ProgrammableMarketing/withprotege.ai (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 6
  • This quarter · 10
  • Big bets · 3
  • Wildcards · 3
  • How we can help
Generated Oct 23, 2026
  1. Programmable
  2. Playbooks
  3. Protege
Protege logo

growth playbook · Oct 23, 2026

withprotege.ai (opens in a new tab)

Protege growth playbook

Diagnosis

Protege has raised $65M, built a network of data partners across healthcare and media, and launched Independent Evaluations for healthcare AI on October 21, 2026.

Its hiring shows the next steps: a commercial lead for assessments, a product and content marketer, a Head of Research for DataLab, and partnership managers for revenue cycle and healthcare administration data.

The site does not yet turn evaluations into something buyers can check, sends evaluation requests to a general contact form, and leaves an expired webinar invitation and confusing dates on key pages. Start with a public evaluation registry with attestation pages, an evaluation intake and targeted outreach; then build buyer-side demand, a living Disease Odyssey leaderboard, structured data-holder intake and vertical sales kits.

Growth Score

8/10

Biggest Opportunity

Becoming the checkable, independent standard for healthcare AI evaluation, with a public registry that every attested builder points buyers to.

Biggest Risk

Trust is the product: any doubt about data rights, privacy or independence would hurt both the data and evaluation businesses, while Scale, Vals AI, Truveta and Datavant compete for the same customers and partners.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    A public registry of Protege evaluations, with a page per attestation that buyers can check

    Draft the registry page with the three published evaluations.

    High impact First result in 1 month
  2. 2

    Turn "See how your model scores" into a short evaluation intake, routed to the Assessments lead

    Write the five questions with the Solutions Lead.

    High impact First result in 2 weeks
  3. 3

    Replace the past webinar banner on the data provider page with the replay

    Swap the banner today.

    Medium impact First result in 1 week

02

Your idea portfolio

22 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 7 ideas are marked Unexpected: plays you are unlikely to have tried.

    1 of these are things Programmable can build and run for you.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 3Big bets · 11Fill-ins · 4Reconsider · 4
Effort →

Showing 22 of 22 ideas

This week 6

Quick wins with a clear first action.

  • 01

    Protege has just launched Independent Evaluations and is hiring a commercial lead to sign its first assessments customers from 15 to 20 vertical healthcare AI builders. The evals page promises them "score, report, attestation" and "a credible go-to-market signal" to "win health systems and payers". But there is nowhere for a health system to check that signal: the three published results are separate blog posts, and there is no list of evaluations and no page that confirms a given attestation. The launch post compares Protege to JD Power, FICO and Gartner, and those are valuable because anyone can look a result up. Vals AI already lets visitors "Explore benchmark results" on its homepage.

    High impactMedium effort First result in 1 month GTM Commercial Lead

    First action

    Draft the registry page with the three published evaluations.

    How to measure it

    Assessments deals closed

    Now
    –
    Target
    5
    By
    2 quarters

    Steps

    1. 1

      Agree the fields

      What each evaluation card and attestation page must show.

      Head of Research, DataLab 2 days

    2. 2

      Legal wording

      What the attestation does and does not claim.

      Sr. Counsel, Commercial and Data 1 week

    3. 3

      Build /evaluations

      Cards for the three published results.

      Web developer 1 week

    4. 4

      Attestation pages

      Permanent URL, ID, test conditions, expiry.

      Web developer 1 week

    5. 5

      Badge

      Linked badge and usage rules for builders.

      Technical Product & Content Marketing Lead 1 week

    6. 6

      Buyer guide

      One page on reading an evaluation, for health systems and payers.

      Technical Product & Content Marketing Lead 1 week

    7. 7

      Sell with it

      Registry link in every outreach to the 15-20 target builders.

      GTM Commercial Lead, Assessments ongoing

    8. 8

      Measure

      Registry visits, badge clicks and deals that cite it.

      Technical Product & Content Marketing Lead monthly

    First result
    1 month
    Cost
    Staff time only
    Owner
    GTM Commercial Lead, Assessments

    Smallest test

    Publish the registry with the three existing results and send it to five target builders.

    Why it works

    Publish /evaluations: one card per published evaluation (model, builder, workflow, data scope, date, headline result, link to the full report) and a permanent page per attestation with an ID, the test conditions (double-blind, held-out, at most two runs) and the date it expires or needs a re-run. Give each customer a badge that links to their attestation page, and a one-page "how to read a Protege evaluation" for buyers. Private evaluations stay private; the registry only lists what builders choose to publish.

    A builder buys an evaluation to win health systems. A public, checkable page is what makes it worth showing to them, and every badge shown in a sales deck sends a buyer back to Protege. It also gives the new Assessments lead something concrete to sell in the first 60 days.

    Ready to use

    ChecklistEvaluation registry launch

    [ ] /evaluations page with one card per published evaluation
    [ ] Attestation page: ID, model and version, workflow, data scope, conditions, date, expiry
    [ ] Linked badge with usage rules
    [ ] "How to read a Protege evaluation" for health systems and payers
    [ ] Legal review of attestation wording
    [ ] Link from the Healthcare Evals page and launch post
    [ ] Registry link in Assessments outreach

    Risks and how to handle them

    • Builders want results kept private

      List only results they opt to publish; private reports stay off the registry.

    • Badges outlive model updates

      Show the model version and an expiry or re-run date on each page.

    Evidence

    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • Your websiteLaunch post "Announcing Independent Evaluations by Protege" (published October 21, 2026): "91% of healthcare organizations cannot verify AI governance"; compares the role to "JD Power", "the FICO score" and "Gartner"; Protege's partners compile "multi-modal data covering 300 million Americans"; HealthSAW "outperformed leading frontier models on 14 of the 16 disease-risk prediction workflows"; on the Disease Odyssey benchmark "Today's leading models get the diagnosis right only 51-67% of the time" for conditions such as sleep apnea and anemia; and "evaluations need to be refreshed as often as the models they are testing".Source
    • HiringGTM Commercial Lead, Assessments (remote, posted September 12, 2026): owns "the commercial strategy and execution for our benchmarking and assessments business"; in 30 days "Build a target account map of 15 to 20 vertical healthcare AI builders"; in 60 days "Close first tranche of assessments customers"; reports to the head of GTM Partnerships.Source
    • Competitor researchRivals' own homepages: Vals AI "independently tests AI models on real-world work. Explore benchmark results, model comparisons"; Scale delivers "data, evaluations, and outcomes to AI labs"; Defined.ai is an "AI Training Data Platform for Enterprise AI"; Truveta offers "real-time data from 130M+ patients across US health systems"; Datavant runs a network of "350+ real-world data partners".Source
    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source
  • 02

    Both "SEE HOW YOUR MODEL SCORES" buttons on the evals page go to a general contact form. The Assessments lead needs to qualify builders by fit, and the Solutions team needs the workflow and data needs before scoping.

