A department chair is sitting in a conference room at 3 p.m. reviewing three ambient-AI scribe contracts, none of which she fully understands, with a decision due to the CFO by Friday.
She is excellent at medicine. She has run this department for six years. What she is being asked to evaluate now is a vendor contract with liability language she has never seen before, a data-use clause she cannot parse, and a rollout timeline that three different sales reps have described three different ways. Somewhere in the country, almost certainly, is a physician leader who deployed this exact vendor's product for a department of similar size eighteen months ago and can tell her, in ten minutes, what the sales deck leaves out.
She does not know that person's name. She has no way to search for them. What she has instead is a LinkedIn post from a physician she has never met, enthusiastically praising a different vendor, that may or may not be sponsored content. She has a consultant's proposal for $400 an hour to run an "independent" evaluation. She has a specialty listserv thread from eight months ago where someone asked a similar question and got two vague replies before the thread went quiet.
This is not a rare moment in her week. She has faced a version of this same blind evaluation for the EHR upgrade two years ago, for a CMS audit response last year, and for a physician staffing model change the year before that. Each time, the person who had already solved exactly this problem existed somewhere in American medicine. Each time, she never found them.
The knowledge that would save her the most time and the most risk this week is not clinical, and that is precisely why nothing in medicine is built to route it to her.
The volume of non-clinical decisions has exploded
This is not a niche complaint from a handful of physician executives. It describes a load that has grown sharply and fast.
Eighty-one percent of physicians now use AI in practice, up from 38 percent in 2023, according to the American Medical Association's 2026 Physician Survey on Augmented Intelligence. Ninety-two percent want more education on it. Physicians are not evaluating AI tools as a hypothetical anymore; the majority are already making live decisions about vendors, contracts and rollouts, often without training for any of it.
The same survey found 85 percent of physicians want to be consulted on AI adoption decisions. And yet the AMA's own data finds peer recommendations carry comparatively weak influence next to validated clinical evidence, a finding worth sitting with rather than accepting at face value: it likely reflects that physicians distrust unverified peer opinion on platforms like LinkedIn, not that verified peer experience would fail to help if it were actually reachable.
The pressure is not limited to AI. The shift toward physician employment, private-equity consolidation of practices, expanding federal interoperability and AI regulation, and the sheer pace of EHR and vendor churn mean the volume of non-clinical, operational and regulatory decisions any given physician leader faces has grown substantially, while the informal mechanism for getting help with them, essentially unchanged since the specialty listserv, has not kept pace at all.
What this actually costs when it goes wrong
The financial stakes are not abstract.
A single failed EHR or AI implementation can cost a mid-size health system millions of dollars in rework, lost productivity and clinician attrition. A physician-founder's regulatory misstep, a poorly scoped 510(k) submission or an FDA filing built on the wrong assumptions, can cost a company a full year of runway. These are not hypothetical worst cases; they are the ordinary, well-documented failure modes of operational and regulatory decisions made without access to someone who has already navigated the identical terrain.
Physician leaders currently pay consultants $300 to $600 an hour for advice that a peer who has already lived through the exact same decision could often give in twenty minutes, at no cost beyond the reciprocal favor of doing the same for someone else later. If even 5,000 physician leaders each avoided a single $10,000 consulting engagement per year through properly routed peer expertise, that is $50 million of value left entirely on the table today, because no mechanism exists to capture it.
And there is a related cost this series has documented elsewhere: physicians leave clinical practice at a mean age of 48.1, citing "hassle factor" at 44.7 percent and stress at 44.5 percent, according to a 2025 Permanente Journal study. Much of that hassle is precisely this category of operational and regulatory burden, and the knowledge that would reduce it is sitting, unroutable, in the heads of peers who already carried it.
Two kinds of expertise directories fail to index
Directories in medicine are built around one axis: clinical specialty. "Cardiology." "Orthopedic surgery." That axis is well developed, extensively verified through board certification, and, imperfectly, searchable.
There is no equivalent axis for what this article is about. Nothing indexes "led an ambient-AI rollout for 400 physicians." Nothing indexes "closed a 510(k) for a Class II device." Nothing indexes "negotiated a private-equity buyout of a 12-physician practice and would tell you honestly what the term sheet actually meant." This knowledge exists in real, specific people. It has simply never had a field to live in.
The distinction that makes this expertise especially hard to substitute for is the same one this series has found in adjacent problems: the most valuable version of it is candid failure, not curated success. "We tried that vendor and it was a disaster, here is exactly why" is worth far more to the physician evaluating the same vendor than any polished case study, and it is precisely the kind of statement nobody will post publicly, because it risks a vendor relationship, a reference call, or simple professional awkwardness.
Why existing channels do not solve this
LinkedIn is public and unverified. A post praising a vendor could be genuine enthusiasm, sponsored content, or a favor to a friend at the company, and there is no way for a reader to tell which. Public visibility is also precisely why the most useful information, candid failure, almost never appears there.
