I once watched a referral coordinator work through a stack of 200 faxes. She keyed demographics into the EHR by hand. She called the same practice three times for a chart note that should have come with the order. She was good at her job. The job was the problem.
Today a model can read that fax in seconds. Good. That should have existed 10 years ago, and the AI-native teams building it are talented and moving fast.
But automate the fax well and you can still lose the patient. The referral lives between two organizations, and they don’t see the same thing. The sender thinks it’s handled. The specialist is waiting on a record. The patient is waiting on a call nobody owns.
The fax was a symptom. The gap between organizations is the problem.
Features get copied
Two companies in our corner of the market have raised about $384 million since 2023 to automate referral paperwork, patient access, and intake. That money is aimed at a real problem, and it will produce good products.
It will also produce a lot of similar ones. Document reading, call handling, and intake parsing run on the same few models, and those models keep getting cheaper. In about two years, I expect most of this to come bundled with platforms you already pay for, or close to free.
If you’re buying, that’s good news. It also means “we read faxes with AI” won’t be a reason to pick a vendor for long.
The network is what’s left
Automation helps one organization work faster. Getting two organizations working from the same information takes a network.
We’ve been building ours since 2012. It’s now 30,000+ organizations, 60,000+ care settings, and 125,000+ providers, from health systems to single-physician practices. It grew one organization at a time, EHR by EHR, practice by practice. Plenty of administrators had been burned by vendors before us and weren’t looking to try another. They had good reason. We had to earn each connection. It wasn’t glamorous, and some of it was demoralizing. Honestly, a lot of it was fun.
What it looks like
Pioneer Physicians Network is an independent primary care network with 109 providers. Their staff stopped chasing faxes and can follow a referral from sent to closed.
The results: referral response time dropped from 108 hours to 35 hours. Days to schedule fell 29%. Care coordination time fell 75%.
Pioneer put it this way: “Having full transparency into our referral data provided the clarity we didn’t know we were missing.” Read the full Pioneer story.
On the Network, both sides of a referral see the same thing.
Take away the network and AI only sees one side
Put AI inside one practice and you get automation: useful, and eventually cheap. Put it on a connected network and it can see what one practice can’t. A coordinator skips a call because the answer is already on the other end. You can see where referrals are leaking, because both sides sit in the same system.
Take away the network and neither works.
Three questions to ask any vendor
Whose model is better will matter less each year. Ask these instead.
- How many of my referral partners are already connected? A tool is only as good as who’s on the other end.
- What happens when the practice across town runs a different system? If the answer is “they’d have to switch,” the gap is still there.
- If this feature is free in two years, what am I still paying for? Push for a specific answer. If it’s “more features,” keep asking.
Where this goes
I expect we’ll still be here, with a much larger Network than we have today. Several of the teams raising money right now are excellent, and I’d hire from them. We’ll be here because we spent 14 years building something a funding round can’t buy, and because the problem doesn’t go away when models get cheap.
In five years, I doubt anyone will pay extra for a tool that reads a fax. They’ll pay for knowing the practice across town is already on the other end. I have a drawer full of t-shirts from healthcare startups that no longer exist. Most had better slide decks than we did. I’d like ours to still be in rotation.
Talk to us about what a connected referral looks like.