    High impactLow effort First result in 2 weeks GTM Commercial Lead

    First action

    Write the five questions with the Solutions Lead.

    How to measure it

    Qualified requests

    Now
    –
    Target
    +50%
    By
    60 days

    Steps

    1. 1

      Questions

      Five fields.

      GTM Commercial Lead, Assessments 1 day

    2. 2

      Scope check

      What Solutions needs to start.

      Solutions Lead, Healthcare 1 day

    3. 3

      Build

      Form on the evals page.

      Web developer 2 days

    4. 4

      Route

      To the Assessments lead with booking link.

      GTM Commercial Lead, Assessments 1 day

    5. 5

      Source tags

      Launch post, webinar, outreach.

      Technical Product & Content Marketing Lead 1 day

    6. 6

      Review

      Request quality weekly.

      GTM Commercial Lead, Assessments weekly

    First result
    2 weeks
    Cost
    $0
    Owner
    GTM Commercial Lead, Assessments

    Smallest test

    Swap the form on the evals page this week.

    Why it works

    Replace the contact form with five questions: model type and access (API, hosted, open weights), clinical or admin workflow, target buyer (health systems, payers, life sciences), timeline, and whether results would be published. Route answers to the Assessments lead with a booking link, and tag the lead source.

    Qualified requests shorten scoping and let the team focus on builders who fit.

    Risks and how to handle them

    • Fewer form fills

      Keep three fields required, two optional.

    Evidence

    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • HiringGTM Commercial Lead, Assessments (remote, posted September 12, 2026): owns "the commercial strategy and execution for our benchmarking and assessments business"; in 30 days "Build a target account map of 15 to 20 vertical healthcare AI builders"; in 60 days "Close first tranche of assessments customers"; reports to the head of GTM Partnerships.Source
    • HiringOther open roles: Solutions Lead, Healthcare (cohorts across "EHR, claims, radiology, pathology, and unstructured text", with "HIPAA-aware handling"), Strategic Account Manager, Applied Healthcare Researcher, Forward Deployed Engineer, Audio, and Sr. Counsel, Commercial and Data. Every post describes "the secure, efficient, and privacy-centric exchange of AI training data".Source
  • 03

    The healthcare page for data providers still invites visitors to a live webinar on October 13 with "Register Here", while the model builders' page already links the replay. A data holder arriving now sees an expired invitation.

    Medium impactLow effort First result in 1 week Technical Product & Content M…

    First action

    Swap the banner today.

    How to measure it

    Replay views

    Now
    –
    Target
    200
    By
    30 days

    Steps

    1. 1

      Swap banner

      Replay and summary.

      Technical Product & Content Marketing Lead 1 hour

    2. 2

      Add CTA

      Become a data partner.

      Technical Product & Content Marketing Lead 1 hour

    3. 3

      Check pages

      Other expired banners.

      Web developer 2 hours

    4. 4

      End dates

      Auto-switch for events.

      Web developer 1 day

    5. 5

      Clip

      Short cuts for social.

      Technical Product & Content Marketing Lead 1 week

    6. 6

      Measure

      Views and partner inquiries.

      Technical Product & Content Marketing Lead monthly

    First result
    1 week
    Cost
    $0
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Change the banner today.

    Why it works

    Swap the banner for the replay with a short summary and a "become a data partner" link, and set an end date on future event banners so they switch to the replay automatically.

    Data holders judge partners on care and detail, and the replay is useful content for them too.

    Risks and how to handle them

    • None significant

      Routine content fix.

    Evidence

    • Your websiteHealthcare page for data providers (/data-provider/healthcare, checked October 22, 2026) still shows "Live Webinar on Oct 13, 12:30pm EST: Learn how Protege better evaluates healthcare AI with real-world data. Register Here", while the model builders' healthcare page links the replay. The page lists "5B+" total patient encounters and "700K+" average tokens per patient, and a four-step process ending in "Delivery & Revenue Recognition", where Protege pays "revenue share to the participating partner(s)".Source
    • HiringTechnical Product & Content Marketing Lead (remote, posted August 27, 2026): "Translate Protege's Product efforts across verticals into clear public-facing positioning"; turn "DataLab's technical research into buyer language"; "Own end-to-end launch campaigns for new Product, DataLab, and Commercial releases"; build "decks, webinars, videos, social media content, one-pagers, case studies, talk tracks, landing pages, and launch materials"; partner with the Head of Community; and use "the right growth channels to distribute your work".Source
  • 04

    Protege dates articles as numbers in US order: the January 8, 2026 funding release shows "01 07 2026", which readers in Europe read as 1 July. Protege partners with broadcasters and data holders in EMEA, Asia and Africa, and the footer still says © 2025.

    Medium impactLow effort First result in 1 week Technical Product & Content M…

    First action

    Change the date format in the theme.

    How to measure it

    Pages fixed

    Now
    –
    Target
    100%
    By
    1 week

    Steps

    1. 1

      Theme

      Date format setting.

      Web developer 1 hour

    2. 2

      Listings

      Article cards.

      Web developer 1 hour

    3. 3

      Footer

      Automatic year.

      Web developer 30 minutes

    4. 4

      DataLab site

      Same format.

      Web developer 1 hour

    5. 5

      Check

      Structured data dates.

      Web developer 1 hour

    6. 6

      Review

      Spot-check 10 pages.

      Technical Product & Content Marketing Lead 1 hour

    First result
    1 week
    Cost
    $0
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Fix the format in the theme today.

    Why it works

    Show dates as "8 January 2026" or "Jan 8, 2026" across articles and listings, and make the footer year automatic.

    Clear dates make news look current to international partners and journalists.

    Risks and how to handle them

    • None significant

      Theme setting only.

    Evidence

    • Your websiteHomepage: "Real World Data for AI Development" for pre-training, post-training and "Evaluation & Benchmarks"; separate paths for model builders and data providers; a Siemens Healthineers quote; article dates are printed as numbers in month-day-year order (the January 8, 2026 funding release shows "01 07 2026", an October 2025 case study "10 08 2025"); the footer reads "© 2025 Protege Health, Inc." The site runs on WordPress.Source
    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source
  • 05

    Protege's pages give several scale figures: "~40M patients" in the evaluation sandbox, partners covering "300 million Americans", "5B+" encounters and "hundreds of millions of patients". Each may be right for what it counts, but buyers and journalists will compare them.