Doximity is clinical and ad-funded. It has no operational-exposure profile field, and its advertiser base includes many of the vendors physicians would be evaluating, which is not a conspiracy so much as a structural conflict that makes it an awkward home for candid vendor critique.
AMDIS and CHIME serve informatics executives specifically, a narrow slice of the physician population who face these decisions, and even within that population, they function as listservs and conferences rather than a routed, searchable, verified expertise graph.
Consultancies monetize the opacity directly. Their business model depends on physician leaders not having a free, trusted alternative; a $400-an-hour engagement is precisely the cost this article is describing as avoidable, and the consulting industry has no incentive to make it more avoidable.
Vendors curate their own references. Ask any vendor for a reference customer and you will be given their happiest customer, not a random or representative one. This is not dishonest, it is simply not independent, and physicians evaluating a vendor need independence more than they need enthusiasm.
Specialty societies run listservs, not routing. A post to a specialty listserv reaches whoever happens to be reading that week, with no mechanism to specifically surface the one person, out of tens of thousands of members, who has actually done the exact thing being asked about.
The structural failure: no credential, so no directory field
Follow the reasoning for why this gap persists rather than simply being an oversight someone will eventually fix.
Non-clinical expertise has no credential, and directories are built around credentials, because credentials are the verifiable signal that makes a listing trustworthy at scale. There is no board certification in "has deployed three different ambient-AI scribes and can compare them honestly." That means there is no natural field for this knowledge to occupy in any existing system built around licensure and board status.
It is acquired by doing, and it is time-bound. The 2023 ambient-scribe market bears little resemblance to the 2026 market. A directory entry from three years ago, even if one existed, would already be stale. This is expertise that decays quickly and needs to be current to be useful, which is a harder property to maintain than clinical specialty, which changes far more slowly.
Its most valuable form requires confidentiality that public platforms cannot provide. A candid account of a failed vendor rollout is professionally risky to share in public, both because it may sour a vendor relationship and because it can read, out of context, as an admission of poor judgment. No open, ad-funded, permanently indexed platform can host this kind of disclosure safely, which is why it stays in private conversations between people who already trust each other, rather than anywhere searchable.
The AMA's own finding, that physicians rate peer recommendations as comparatively weak, is consistent with this: it is not that peer knowledge lacks value, it is that the peer knowledge physicians currently encounter is unverified, unrouted, and largely public, which are exactly the conditions under which candid, high-value information does not get shared in the first place.
What would actually work
Verified operational exposure, attested alongside clinical credentials. A member states specifically what she has done: which vendor, what scale, what year, what outcome, verified the same way clinical board certification is verified, so a searcher knows the exposure is real rather than self-promotional.
A confidential channel, not a public posting. Candid failure needs a space that cannot be indexed publicly and cannot be quoted back to a vendor, so the person sharing it can speak honestly about what actually went wrong.
Routing on the specific exposure, not the general specialty. The question is not "who is a CMIO" but "who specifically deployed this vendor, at this scale, in the last two years," and the routing mechanism needs to match on that narrow, current attribute.
Reciprocity as the incentive, not payment for every interaction. A senior or retired physician leader answering a twenty-minute question is far more sustainable as a norm of mutual obligation within a trusted network than as a per-question paid transaction, which introduces exactly the incentive problems that make vendor references untrustworthy.
Firewalled from vendor influence entirely. Any version of this that accepts vendor sponsorship for placement or visibility recreates the exact conflict that makes existing reference programs untrustworthy; the value of the mechanism depends entirely on its independence.
A living record that decays and gets refreshed, not a permanent directory entry. Given how fast the underlying technology and vendor landscape changes, exposure attestations need built-in staleness signals, so a three-year-old entry is visibly less current than a six-month-old one.
Time-bound engagement, matched to how physicians actually have time to give. A structured twenty-minute call or written debrief is a request most busy physician leaders can say yes to; an open-ended standing consulting relationship is not, and conflating the two discourages participation.
What you can do now
If you are a physician leader facing a non-clinical decision
Ask your own network specifically, not generally. Instead of "does anyone have thoughts on AI scribes," ask "has anyone deployed [specific vendor] for a department over 200 physicians in the last 18 months." Specificity produces useful answers where generality produces silence.
Treat vendor references as a floor, not a ceiling. Ask the vendor for a reference, take the call, and then separately try to find someone outside the vendor's selection, through your own network or a specialty listserv, who has used the product and was not handed to you by the company selling it.
Write down what you learn from every implementation, even the ones that go well, in a form you would be willing to share with a trusted peer later. This is the same discipline this series recommends for clinical scar tissue, applied to operational and regulatory experience, and it is the raw material any future routing mechanism will need from people like you.
If you lead a department, health system or physician group
Name the real cost of unroved evaluation out loud. A $300-to-$600-an-hour consulting engagement for advice a peer could give in twenty minutes is a quantifiable, recurring cost your organization is already paying; naming it changes how seriously your leadership takes building or joining a better channel.
