If you’re not following Brendan Keeler, then you’re missing out. While I find most of what Brendan publishes quite interesting (Yes, I’m an #HITNerd), I was particularly impressed by his recent piece entitled “Make Networks Dumb Again.” If you’re into healthcare networks and wondering what the future of interoperability looks like you should go and read the whole piece. For those that are too lazy to read it or ask their favorite Gen AI bot to summarize it, I wanted to point out a few points he makes that really stand out to me along with some additional commentary.
One of the best points he makes is about the value of ubiquity when it comes to networks. Here’s an excerpt from the article:
Networks are judged first and foremost on ubiquity. Reaching one person with the coolest messaging app is a novelty, whereas a simple SMS can be relied upon to work. A technically elegant network that reaches 70% of its intended counterparties is decidedly less useful than an ugly, expensive one that reaches 99%.
This is why whenever anyone slams on faxes in healthcare, I always rail back at them. Fax isn’t an elegant solution. Fax has its weaknesses. However, it has the two things that every network wants: everyone can send and receive and we all know the standard. The ubiquity of Fax is what makes it so powerful. If there was another alternative that everyone adopted, we’d likely see the fax go away. Just look at the pager to see what I mean. Pagers are almost gone in healthcare because the smart phone came along and is essentially ubiquitous.
The problem here is that ubiquity is much harder to achieve than you think. Plus, creating a standard payload for the network that’s useful is a challenge as well. Or is it?
Does AI change things so that the payload can be as crude as you want and is it flexible enough to create an ubiquitous network?
Keeler in his article shared an example where this is happening with voice calls in healthcare. Talk about a crude network, but every organization has a phone and call centers answering those phones. Apply AI to the phone call as we’re seeing happen across healthcare, and you have a very crude standard (a phone call) and a nearly ubiquitous network. He described the challenge with this kind of network happening in systems that weren’t designed for this type of scale:
AI lets us reach that point extraordinarily quickly, because the incremental participant does not even need to be malicious. Now both ends of our analog networks are automating in an endless loop following their incentives:
- Consumers and businesses are using outbound agents to work through phone trees.
- Meanwhile, their counterparties are investing in reciprocal voice AI to handle the inbound volume.
- Both sides book ever expanding vendor contracts while volume explodes and the settlement rate on the underlying workflow stays flat.
The casualties are unfortunately us humans, the rapidly diminishing portion of the network volume. We queue behind tireless machines that never get frustrated and to whom another attempt is never not worth the effort. The defensive systems built to absorb and deflect all that traffic inevitably make the network harder for humans to use.
That’s right, the AI bots are going to overwhelm these phone systems, which were built for humans who are impatient and are happy to avoid making a call. AI bots are exhaustingly patient and happy to tie up phone lines for hours until they achieve their desired outcome.
This reminds me of one of the first companies I met who built an AI bot to call insurance companies. They had a massive need to get information from the payers for their provider clients and the payers didn’t want to work with them with any sort of API. The CEO finally said, “Screw it, we’re just going to build a bot to call the insurance company.”
They set the bot to work calling the insurance company to obtain the information their clients needed. After a little bit of time, the payer reached out to the company because the payer had seen such a high volume of calls coming to their call centers. The payer asked the company if they wanted to use an API to the payer instead of tying up the payer call center agents.
I think we can all appreciate the irony of this. Although, what’s interesting is what the CEO then told me next. He said that the payer wanted to give them free API access. He told them that he didn’t want the API access to be free. He wanted to pay the payer for the service. He figured if humans used to cost $5 per call to the payer and his AI bots calling cost 75 cents per call, then he could reasonably pay the payer 25 cents per API call (these are round numbers since I can’t remember his actual numbers, but they’re directionally accurate).
Why would this company want to pay for API access? The CEO told me that if he got API access for free, one day some executive at the payer would come along and say “Why are we paying so much to provide an API? That’s expensive, so let’s just stop doing it.” The CEO surmised that if the API was a profit center for the payer or at least covered its costs, it was more likely to stick around.
What’s fascinating is that we’re seeing this happen all over healthcare thanks to AI. Voice calls is one area. Bots leveraging the EHR user interface is another area. Screen scraping is about as crude an interface as you can get, but it’s ubiquitous which we’ve highlighted is a super power. Well, it’s a super power until it starts overwhelming the EHR which designed the UI for humans and not bots. We’ll see how that battle plays out technically and legally with things like information blocking.
Keeler also pointed out what we really want in our networks:
AI native networks can focus on the shared problems, in both healthcare and beyond! Verizon or USPS don’t attempt to know what you’re saying (hopefully). They (and all dumb networks) just do the middle boring stuff:
- Identity: who you are in a way the other end can verify
- Permissions: who you are acting for and what they authorized you to do
- Routing: how to find the counterparty you are trying to reach
- Trust: whether both ends have agreed to the same rules of the road
- Receipts: what happened provable after the fact
What do you think of this list for what a network needs to provide? Are we overcomplicating things like healthcare interoperability? Sure, we’d all love to have the perfect standard that everyone follows. However, that’s a pipe dream that Keeler appropriately points out will never be “finished.” Everyone connected with even a simple standard is far more valuable than 70% of people connected with an elegant standard as we shared at the beginning.
Keeler does finish with the assertion that the federal government needs to be the one to push this type of ubiquitous network in healthcare. I think he’s right. No one else in healthcare has the power and reach to really push this forward. In fact, I argued that this is what ONC should have done with the $36 billion of meaningful use money. Unfortunately, they took a much more prescriptive approach instead of incentivizing a network of healthcare organizations sharing data where there’s not a natural incentive to do so. That ship has passed, but they still seem like the only ones who could push widespread adoption.