    High impactLow effort First result in 1 month Technical Product & Content M…

    First action

    List every scale number on the site.

    How to measure it

    Pages aligned

    Now
    –
    Target
    100%
    By
    30 days

    Steps

    1. 1

      Collect

      All figures.

      Technical Product & Content Marketing Lead 1 day

    2. 2

      Define

      What each counts.

      Head of Research, DataLab 2 days

    3. 3

      Legal check

      Wording on de-identified data.

      Sr. Counsel, Commercial and Data 2 days

    4. 4

      Publish

      One block reused.

      Web developer 2 days

    5. 5

      Decks

      Same block.

      Technical Product & Content Marketing Lead 1 day

    6. 6

      Review

      Quarterly.

      Technical Product & Content Marketing Lead quarterly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Write the definitions block.

    Why it works

    Write one short "our data network in numbers" block that defines each figure (de-identified records available for evaluation, partner network coverage, encounters) with a date, and use the same wording on every page and in decks.

    Clear definitions prevent doubt in diligence and help the marketing lead keep one story.

    Risks and how to handle them

    • Numbers change

      Date each figure and review quarterly.

    Evidence

    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • Your websiteLaunch post "Announcing Independent Evaluations by Protege" (published October 21, 2026): "91% of healthcare organizations cannot verify AI governance"; compares the role to "JD Power", "the FICO score" and "Gartner"; Protege's partners compile "multi-modal data covering 300 million Americans"; HealthSAW "outperformed leading frontier models on 14 of the 16 disease-risk prediction workflows"; on the Disease Odyssey benchmark "Today's leading models get the diagnosis right only 51-67% of the time" for conditions such as sleep apnea and anemia; and "evaluations need to be refreshed as often as the models they are testing".Source
    • Your websiteHealthcare page for data providers (/data-provider/healthcare, checked October 22, 2026) still shows "Live Webinar on Oct 13, 12:30pm EST: Learn how Protege better evaluates healthcare AI with real-world data. Register Here", while the model builders' healthcare page links the replay. The page lists "5B+" total patient encounters and "700K+" average tokens per patient, and a four-step process ending in "Delivery & Revenue Recognition", where Protege pays "revenue share to the participating partner(s)".Source
  • 06

    The Assessments lead must map 15 to 20 vertical healthcare AI builders and start conversations. The launch shows leading models get the diagnosis right only 51-67% of the time on Disease Odyssey, which is a reason for any builder in those areas to want its own test.

    High impactLow effort First result in 1 month GTM Commercial Lead

    First action

    Map the first ten target builders to the closest published result.

    How to measure it

    Scoping calls

    Now
    –
    Target
    10
    By
    30 days

    Steps

    1. 1

      Map

      Builders to results.

      GTM Commercial Lead, Assessments 2 days

    2. 2

      Write notes

      One per builder.

      Technical Product & Content Marketing Lead 3 days

    3. 3

      DataLab slots

      Research partner on calls.

      Head of Research, DataLab 1 week

    4. 4

      Send

      From the Assessments lead.

      GTM Commercial Lead, Assessments 1 week

    5. 5

      Follow up

      Twice.

      GTM Commercial Lead, Assessments 2 weeks

    6. 6

      Track

      Calls and proposals.

      GTM Commercial Lead, Assessments weekly

    First result
    1 month
    Cost
    Staff time only
    Owner
    GTM Commercial Lead, Assessments

    Smallest test

    Send five notes this week.

    Why it works

    For each target account, write a short note that links the result most relevant to their area, explains what a double-blind evaluation of their model would show, and offers a 30-minute scoping call with DataLab.

    A specific result is a better opener than a generic launch email, and DataLab in the room adds credibility.

    Risks and how to handle them

    • Builders read results as criticism

      Frame as an opportunity to prove their model.

    Evidence

    • HiringGTM Commercial Lead, Assessments (remote, posted September 12, 2026): owns "the commercial strategy and execution for our benchmarking and assessments business"; in 30 days "Build a target account map of 15 to 20 vertical healthcare AI builders"; in 60 days "Close first tranche of assessments customers"; reports to the head of GTM Partnerships.Source
    • Your websiteLaunch post "Announcing Independent Evaluations by Protege" (published October 21, 2026): "91% of healthcare organizations cannot verify AI governance"; compares the role to "JD Power", "the FICO score" and "Gartner"; Protege's partners compile "multi-modal data covering 300 million Americans"; HealthSAW "outperformed leading frontier models on 14 of the 16 disease-risk prediction workflows"; on the Disease Odyssey benchmark "Today's leading models get the diagnosis right only 51-67% of the time" for conditions such as sleep apnea and anemia; and "evaluations need to be refreshed as often as the models they are testing".Source
    • HiringHead of Research, DataLab (remote, posted October 21, 2026): DataLab publishes "on the questions that matter: what agentic AI should actually be trained to do, how to quality-control large-scale corpora, and how to build evaluation datasets that reflect the real world rather than the leaderboard"; the head will "Serve as a customer-facing research partner to GTM" and "Represent Protege externally".Source
    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source

This quarter 10

The core plays for the next three months.

  • 07

    The evals page is written for builders, but the value of an attestation depends on buyers asking for it. The launch post says 91% of healthcare organizations cannot verify AI governance, and the sitemap has no page for health systems or payers.

    High impactMedium effort First result in 1 quarter Technical Product & Content M…

    First action

    Draft the sample RFP clause.

    How to measure it

    Buyer-driven requests

    Now
    –
    Target
    5 a quarter
    By
    2 quarters

    Steps

    1. 1

      Interview

      Three health system AI leaders.

      GTM Commercial Lead, Assessments 1 month

    2. 2

      Write page

      Buyer guide and RFP clause.

      Technical Product & Content Marketing Lead 2 weeks

    3. 3

      Legal

      Clause wording.

      Sr. Counsel, Commercial and Data 1 week

    4. 4

      Publish

      Linked from registry.

      Web developer 1 week

    5. 5

      Webinar

      For health system AI leaders.

      Technical Product & Content Marketing Lead 1 month

    6. 6

      Track

      Requests citing buyers.

      GTM Commercial Lead, Assessments monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Publish the RFP clause as a short post.