Build a habit of debriefing implementations candidly within your own leadership team, then extend it outward. Most organizations do not even capture this knowledge internally before a leader who holds it departs; internal capture is a prerequisite for any external sharing later.
If you build systems, run a specialty society, or lead a physician network
Do not build another directory of clinical specialties. That axis is already well served. Build the missing axis: verified, current, specific operational and regulatory exposure, with confidentiality protections strong enough that candid failure stories are actually shared rather than sanitized into uselessness.
Firewall vendor money from routing and visibility from day one. The single fastest way to destroy trust in a peer-reference mechanism is to let a vendor pay for placement within it, even indirectly; building this constraint in early is far easier than retrofitting it after credibility has already been damaged.
Frequently asked questions
How should physicians evaluate an AI scribe or other clinical AI tool? Most physicians currently rely on vendor demonstrations, vendor-selected references, and informal peer opinion from unverified sources like LinkedIn. The American Medical Association's 2026 Physician Survey on Augmented Intelligence found 81 percent of physicians now use AI in practice and 92 percent want more education, indicating current evaluation resources are widely seen as inadequate.
What percentage of physicians use AI in their practice? Eighty-one percent, according to the AMA's 2026 Physician Survey on Augmented Intelligence, up sharply from 38 percent in 2023. The same survey found 85 percent of physicians want to be consulted on AI adoption decisions in their own organizations.
Where do physician leaders get peer advice confidentially? There is no dedicated, verified, confidential channel for this today. Physicians typically rely on informal contacts, specialty listservs such as those run by AMDIS or CHIME, or paid consultants charging $300 to $600 an hour, none of which reliably routes a specific question to the specific peer who has already solved it.
Who can a physician ask about starting a medical device company or filing a 510(k)? Currently, mostly whoever they happen to know personally, since no verified, searchable network indexes physicians by specific regulatory or entrepreneurial exposure such as having completed an FDA 510(k) submission. This gap is one reason physician-founded device and digital-health companies frequently repeat avoidable regulatory missteps that a peer who already navigated the process could have flagged in advance.
Why don't LinkedIn or Doximity solve this problem already? LinkedIn is public and unverified, which discourages the candid, sometimes unflattering accounts of vendor failures that are most valuable, and does not verify claimed expertise. Doximity is built around clinical identity and is ad-funded, with advertisers that include many of the vendors physicians would be evaluating, creating a structural conflict that makes it a poor home for independent peer critique.
What does a failed health-system AI or EHR implementation actually cost? Published figures vary by scale and vendor, but a single failed rollout at a mid-size health system can run into the millions of dollars in rework, lost productivity, and clinician attrition; the underlying driver in many cases is a decision made without access to a peer who had already implemented the same tool elsewhere and could flag the risks in advance.
The bottom line
A physician leader facing a vendor contract, an FDA filing, a CMS audit, or a practice sale is, in nearly every case, facing a decision someone else in American medicine has already made. That person's hard-won knowledge, including exactly the candid failure detail that would matter most, exists somewhere, right now, reachable by nobody who needs it.
Eighty-one percent of physicians already use AI in practice, and the volume of decisions like this is only rising as employment consolidation, private equity, and regulatory complexity all compound at once. Directories index clinical specialty well. They index none of this, because none of this fits the credential-based model directories were built around, and because its most valuable form, candid disclosure of what went wrong, cannot survive on a public, ad-funded, permanently indexed platform.
Consultants charge $300 to $600 an hour to fill the gap. Vendors offer curated references that answer a different question than the one physicians are actually asking. LinkedIn offers volume without verification. None of it is what a department chair actually needs at 3 p.m. on a Wednesday with a decision due Friday.
The chair in the conference room is not lacking clinical judgment. She has plenty. The knowledge that would save her the most time and the most risk this week is not clinical, and that is precisely why nothing in medicine is built to route it to her.
Part of a series on the missing professional infrastructure of healthcare. Previously: The M&M Room Has No Door
Evidence note: sources include the American Medical Association's 2026 Physician Survey on Augmented Intelligence; a 2025 Permanente Journal study on physicians leaving clinical practice at a mean age of 48.1, citing hassle factor and stress as leading reasons; and a 2009 Vanson Bourne survey of 170 UK organizations finding 68 percent wanted to search colleague expertise while only 39 percent could, a dated but sector-relevant figure with no more recent US equivalent identified. Dollar figures on consulting costs and the value of avoided engagements ($300 to $600 per hour, a hypothetical $50 million in aggregate avoided spend across 5,000 physician leaders) are order-of-magnitude estimates built from typical published consulting rates and stated assumptions, not a measured market total, and should be read accordingly. Claims about a specific number of board-certified clinical informaticists could not be independently verified in this research pass and are omitted rather than estimated.
