A Mumbai hospital now generates 90% of its progress notes and nurse handovers with ambient AI — and the CIO's advice is that it only worked because they spent six months on governance before scaling a single ward
August 28, 2026 · 4 items
A Mumbai hospital now generates 90% of its progress notes and nurse handovers with ambient AI — and the CIO's advice is that it only worked because they spent six months on governance before scaling a single ward
Healthcare IT News · Adam Ang · August 28, 2026Practice operations
The numbers, all from the hospital's own deployment: Sir HN Reliance Foundation Hospital and Research Centre (Mumbai, Reliance Foundation) has run ambient AI at enterprise scale since January on a single platform integrated with its EMR. It now generates nearly 90% of progress notes and shift handovers, with five use cases live and more in development. Nurse handover time is down by nearly half. A multidisciplinary tumour board record — case parameters, competing opinions, final consensus — is produced in about 30 seconds, work that previously took hours of registrar time. The platform supports nine Indian languages and handles conversations that switch between English and a regional language mid-sentence.
The sequencing is the actual lesson, and it's the opposite of how this usually gets sold. CIO Dr Roheet Rao, verbatim: "Start with a platform thesis, not a demo." The hospital spent six months piloting and establishing governance, infrastructure, clinical safety and information-risk processes before scaling. Why that mattered: "We did establish a standard governance process to ensure that we can grow when required, plug and play, single location, multi-location" — and it can now stand the system up at a newly onboarded location "within a matter of days."
The safety design, spelled out item by item: a nurse or clinician reviews and approves every AI-generated note before it enters the EMR. Medications, allergies and lab values are checked against the patient's record rather than generated from the conversation alone. High-risk data are excluded outright. Reported result: zero safety incidents so far, and the hospital's latest quality audit found AI-generated notes to be of higher quality than the manual notes they replaced. Usage, note quality, defects, failures and template drift are monitored continuously by IT and quality teams before any further scaling.
The commercial detail nobody talks about: HN Reliance bought on a fixed-fee, not per-user model. Rao says that structure could make the cost of generating each AI note "almost five times cheaper than today" if projected FY2027 volumes are reached. His closing framing: "This was a deliberate path to being a fully ambient hospital for every profile within our system, for every conversation and every documentation that gets ambiently recorded and pushed into our analytics pipeline."
Why it matters for an independent practice: Yesterday's Ardent item was about adoption; this one is about architecture, and it's the more transferable of the two for a small practice. Three things. One, the per-user pricing trap is real and it's the single most actionable line here. Almost every ambient scribe and AI front-desk tool pitched to our clients is priced per seat, which means the cost of the tool rises exactly as the practice succeeds with it. A five-doctor practice should ask every vendor for the fixed-fee or volume-tiered version of the quote — not because it's always better, but because the answer tells you whether the vendor's economics are aligned with your growth. Two, "a nurse approves every note before it enters the record" is the sentence to put in a client's own AI policy. It's cheap, it's true, it's auditable, and it's the exact disclosure Wednesday's Pew data said 72% of patients want. A practice that can say it out loud has a differentiator its competitors have to build. Three, the honest scale caveat: this is a large private hospital with a dedicated CIO and a quality department. A solo practice can't run template-drift monitoring. But it can do the small version of the same thing — pick one workflow, run it for a month, check the output against reality, then add the second. The failure mode isn't going too slow; it's buying five tools at once and having no way to tell which one broke. Healthcare IT News — Private Indian hospital takes platform-first approach to ambient AI
Most physicians now use AI every day or every week — and more than 40% think they, not the practice, should keep the money it saves
Doximity 2026 Physician Compensation Report · report published Aug 28, 2026 · August 28, 2026Practice operations
The adoption baseline: more than 65% of physicians use AI on a daily or weekly basis for clinical or administrative work. Use is higher among non-surgical specialists, and younger physicians are both more likely to use AI and more likely to say it is reducing their workload.
The compensation question, which is the new part and the reason this is here: more than 40% of doctors think they should capture most of the cost savings if AI helps them work faster. Nearly half said physicians should be the primary beneficiaries if AI lets them complete more clinical work in the same time — against just 17% who said the savings should become lower costs for patients and 20% who said they should be shared. On whether pay should move at all: more than a third said compensation for a service should change if AI substantially cuts the time or effort it takes; 43% said it should stay the same. They are genuinely split.