    Why it works

    Publish a buyer page: what a double-blind evaluation tests, questions to ask vendors, how to read an attestation, and a sample RFP clause asking for independent evaluation. Promote it with a webinar for health system AI leaders.

    When buyers ask vendors for a Protege evaluation, demand comes to Protege through the builders.

    Risks and how to handle them

    • Seen as self-serving

      Recommend independent evaluation generally, with Protege as one option.

    Evidence

    • Your websiteLaunch post "Announcing Independent Evaluations by Protege" (published October 21, 2026): "91% of healthcare organizations cannot verify AI governance"; compares the role to "JD Power", "the FICO score" and "Gartner"; Protege's partners compile "multi-modal data covering 300 million Americans"; HealthSAW "outperformed leading frontier models on 14 of the 16 disease-risk prediction workflows"; on the Disease Odyssey benchmark "Today's leading models get the diagnosis right only 51-67% of the time" for conditions such as sleep apnea and anemia; and "evaluations need to be refreshed as often as the models they are testing".Source
    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source
  • 08

    Disease Odyssey tests whether models spot a serious illness in the record before diagnosis, and the launch says evaluations must be "refreshed as often as the models they are testing". Vals AI publishes model comparisons on its homepage.

    High impactMedium effort First result in 1 quarter Head of Research

    First action

    Publish the current results as a table.

    How to measure it

    Model submissions

    Now
    –
    Target
    10
    By
    2 quarters

    Steps

    1. 1

      Table

      Current results.

      Applied Healthcare Researcher 1 week

    2. 2

      Method notes

      How it is run.

      Head of Research, DataLab 1 week

    3. 3

      Page

      Leaderboard on the site.

      Web developer 2 weeks

    4. 4

      Re-run plan

      On major releases.

      Head of Research, DataLab ongoing

    5. 5

      Submissions

      Builder form.

      GTM Commercial Lead, Assessments 2 weeks

    6. 6

      Promote

      Each update.

      Technical Product & Content Marketing Lead ongoing

    First result
    1 quarter
    Cost
    Compute and staff time
    Owner
    Head of Research, DataLab

    Smallest test

    Publish the current Disease Odyssey table.

    Why it works

    Publish Disease Odyssey results by model on a leaderboard page, with method notes and the date of each run, and re-run when major models are released. Invite builders to submit their models.

    A living leaderboard earns repeat visits, press and citations, and every new model release is a reason to post.

    Risks and how to handle them

    • Test contamination over time

      Rotate held-out cases and say so.

    Evidence

    • Your websiteLaunch post "Announcing Independent Evaluations by Protege" (published October 21, 2026): "91% of healthcare organizations cannot verify AI governance"; compares the role to "JD Power", "the FICO score" and "Gartner"; Protege's partners compile "multi-modal data covering 300 million Americans"; HealthSAW "outperformed leading frontier models on 14 of the 16 disease-risk prediction workflows"; on the Disease Odyssey benchmark "Today's leading models get the diagnosis right only 51-67% of the time" for conditions such as sleep apnea and anemia; and "evaluations need to be refreshed as often as the models they are testing".Source
    • Competitor researchRivals' own homepages: Vals AI "independently tests AI models on real-world work. Explore benchmark results, model comparisons"; Scale delivers "data, evaluations, and outcomes to AI labs"; Defined.ai is an "AI Training Data Platform for Enterprise AI"; Truveta offers "real-time data from 130M+ patients across US health systems"; Datavant runs a network of "350+ real-world data partners".Source
    • HiringHead of Research, DataLab (remote, posted October 21, 2026): DataLab publishes "on the questions that matter: what agentic AI should actually be trained to do, how to quality-control large-scale corpora, and how to build evaluation datasets that reflect the real world rather than the leaderboard"; the head will "Serve as a customer-facing research partner to GTM" and "Represent Protege externally".Source
    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
  • 09

    Two new partnership managers will run partners "from first conversation through NDA, sample intake, Data Lab evaluation". The data provider pages end with "Become a data partner", but there is no structured way to describe a dataset before the first call.

    High impactMedium effort First result in 1 quarter Data Partnerships Manager

    First action

    Draft the intake fields with counsel.

    How to measure it

    Qualified data partner leads

    Now
    –
    Target
    +40%
    By
    2 quarters

    Steps

    1. 1

      Fields

      Asset-level only.

      Data Partnerships Manager 1 week

    2. 2

      Legal

      Wording and no-sample rule.

      Sr. Counsel, Commercial and Data 1 week

    3. 3

      Build

      Form on provider pages.

      Web developer 1 week

    4. 4

      Route

      By domain.

      Data Partnerships Manager 1 week

    5. 5

      Reply

      Review steps and timing.

      Partnerships Manager, Healthcare Admin & Operations 1 week

    6. 6

      Measure

      Lead to NDA rate.

      Data Partnerships Manager monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Data Partnerships Manager

    Smallest test

    Launch the intake on the healthcare provider page.

    Why it works

    Add an intake for data holders that asks only about the asset: domain, modality, format, volume, time span, geography, refresh cadence, rights and licensing position, de-identification already applied. Never ask for samples or records in the form. Route by domain to the right partnerships manager, and reply with the review steps and timelines.

    Structured intake speeds qualification and shows data holders a careful process from the first step.

    Risks and how to handle them

    • Someone uploads sensitive data anyway

      No upload field; state clearly not to send records.

    Evidence

    • HiringTwo partnership roles posted October 22, 2026, Data Partnerships Manager ("with a focus in RCM") and Partnerships, Healthcare Admin & Operations: turn "an unmet data need into a clear sourcing strategy"; build relationships with "data holders, aggregators, brokers, and channel partners"; move partners "from first conversation through NDA, sample intake, Data Lab evaluation, feedback, diligence, deal structuring, contracting, activation"; and assess "PHI and PII exposure, de-identification methodology, re-identification risk" and "any restrictions on AI training or model evaluation".Source
    • Your websiteHealthcare page for data providers (/data-provider/healthcare, checked October 22, 2026) still shows "Live Webinar on Oct 13, 12:30pm EST: Learn how Protege better evaluates healthcare AI with real-world data. Register Here", while the model builders' healthcare page links the replay. The page lists "5B+" total patient encounters and "700K+" average tokens per patient, and a four-step process ending in "Delivery & Revenue Recognition", where Protege pays "revenue share to the participating partner(s)".Source
    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source
  • 10

    Protege has announced about 40 partnerships, each as a separate news post. A model builder looking for pathology or imaging sources, or a data holder checking who else works with Protege, has to read them one by one.

