A few months ago, I started therapy with a therapist I found on Headway. After adding my insurance information, the estimated copay was $5 per session. I’d been having difficulty finding an affordable therapist so this was a huge relief. I started seeing my therapist weekly, paying the $5 after every visit.

A few weeks later, I got this email:

My therapy sessions had a billing error after being processed by my insurance company. The updated costs were now between $130-$190 - a huge difference from the $5 per session I was paying initially.

You can imagine the double take I did when reading this email, and I want to tell you about it. But first let’s talk about how health tech companies work with insurance companies in the first place.

A Little Background Info

One of the most common financial models for health tech companies is to partner with insurance companies. The idea is simple - health insurance companies have a pool of patients that subscribe to them. Health tech companies can tap into this by building a partnership with the insurance company. They either get paid directly by the insurance company or the patient pays a copay or fee for the digital health product.

Sometimes patients find health tech companies on their own and use their insurance to access services. These providers take your insurance, like your PCP does. The difference is the cost can be hard to predict. Whereas you know the copay for your PCP visit will be X amount, some digital health companies provide estimated copay predictions.

You see this a lot with aggregator platforms for mental health services. There are companies like Headway, Alma, or Octave that onboard various providers who all accept different insurance. They try to predict the cost, but when they’re wrong it can be a nightmare for patients.

So what exactly happened with my case? And why did it happen and how can estimates be so off for a mental health visit?

What Happened

After getting the email that my therapy costs had gone up 30x, I was taken aback but not entirely surprised. After all, I used to work for a payer and know that tech company backends are prone to errors. So, I emailed customer support to find out what was going on.

They confirmed the initial message saying that now that the claims were actually processed with my insurance company, the original copay was an underestimate. The provider I had chosen was in a different tier than what they thought at first. Because of this, I would have to first hit my out-of-pocket max before my insurance would cover additional visits.

We went back and forth and I mentioned that I shouldn’t be held accountable to pay 30x the original cost and that there should be a margin of error or an upper limit on the estimates to prevent something like this.

I also threw in this gem:

Ok thank you. I still have two additional appointments that were on 2/1 and 2/8. I paid a $5 copay for those too. Is that included here? I just don’t want to be charged for those later if you guys have claims lag.

They ended up waiving the updated claims and honoring the $5 copay, but said they couldn’t waive claims that were still in process (update: they eventually waived those as well).

It was a good ending for me (though I had to stop seeing that therapist which was not very helpful). However, many people may not have reached out to customer support to question the situation and just paid the extra money.

And all because of inaccurate predictions and claims lag.

Why It Happened

Estimating out-of-pocket costs is not an easy task. I totally get this! It’s hard to predict cost sharing especially when dealing with multiple clinicians and insurance companies. The same clinician could be in a different provider tier for different insurance companies. Even within the same insurance company there are multiple plan skews where different providers are different tiers. Because of this, you can’t always predict what the copay for a given provider will be for any insurance plan.

Digital health companies likely model unknown costs by identifying known parameters and analyzing historical data to make an educated guess. In essence, this is a data science and modeling problem. Maybe you have a lot of patients with a specific insurance plan and can extrapolate that insurance type to others with the same insurance. Of course, there are many skews of plans within an insurance provider so you may not always be right but you should be close. In my case, the estimation algorithm broke. And it broke by a lot.

The other reason this happened is due to claims lag. I had weekly therapy sessions, paying a $5 copay each time. Those claims would be sent to my insurance company but would take a few weeks to process. By the time they were processed and Headway found out their estimate was wrong, I had already had 5 additional sessions. That put them in the position of trying to collect the difference from me weeks after. If they had known about the real cost sharing sooner, they could have intervened much sooner.

It’s hard to say how many people this impacts and what the outcomes are for each mishap. In my case, 5 sessions went from $5 to an average of $150. This means I would have to pay over $750 for care I had already received, when I thought it should cost me $25 total! Of course, the companies don’t release data on this but the downside for patients is that these large bills can be very costly and may lead people to discontinue care because they can’t afford it any longer.

How Do You Fix This?

I’m not writing about this experience to throw shade at Headway. They’re in a tough spot and doing the best they can with the information that they have. Healthcare data is messy and most of the time gated so finding ways to make proxy estimations is sometimes the best you can do as a health tech company. However, it’s a good thought exercise to think about how to improve on this customer experience and what we may actually be able to do to solve a tough data problem.

Fixing Claims Lag

This is a tough one for a health tech company to do. Mostly because they are a small player for a multitude of insurance companies. The insurers don’t have a huge financial incentive to help them, and they don’t control how quickly claims get processed and are on the receiving end of what happens at the insurance company level. Integration is the solution - but it requires significant growth and possibly an acquisition to get there. This is generally where pay-vider models work better but then you are giving up larger reach to multiple payers for straight integration with one payer.

Improving Estimation

In a world with claims lag, how can you improve your estimation? Using historical data for your patient population is key, but can be challenging for insurance products with small sample sizes (like my situation). In those cases, having ceiling prices for the customer is important. As a customer, I shouldn’t be held accountable for an estimation that was 30x off the actual cost. Having an upper ceiling is helpful and would probably make customers more likely to pay if they knew they were signing up for a specific limit. The flip side of this is being more conservative in your estimates. Companies likely already do this but there is always room for improvement, especially for edge cases.

The two other pieces to think about when it comes to estimation are better understanding insurance plan skews and validating estimations with other players in the market. Understanding plan skews is tough (and would likely need an ROI analysis to see if it is worth it). However, mapping this information out and understanding how plans vary within an insurance company could improve the estimation that you are providing your customers. The other point is to validate your estimates with those of other companies. This is a web scraping problem (also likely would need ROI analysis) but at least this gives you an idea if your price is way out left field compared to what others are doing in the market.

The goal with all of this isn’t necessarily to be 100% correct all the time. It’s to try and be less wrong and wrong with a lower variance (wow sorry that is such a data science sentence but read it again, it makes sense).

Some Parting Thoughts

All of these solutions are band-aids to deal with a system where data doesn’t flow as freely as it should. The ideal solution would be companies accessing provider databases from insurance companies to know upfront what the copays will be based on the tier of the provider. That information is gated and ultimately ends up hurting patients. In my case, I didn’t have to pay the additional money that they charged, but I did end up stopping therapy because I can’t afford the current cost.

Ultimately, the goal of having insurance and using these products is to make care easier, more accessible, and more affordable. Dealing with insurance complicates this, which is why many companies prefer the monthly fee route. But as patients, if we are already paying for insurance, digital health companies should provide cost estimates or limit our risk of high bills when we use their services.