The pressure is already showing up: a fifth of respondents say they face higher productivity expectations today because of AI, and 42% expect to. At the same time the vast majority don't see AI as a threat to job security — nearly a quarter expect AI to increase their compensation within the next year, and 39% say AI proficiency is already a factor in hiring decisions within their specialty.
The pay context, and one number that matters for recruiting: average physician pay rose 2% last year, down from 3.7% in 2024. Compensation rose in 41 of the 60 metro areas covered, more than half of those by at least 3%. Surgical specialties (neurosurgery, thoracic) top the table; pediatrics and primary care sit at the bottom.
:warning: The date caveat, stated up front because we just made a rule about this. The report is one day old. The underlying data is not new: Doximity drew on more than 250,000 compensation survey responses collected over seven years, including nearly 23,000 U.S. physicians surveyed last year alone. Healthcare Dive does not say when the AI-specific questions were fielded and I could not confirm it — so treat "65% use AI weekly" as a figure with a fielding date I can't verify, not a fresh August reading. Yesterday's trap was a new article about a seven-month-old survey; this is the same shape, just closer in.
Why it matters for an independent practice: This is the item to read as a business owner rather than a clinician. Three things. One, the recruiting number is the immediately usable one: 39% say AI proficiency factors into hiring in their specialty. For an independent practice competing against a health system for an associate, "here is the AI stack you'll work with, and here's who supports it" is now a recruiting line, not a tech line — the same signal Ardent reported yesterday when candidates started asking whether ambient documentation was available. Two, the split on compensation is a conversation to have before it becomes a grievance. If a practice deploys a scribe and volume goes up, someone will eventually ask who that gain belongs to. This survey says a plurality of doctors think the answer is "me," and a fifth already feel the productivity expectation without any of the upside. Deciding that question in advance and in writing costs nothing today and is very expensive later. Three, the marketing read for MMR: pediatrics and primary care sit at the bottom of the pay table and are exactly where efficiency tools get pitched hardest. That's the audience most exposed to a vendor promising throughput — and the one most in need of somebody asking what the practice actually keeps. Healthcare Dive — Physicians want compensation boost from AI productivity gains: survey
Google's August spam update made a top-10 page 1.8× more likely to disappear past position 100 — and healthcare was one of the steadier niches, not one of the wrecked ones
Search Engine Land · Danny Goodwin, Editorial Director · August 27, 2026Patient acquisition
The headline number: during Google's August 2026 spam update there was an 82% increase in top-10 URLs dropping out of the top 100 compared with a period with no confirmed ranking update. In hard terms: 16.71% of URLs that had ranked in the top 10 fell beyond position 100 for the same keyword, versus 9.2% during a comparable five-day stretch in July. That's a top-10 URL being about 1.8× more likely to vanish entirely.
It cut both ways, which is the part worth noticing: SE Ranking also recorded a 12% increase in the share of URLs ranking top 3 after the update that hadn't cracked the top 20 before it. Pages didn't just fall out — new pages arrived in the top three from outside the top twenty. Displacement, not just deletion.
No niche was spared, but the spread matters. Top-10 volatility ranged from 74.64% in real estate to 85.55% in fashion and beauty across the 20 industries tracked. SE Ranking notes that YMYL categories — including healthcare — were among the steadiest. Steadiest is not stable: the floor of that range is still ~75%.
The method, and what it can't tell you: SE Ranking analyzed U.S. organic results for the same 100,000 keywords across 20 industries, comparing Aug 17 vs Aug 22 (immediately before and after the update) against a July 26 vs July 31 baseline when Google had no confirmed update running. It tracked positions 1–100 only, so it cannot say whether the vanished URLs were deindexed or simply fell below 100 — and it did no URL- or domain-level analysis, so it cannot say what kind of site or tactic got hit. Conflicts to state: Search Engine Land is Semrush-owned, and SE Ranking is a rank-tracking vendor whose product exists to measure exactly this.