    Medium impactMedium effort First result in 1 quarter Data Partnerships Manager

    First action

    Ask the first ten partners for a directory listing.

    How to measure it

    Partner inquiries citing the directory

    Now
    –
    Target
    20 a quarter
    By
    2 quarters

    Steps

    1. 1

      Consent

      Partner approval.

      Data Partnerships Manager 2 weeks

    2. 2

      Cards

      Domain and modality.

      Technical Product & Content Marketing Lead 2 weeks

    3. 3

      Page

      With filters.

      Web developer 2 weeks

    4. 4

      Process

      Add on signing.

      Data Partnerships Manager ongoing

    5. 5

      Promote

      In outreach.

      Partnerships Manager, Healthcare Admin & Operations ongoing

    6. 6

      Measure

      Views and inquiries.

      Technical Product & Content Marketing Lead monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Data Partnerships Manager

    Smallest test

    Publish the directory with ten partners.

    Why it works

    Publish a directory with one card per partner who agrees: domain, modality, a one-line description and a link to the announcement. Filter by domain. Add new partners on signing.

    The network is Protege's moat; showing it in one place helps both sides of the marketplace.

    Risks and how to handle them

    • Partners prefer discretion

      List only those who opt in.

    Evidence

    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source
    • HiringTwo partnership roles posted October 22, 2026, Data Partnerships Manager ("with a focus in RCM") and Partnerships, Healthcare Admin & Operations: turn "an unmet data need into a clear sourcing strategy"; build relationships with "data holders, aggregators, brokers, and channel partners"; move partners "from first conversation through NDA, sample intake, Data Lab evaluation, feedback, diligence, deal structuring, contracting, activation"; and assess "PHI and PII exposure, de-identification methodology, re-identification risk" and "any restrictions on AI training or model evaluation".Source
    • Your websiteFunding release (January 8, 2026): a $30 million Series A extension led by Andreessen Horowitz, bringing "total funding to $65 million since the company's founding in 2024"; Protege supplies "media content, audio recordings, de-identified health records, and medical imaging" through licensing agreements with data providers.Source
  • 11

    Protege's case studies mostly tell the data provider story (a broadcaster earning $1M+, a sports rights distributor) or a delivery feat. Model builders want to know how much faster they shipped or how their model improved.

    High impactMedium effort First result in 1 quarter Technical Product & Content M…

    First action

    Ask three model builder customers for a story.

    How to measure it

    Stories live

    Now
    –
    Target
    3
    By
    2 quarters

    Steps

    1. 1

      Pick customers

      Three domains.

      Strategic Account Manager 2 weeks

    2. 2

      Interview

      Results and timing.

      Technical Product & Content Marketing Lead 1 month

    3. 3

      Approve

      Customer and counsel.

      Sr. Counsel, Commercial and Data 1 month

    4. 4

      Publish

      Story pages.

      Web developer 1 week

    5. 5

      One-pagers

      For sales.

      Technical Product & Content Marketing Lead 1 week

    6. 6

      Use

      In deals.

      Strategic Account Manager ongoing

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Secure one builder story this month.

    Why it works

    Write three builder stories (one healthcare, one media, one spatial) with time to dataset, iterations saved and model results, approved by the customer. Turn each into a one-pager and talk track.

    Buyer-side proof is what the commercial team needs in competitive deals against Scale and other data providers.

    Risks and how to handle them

    • Customers keep data sources confidential

      Anonymise the customer if needed and keep the numbers.

    Evidence

    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source
    • Your websiteHomepage: "Real World Data for AI Development" for pre-training, post-training and "Evaluation & Benchmarks"; separate paths for model builders and data providers; a Siemens Healthineers quote; article dates are printed as numbers in month-day-year order (the January 8, 2026 funding release shows "01 07 2026", an October 2025 case study "10 08 2025"); the footer reads "© 2025 Protege Health, Inc." The site runs on WordPress.Source
    • HiringTechnical Product & Content Marketing Lead (remote, posted August 27, 2026): "Translate Protege's Product efforts across verticals into clear public-facing positioning"; turn "DataLab's technical research into buyer language"; "Own end-to-end launch campaigns for new Product, DataLab, and Commercial releases"; build "decks, webinars, videos, social media content, one-pagers, case studies, talk tracks, landing pages, and launch materials"; partner with the Head of Community; and use "the right growth channels to distribute your work".Source
    • Competitor researchRivals' own homepages: Vals AI "independently tests AI models on real-world work. Explore benchmark results, model comparisons"; Scale delivers "data, evaluations, and outcomes to AI labs"; Defined.ai is an "AI Training Data Platform for Enterprise AI"; Truveta offers "real-time data from 130M+ patients across US health systems"; Datavant runs a network of "350+ real-world data partners".Source
  • 12

    DataLab publishes research on corpora quality, egocentric video and clinical model evaluation, and the new Head of Research must bring research into sales. The marketing lead is asked to turn research into buyer language.

    High impactMedium effort First result in 1 quarter Head of Research

    First action

    Pick the first finding from recent DataLab work.

    How to measure it

    Substack subscribers

    Now
    –
    Target
    +2,000
    By
    2 quarters

    Steps

    1. 1

      Pick finding

      Monthly.

      Head of Research, DataLab monthly

    2. 2

      Write

      Buyer language.

      Technical Product & Content Marketing Lead monthly

    3. 3

      Chart

      One per brief.

      Applied Healthcare Researcher monthly

    4. 4

      Publish

      Substack and site.

      Technical Product & Content Marketing Lead monthly

    5. 5

      Live session

      Quarterly.

      Head of Research, DataLab quarterly

    6. 6

      Measure

      Subscribers and replies.

      Technical Product & Content Marketing Lead monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Head of Research, DataLab

    Smallest test

    Send the first brief.

    Why it works

    Publish a monthly brief on Substack and the site: one finding, a chart, three implications for model builders and a link to the relevant data or evaluation offer. Follow it with a 30-minute live session each quarter.

    Regular, useful research keeps Protege in front of research teams between deals.

    Risks and how to handle them

    • Research cadence slips

      Keep a backlog of three briefs.

