+ of calls handled by voice AI end-to-end.

Lastly, there is a revenue uptick for practices that invest in our voice AI - typically 10-15% improvement in payment collections - and we see that financial ROI within a few months.

3) Wearing my investor hat, I’d love to know what you think about creating a durable moat. When it comes to AI, what is going to provide companies with that true competitive edge?

CR:  The CEO of Anthropic has come out and said the models will be able to do most of the tasks that software engineers can do within months. So what models aren’t good at is the plumbing aspect of the end outcome that a practice wants to achieve within a workflow.

In healthcare, the real challenge is talking to EHRs for reading and writing conversations in real time, and tapping into fragmented phone systems without latency. It’s also a highly fragmented market with hundreds of EHR and PMS combinations. There are not enough standardized processes in the midmarket segment compared to large health systems that have centralized IT teams and infrastructure.

If you look at midmarket, especially the private equity-owned segment, there has historically been limited investment in SOPs and operational systems, which makes that plumbing layer significantly harder than people assume.

Bottom line, companies building in healthcare need to go deep into workflow integration. That includes provider preferences, insurance rules, patient journeys, outreach logic, and seamless EHR writeback, ideally with near real-time response speeds. Without that depth, the solution becomes surface-level.

How we landed on voice AI came from shadowing practices and seeing that the average front office staff member spends 3 to 4 hours per day on the phone. Solving that first made logical sense because it represents more than half their day.

For us, there is another potential moat, which is the data flywheel. Many peers are collecting data, but have different views on what to do with it. Some focus on memory, some on operational insights, some on broader intelligence layers. Our approach is simpler and more practical. If we know that Christina Farr cancelled her last appointment and is experiencing pain, can we proactively reach out so she comes back into the office? That closed-loop intelligence, built on real interaction data, becomes harder to replicate over time.

4) You brought in a head of partnerships relatively early in your trajectory. So let’s bring him in. Chetan and Raihan (Faroqui), why do you think partnerships are so important for companies like yours?

CR: We don’t call it partnerships. We call it embedded distribution. Because the term “partnerships” is overused, and it is often unclear what it really means. Companies assume that hiring someone in that role will magically turn them into a multimillion-dollar company overnight. That is rarely how it works.

For us, embedded distribution means aligning with organizations that already operate at scale in healthcare. Instead of selling one location at a time, we look for pathways where deployment can happen across multi-location groups. That is how we have been able to expand from single-site pilots to 100+ locations within months in some cases.

It is not about announcements. It is about repeatable expansion and measurable outcomes. When a solution can show 4 to 6 week time-to-impact and consistent operational lift across locations, those relationships naturally deepen.

RF: One example of embedded distribution for us is with EHR companies. They already own core workflows like scheduling, payment collections, and follow-ups. The question becomes, how do you leverage the EMR as the system of record and turn it into a true system of action? When we integrate deeply (e.g., both read and write functions), we are not replacing that infrastructure; we are activating it.

We are also engaging with private equity firms with specialty care roll-ups where we can scale across their large MSO, DSO, CIN, and group provider assets. These funds are strong collaborators because they are incentivized to implement tools like ours that deliver measurable EBITDA impact within months, not years. They are also looking to implement repeatable, scalable AI operating layers like ours across dozens of locations with consistent results. In some of our live deployments, we have seen 4 to 6 weeks' time-to-impact and meaningful operational lift across multi-location groups.

We host dinners, small private events, and rely heavily on warm introductions to operating partners and IT/Tech advisors. In healthcare, trust compounds in person. Those real-world relationships matter, and they often determine whether you get access to the operating layer of an organization or not.