Ben Aylesworth.
Prepared for Microsoft AI Health, Strategy & Commercialization · November 2026

MAI‑DxO's first paying health systems should come from Dragon Copilot's existing customers.

Its four heaviest users carry their own financial risk, so a better diagnosis can pay them before any billing code exists.

Microsoft's four heaviest Dragon Copilot users already run thousands of clinicians on it, and each carries its own risk.

One square per 500 Dragon Copilot users, as each system reports them (users, providers, physicians or clinicians). Microsoft says ten systems passed 1M encounters this year but has named only these four.1

Also named as customers, at enterprise scale: Stanford, Northwestern, Michigan, Vanderbilt, Mount Sinai, Northeast Georgia. In pilots or departments: Atrium2, UNC, Brown, OCHIN, Tampa General, Ohio State, Community Health Network, WellSpan, Montefiore, Providence, Cooper, Overlake.

Payment for diagnostic AI is still at the proposal stage, so the first value has to land inside a risk contract.

  1. CMS's CY2027 rules propose a "Software as a Medical Service" category for algorithm-based decision support, priced largely by Medicare contractors.
  2. MAI-DxO's results published in Science: 83.2% on NEJM cases, against 36.1% for generalists. Retrospective, text-only; no named commercial customers yet.
  3. Dragon Copilot spread without a billing code. Diagnostic support can take the same route through a risk contract.

These four absorb the cost of a missed diagnosis, so they're the likeliest to see the return first.

  1. Intermountain Health

    Pilot where the plan and the hospital share one P&L

    A missed diagnosis costs Intermountain twice, as provider and as payer, so the ROI case is easiest to see here.

    First move. A bounded MAI-DxO pilot on complex or atypical cases in one service line, measured on Select Health members.

  2. Mass General Brigham

    Add diagnostic reasoning on top of Dragon Copilot

    MGB already runs Dragon Copilot at scale. A diagnostic layer in the same workflow gives the account a new reason to expand.

    First move. Bring MAI-DxO in through the existing Dragon Copilot account team as the next layer, not a separate sale.

  3. Mercy

    Show up where shared savings are already counted

    Mercy reports its ACO savings publicly ($103M in 2024) and plans to take Dragon Copilot to all 5,000 of its providers.

    First move. Tie the pilot's outcome to downstream workups inside the ACO population.

  4. Advocate Health

    Scale the pilot across the largest named user base

    A win here is the reference case for the six million-encounter systems Microsoft hasn't named.

    First move. Once one of the first three reads out, expand through Microsoft's existing enterprise sales team.

How I've tackled this before

At Bayer I moved 300+ hospital accounts onto a cloud clinical AI platform. At Aidoc I built the reimbursement and ROI models that COOs and CFOs signed off on.

Next step: a 90‑day MAI‑DxO pilot at Intermountain, measured in diagnostic workups against a baseline.

One service line where workups are frequent and expensive. Count workups for those case types in the quarter before go-live, compare the pilot quarter against it, and report to Select Health's finance lead at day 90. I can walk your team through the design in 20 minutes.

ben.aylesworth@gmail.com