    Evidence

    • HiringHead of Research, DataLab (remote, posted October 21, 2026): DataLab publishes "on the questions that matter: what agentic AI should actually be trained to do, how to quality-control large-scale corpora, and how to build evaluation datasets that reflect the real world rather than the leaderboard"; the head will "Serve as a customer-facing research partner to GTM" and "Represent Protege externally".Source
    • HiringTechnical Product & Content Marketing Lead (remote, posted August 27, 2026): "Translate Protege's Product efforts across verticals into clear public-facing positioning"; turn "DataLab's technical research into buyer language"; "Own end-to-end launch campaigns for new Product, DataLab, and Commercial releases"; build "decks, webinars, videos, social media content, one-pagers, case studies, talk tracks, landing pages, and launch materials"; partner with the Head of Community; and use "the right growth channels to distribute your work".Source
    • Your websiteDataLab site (datalab.withprotege.ai): research on grounding egocentric video for robot learning, a quality framework for motion-capture data and "Counterfactual Evaluation Reveals Hidden Capability Profiles in Clinical LLMs and Agents"; news such as "Healthcare AI Needs An Independent Referee"; calls to action "Got Data?" and a Substack signup.Source
  • 13

    Protege's domain pages describe data types in general terms (structured, unstructured, imaging, documents). Research teams shortlisting providers want to see example dataset descriptions: modality, volume ranges, formats, rights and turnaround.

    Medium impactMedium effort First result in 1 quarter Technical Product & Content M…

    First action

    Write three healthcare dataset cards.

    How to measure it

    Card-driven requests

    Now
    –
    Target
    15 a quarter
    By
    2 quarters

    Steps

    1. 1

      Template

      Card fields.

      Technical Product & Content Marketing Lead 1 week

    2. 2

      Write

      Three per domain.

      Solutions Lead, Healthcare 1 month

    3. 3

      Review

      Partner and legal.

      Sr. Counsel, Commercial and Data 2 weeks

    4. 4

      Publish

      On domain pages.

      Web developer 2 weeks

    5. 5

      Request button

      Routed by domain.

      Web developer 1 week

    6. 6

      Measure

      Requests per card.

      Technical Product & Content Marketing Lead monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Publish three healthcare cards.

    Why it works

    Publish a few anonymised dataset cards per domain describing features only (modality, formats, volume range, time span, rights and turnaround), with no records or samples, and a "request this kind of data" button.

    Concrete cards help buyers self-qualify and give search engines specific content to index.

    Risks and how to handle them

    • Partners' confidentiality

      Describe types, never name sources without consent.

    Evidence

    • Your websiteHomepage: "Real World Data for AI Development" for pre-training, post-training and "Evaluation & Benchmarks"; separate paths for model builders and data providers; a Siemens Healthineers quote; article dates are printed as numbers in month-day-year order (the January 8, 2026 funding release shows "01 07 2026", an October 2025 case study "10 08 2025"); the footer reads "© 2025 Protege Health, Inc." The site runs on WordPress.Source
    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • Your websiteHealthcare page for data providers (/data-provider/healthcare, checked October 22, 2026) still shows "Live Webinar on Oct 13, 12:30pm EST: Learn how Protege better evaluates healthcare AI with real-world data. Register Here", while the model builders' healthcare page links the replay. The page lists "5B+" total patient encounters and "700K+" average tokens per patient, and a four-step process ending in "Delivery & Revenue Recognition", where Protege pays "revenue share to the participating partner(s)".Source
  • 14

    Real visitors wait 3.3 seconds for the homepage's main content, above the 2.5-second "good" mark, and the lab test points at the hero image. Everything else (SEO, accessibility, interaction) already scores well.

    Medium impactLow effort First result in 1 month Web developer

    First action

    Compress and preload the hero image.

    How to measure it

    Real-visitor LCP p75

    Now
    –
    Target
    Under 2.5 s
    By
    60 days

    Steps

    1. 1

      Image

      Format and size.

      Web developer 1 day

    2. 2

      Preload

      Hero only.

      Web developer 1 hour

    3. 3

      Scripts

      Defer sliders.

      Web developer 1 day

    4. 4

      Cache

      CDN rules.

      Web developer 1 day

    5. 5

      Retest

      Lab.

      Web developer 1 day

    6. 6

      Field check

      After a month.

      Web developer 1 month

    First result
    1 month
    Cost
    Developer time
    Owner
    Web developer

    Smallest test

    Fix the hero image this week.

    Why it works

    Serve the hero image in a modern format at the right size, preload it and drop any slider scripts it waits for. Recheck field data a month later.

    Research buyers often open the site on the move after a conference or email; a faster first screen keeps them.

    Risks and how to handle them

    • Visual change

      Keep the same design.

    Evidence

    • Google PageSpeedwithprotege.ai: real Chrome visitors' Largest Contentful Paint (p75) is 3.3 seconds on the homepage and 2.8 s across the site, above the 2.5-second "good" mark; interaction (62 ms) and layout shift (0.01) pass. Mobile lab score 85 with LCP 4.2 s on the hero image; SEO and accessibility both 100.Source
  • 15

    The marketing lead must build decks, talk tracks and landing pages for the commercial team across healthcare, video, audio, spatial and other domains. Sellers in each vertical currently draw on news posts.

    High impactMedium effort First result in 1 quarter Technical Product & Content M…

    First action

    Start with the healthcare kit.

    How to measure it

    Win rate

    Now
    –
    Target
    +10 points
    By
    2 quarters

    Steps

    1. 1

      Interview sellers

      Common objections.

      Technical Product & Content Marketing Lead 2 weeks

    2. 2

      Healthcare kit

      Deck, talk track, page.

      Technical Product & Content Marketing Lead 1 month

    3. 3

      Media kit

      Same set.

      Technical Product & Content Marketing Lead 1 month

    4. 4

      Audio and spatial

      Same set.

      Technical Product & Content Marketing Lead 1 month

    5. 5

      Train

      Sellers.

      Strategic Account Manager 1 week

    6. 6

      Review

      Quarterly.

      Technical Product & Content Marketing Lead quarterly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Ship the healthcare kit.

    Why it works

    For each vertical, build one deck, one talk track with objections (rights, privacy, contamination, price), one landing page and one case study. Review with the account managers every quarter.

    Consistent materials raise win rates and shorten ramp time for new sellers.

    Risks and how to handle them

    • Kits go stale

      Quarterly review owner.

    Evidence

    • HiringTechnical Product & Content Marketing Lead (remote, posted August 27, 2026): "Translate Protege's Product efforts across verticals into clear public-facing positioning"; turn "DataLab's technical research into buyer language"; "Own end-to-end launch campaigns for new Product, DataLab, and Commercial releases"; build "decks, webinars, videos, social media content, one-pagers, case studies, talk tracks, landing pages, and launch materials"; partner with the Head of Community; and use "the right growth channels to distribute your work".Source
    • HiringOther open roles: Solutions Lead, Healthcare (cohorts across "EHR, claims, radiology, pathology, and unstructured text", with "HIPAA-aware handling"), Strategic Account Manager, Applied Healthcare Researcher, Forward Deployed Engineer, Audio, and Sr. Counsel, Commercial and Data. Every post describes "the secure, efficient, and privacy-centric exchange of AI training data".Source
    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source
  • 16

    A broadcaster earned over $1M in six months licensing its library through Protege, and the partnership managers source "data holders, aggregators, brokers". Many broadcasters, studios and archives hold similar libraries.