Why it matters for an independent practice: We have spent most of August on AI-search citations. This is a reminder that classic organic ranking is still where most of a practice's inbound patients come from, and that it moved hard nine days ago. Three things. One, do the boring check this week: pull rankings for every client's top 20 commercial terms and compare against the first week of August. The window that matters is Aug 17–22. If a client dropped, this is very likely why, and knowing that is worth more than a month of speculation — it also means you tell them before they ask. Two, the "healthcare was steady" line is a comfort you should not fully accept. YMYL being steadier is consistent with Google being more conservative in medical results, which historically favors established, credentialed sites. That helps a physician-authored practice site and hurts a thin one. It's an argument for the named-author, credentialed-content approach we already push — and a reason to check which of your clients' pages carry a real byline. Three, the displacement number is the opportunity. A 12% rise in pages jumping from outside the top 20 into the top 3 means the SERP for a client's core terms may have new competitors on it right now who weren't there three weeks ago. Look at who replaced whom before deciding what to write next. Search Engine Land — Google's August 2026 spam update hit rankings harder than normal
Salesforce made Claude the default reasoning model across its whole stack — and shipped a plugin that lets you run the CRM without ever opening the CRM
Salesforce Newsroom · press release, Aug 26, 2026, San Francisco · August 26, 2026Buildable AI
What actually shipped: "Claudeforce," an expanded Salesforce–Anthropic partnership. It launches with Salesforce in Claude — a plugin carrying 37 prebuilt sales skills (Salesforce names meeting prep, deal health review and pipeline review) that lets a user reason over live revenue data, update records and take action from inside Claude. Availability, per the release: select pilot customers now, open beta in September 2026, with additional skills from late 2026.
The line in the release that is the whole story: "For decades, enterprise software required users to manually navigate static UI to get work done. Now that agents can access the data, workflows, and rules directly, software is a system that powers every interface." Marc Benioff's version: "Here, the UI is the AI." Dario Amodei's: companies can "point Claude at the customer information and business context that they've been building in Salesforce for decades, and use it to actually run and grow their businesses."
The governance mechanics, which are the part that transfers: actions are routed through Salesforce so business rules are enforced when an action is taken — the reasoning happens in Claude, the permissions stay in the system of record. Setup is admin-connected once, with authentication and permissions managed centrally: "no per-user setup, no new permissions model to build, no re-auditing account by account." Claude is also delivered within the Salesforce Trust Boundary via Amazon Bedrock, which the release positions specifically for regulated industries.
How deep it goes, and one vendor number to treat as a vendor number: Claude now powers the Atlas Reasoning Engine, runs Agentforce Vibes and Agentforce Coworker by default, is available in Agent Builder, and is the default model for Slack — Slackbot, Claude Tag, and Slack Code. Salesforce's own internal claim: Slackbot drove 8.1M hours of annualized productivity gains, up over 2× quarter-over-quarter. That is a company reporting its own results in its own press release about its own partner; read it as marketing, not evidence. Conflict to state plainly: this is a vendor primary source, and both named quotes are the CEOs of the two companies selling it.
Why it matters for an independent practice: Zero medicine in the source — extrapolate. Nearly every practice we touch runs its patient pipeline on a CRM (GoHighLevel, HubSpot, an EHR's built-in module), and this is the clearest signal yet of where those tools are going: the login screen stops being the boundary. Three things. One, the diligence question changes shape. For years the question was "who has a login." When the CRM is reachable through a chat interface with an admin-level connection, the question becomes "what can the assistant do on behalf of someone who has a login, and where is that written down?" Salesforce's answer — actions route through the system so business rules are enforced — is the right answer, and it is the exact property to demand from any smaller vendor pitching the same convenience. Ask them to demonstrate it, not describe it. Two, "no re-auditing account by account" is a feature and a warning in the same sentence. Central setup is genuinely better than per-user chaos. It also means one admin connection can expose the whole record set at once. In a practice, that record set is patient-adjacent — names, contact details, why they called, what they were quoted. Whether that surface sits inside the BAA is a question to answer before someone turns it on, not after. Three, the opportunity, because it isn't only risk: "meeting prep, deal health review, pipeline review" translates almost directly to consult prep, lapsed-patient review, and referral-pipeline review. The enterprise version is being built in public right now; the practice-scale version is exactly the kind of thing we could assemble for a client six months before their competitors think to ask for it. Salesforce Newsroom — Salesforce and Anthropic Announce Claudeforce