    High impactMedium effort First result in 1 quarter Data Partnerships Manager

    First action

    List 50 rights holders by region.

    How to measure it

    New media partners

    Now
    –
    Target
    10
    By
    2 quarters

    Steps

    1. 1

      Target list

      50 rights holders.

      Data Partnerships Manager 2 weeks

    2. 2

      Outreach

      Case study and briefing invite.

      Data Partnerships Manager 1 month

    3. 3

      Briefing

      For archive owners.

      Technical Product & Content Marketing Lead 1 month

    4. 4

      Rights check

      Licensing terms.

      Sr. Counsel, Commercial and Data ongoing

    5. 5

      Sign

      Partner agreements.

      Data Partnerships Manager 1 quarter

    6. 6

      Measure

      Signed and revenue.

      Head of GTM Partnerships quarterly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Data Partnerships Manager

    Smallest test

    Send the case study to ten rights holders.

    Build the prospect list. The list compiled from public sources, verified, deduplicated and delivered to your CRM. Usually live within days.

    Book a working session

    Why it works

    Build a target list of rights holders by region and catalogue type, send the case study with a short estimate of what kinds of content buyers want, and host a briefing for archive owners with rights and licensing protections front and centre.

    A concrete payout figure is the most persuasive message to a rights holder looking for new revenue.

    Risks and how to handle them

    • Results vary by catalogue

      Show ranges, not promises.

    Evidence

    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source
    • HiringTwo partnership roles posted October 22, 2026, Data Partnerships Manager ("with a focus in RCM") and Partnerships, Healthcare Admin & Operations: turn "an unmet data need into a clear sourcing strategy"; build relationships with "data holders, aggregators, brokers, and channel partners"; move partners "from first conversation through NDA, sample intake, Data Lab evaluation, feedback, diligence, deal structuring, contracting, activation"; and assess "PHI and PII exposure, de-identification methodology, re-identification risk" and "any restrictions on AI training or model evaluation".Source
    • Your websiteFunding release (January 8, 2026): a $30 million Series A extension led by Andreessen Horowitz, bringing "total funding to $65 million since the company's founding in 2024"; Protege supplies "media content, audio recordings, de-identified health records, and medical imaging" through licensing agreements with data providers.Source

Big bets 3

Larger plays that can change the trajectory.

  • 17

    The launch argues that evaluations "need to be refreshed as often as the models they are testing", and Protege says its data pipeline refreshes continuously. A one-off attestation ages as soon as the model updates.

    High impactHigh effort First result in 3 quarters Head of Research

    First action

    Price a subscription for the first assessments customers.

    How to measure it

    Subscriptions

    Now
    –
    Target
    8
    By
    4 quarters

    Steps

    1. 1

      Design

      Cadence and scope.

      Head of Research, DataLab 1 month

    2. 2

      Price

      Annual tiers.

      GTM Commercial Lead, Assessments 1 month

    3. 3

      Pipeline

      Fresh held-out cases.

      Applied Healthcare Researcher 1 quarter

    4. 4

      Registry updates

      Versioned pages.

      Web developer 1 month

    5. 5

      Sell

      To first customers.

      GTM Commercial Lead, Assessments 1 quarter

    6. 6

      Review

      Renewals.

      Head of GTM Partnerships quarterly

    First result
    3 quarters
    Cost
    Compute and staff
    Owner
    Head of Research, DataLab

    Smallest test

    Offer a re-run to one existing evaluation customer.

    Why it works

    Offer an annual subscription that re-runs the evaluation on new held-out cases with each model release, updates the attestation page, and alerts the builder and opted-in buyers to changes.

    A subscription turns one-off assessments into recurring revenue and keeps the registry current.

    Risks and how to handle them

    • Held-out data runs out

      Plan supply with the partnerships team.

    Evidence

    • Your websiteLaunch post "Announcing Independent Evaluations by Protege" (published October 21, 2026): "91% of healthcare organizations cannot verify AI governance"; compares the role to "JD Power", "the FICO score" and "Gartner"; Protege's partners compile "multi-modal data covering 300 million Americans"; HealthSAW "outperformed leading frontier models on 14 of the 16 disease-risk prediction workflows"; on the Disease Odyssey benchmark "Today's leading models get the diagnosis right only 51-67% of the time" for conditions such as sleep apnea and anemia; and "evaluations need to be refreshed as often as the models they are testing".Source
    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • HiringGTM Commercial Lead, Assessments (remote, posted September 12, 2026): owns "the commercial strategy and execution for our benchmarking and assessments business"; in 30 days "Build a target account map of 15 to 20 vertical healthcare AI builders"; in 60 days "Close first tranche of assessments customers"; reports to the head of GTM Partnerships.Source
  • 18

    Protege is hiring two partnership managers for revenue cycle and healthcare administration data, and its data already powers Vals AI benchmarks in clinical documentation and medical billing. Administrative AI is a fast-growing buyer group with few trusted tests.

    High impactHigh effort First result in 3 quarters Partnerships Manager

    First action

    Scope one coding benchmark with DataLab.

    How to measure it

    Admin evaluations

    Now
    –
    Target
    5
    By
    4 quarters

    Steps

    1. 1

      Scope

      Three workflows.

      Head of Research, DataLab 1 month

    2. 2

      Supply

      Partner data for tests.

      Partnerships Manager, Healthcare Admin & Operations 1 quarter

    3. 3

      Privacy review

      De-identification and rights.

      Sr. Counsel, Commercial and Data 1 month

    4. 4

      Build

      First benchmark.

      Applied Healthcare Researcher 1 quarter

    5. 5

      Sell

      Admin AI builders.

      GTM Commercial Lead, Assessments 1 quarter

    6. 6

      Publish

      Opted-in results.

      Technical Product & Content Marketing Lead ongoing

    First result
    3 quarters
    Cost
    Research and data costs
    Owner
    Partnerships Manager, Healthcare Admin & Operations

    Smallest test

    Scope one coding benchmark.

    Why it works

    Build independent evaluations for admin workflows (coding, denials, prior authorization), run them double-blind like the clinical ones, and sell them to admin AI builders and the payers and providers who buy from them.

    It extends the Assessments business into a second market with the same method and supply the new hires are sourcing.

    Risks and how to handle them

    • Workflow data is sensitive

      Same privacy and rights standards as clinical work.

    Evidence

    • HiringTwo partnership roles posted October 22, 2026, Data Partnerships Manager ("with a focus in RCM") and Partnerships, Healthcare Admin & Operations: turn "an unmet data need into a clear sourcing strategy"; build relationships with "data holders, aggregators, brokers, and channel partners"; move partners "from first conversation through NDA, sample intake, Data Lab evaluation, feedback, diligence, deal structuring, contracting, activation"; and assess "PHI and PII exposure, de-identification methodology, re-identification risk" and "any restrictions on AI training or model evaluation".Source
    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • HiringGTM Commercial Lead, Assessments (remote, posted September 12, 2026): owns "the commercial strategy and execution for our benchmarking and assessments business"; in 30 days "Build a target account map of 15 to 20 vertical healthcare AI builders"; in 60 days "Close first tranche of assessments customers"; reports to the head of GTM Partnerships.Source
  • 19

    Protege sees demand and supply for AI training data across healthcare, media, audio, motion capture and more, and has raised $65M to scale it. No one else reports on which kinds of real-world data AI teams are asking for.

    High impactMedium effort First result in 3 quarters Technical Product & Content M…

    First action

    Pick ten questions buyers and data holders ask.

    How to measure it

    Downloads

    Now
    –
    Target
    3,000
    By
    1 year

    Steps

    1. 1

      Questions

      Ten.

      Technical Product & Content Marketing Lead 1 week

    2. 2

      Data

      Aggregated marketplace data.

      Head of Research, DataLab 2 months

    3. 3

      Confidentiality review

      No customer or partner identifiable.

      Sr. Counsel, Commercial and Data 1 month

    4. 4

      Write

      Report.

      Technical Product & Content Marketing Lead 1 month

    5. 5

      Launch

      Press and webinar.

      Technical Product & Content Marketing Lead 1 month

    6. 6

      Use

      In sales.

      Strategic Account Manager ongoing

    First result
    3 quarters
    Cost
    Staff time
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Publish one chart on demand by modality.

    Why it works

    Publish a yearly report with aggregated, anonymous marketplace figures (demand by modality, typical rights terms, time to delivery, evaluation trends) plus DataLab commentary. Launch with press and a webinar.

    Original market data earns press and citations and attracts both builders and data holders.

    Risks and how to handle them

    • Revealing deal terms

      Only ranges and aggregates.

    Evidence

    • Your websiteFunding release (January 8, 2026): a $30 million Series A extension led by Andreessen Horowitz, bringing "total funding to $65 million since the company's founding in 2024"; Protege supplies "media content, audio recordings, de-identified health records, and medical imaging" through licensing agreements with data providers.Source
    • HiringHead of Research, DataLab (remote, posted October 21, 2026): DataLab publishes "on the questions that matter: what agentic AI should actually be trained to do, how to quality-control large-scale corpora, and how to build evaluation datasets that reflect the real world rather than the leaderboard"; the head will "Serve as a customer-facing research partner to GTM" and "Represent Protege externally".Source
    • Your websiteSitemap (96 URLs, checked October 22, 2026): about 40 partnership announcements, case studies (a broadcaster that "Earns $1M+ in 6 Months", a sports rights distributor that "Unlocks $250K", imaging "for pre-training in 30 days"), the three evaluation results, and tag pages; there is no page listing every published evaluation or attestation, no page per data partner and no page for health systems or payers that buy AI.Source

Wildcards 3

Unexpected experiments, each with a cheap way to test it.

  • 20

    The evals page already has tabs for Audio & Speech and Spatial & Physical next to healthcare, and Protege is hiring a forward deployed engineer for audio. Speech models face the same "graded their own homework" problem.

    Medium impactMedium effort First result in 2 quarters Forward Deployed Engineer

    First action

    Scope one audio evaluation.

    How to measure it

    Audio evaluations sold

    Now
    –
    Target
    3
    By
    3 quarters

    Steps

    1. 1

      Scope

      One task.

      Forward Deployed Engineer, Audio 1 month

    2. 2

      Data

      Licensed held-out audio.

      Data Partnerships Manager 1 month

    3. 3

      Run

      Leading models.

      Forward Deployed Engineer, Audio 1 month

    4. 4

      Publish

      Registry and post.

      Technical Product & Content Marketing Lead 2 weeks

    5. 5

      Sell

      Speech builders.

      Strategic Account Manager 1 quarter

    6. 6

      Review

      Next domain.

      Head of Research, DataLab quarterly

    First result
    2 quarters
    Cost
    Compute and data
    Owner
    Forward Deployed Engineer, Audio

    Smallest test

    Scope the audio evaluation.

    Why it works

    Design one double-blind evaluation for speech (for example accents or noisy environments) on licensed held-out audio, run it on leading models and publish the results in the registry.

    The method and brand transfer, and audio is a domain where Protege already supplies data.

    Risks and how to handle them

    • Spreading the team thin

      One domain at a time.

    Evidence

    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • HiringOther open roles: Solutions Lead, Healthcare (cohorts across "EHR, claims, radiology, pathology, and unstructured text", with "HIPAA-aware handling"), Strategic Account Manager, Applied Healthcare Researcher, Forward Deployed Engineer, Audio, and Sr. Counsel, Commercial and Data. Every post describes "the secure, efficient, and privacy-centric exchange of AI training data".Source
    • Your websiteDataLab site (datalab.withprotege.ai): research on grounding egocentric video for robot learning, a quality framework for motion-capture data and "Counterfactual Evaluation Reveals Hidden Capability Profiles in Clinical LLMs and Agents"; news such as "Healthcare AI Needs An Independent Referee"; calls to action "Got Data?" and a Substack signup.Source
  • 21

    The evals page has a strong chart: public benchmarks send a model a fraction of one patient's record (median 13 notes per patient), with HealthBench around 170 tokens per case. It is the clearest argument for Protege's approach.

    Medium impactLow effort First result in 1 month Technical Product & Content M…

    First action

    Publish the chart as its own page.

    How to measure it

    Referring domains

    Now
    –
    Target
    30
    By
    2 quarters

    Steps

    1. 1

      Data

      Benchmark figures.

      Applied Healthcare Researcher 1 week

    2. 2

      Build

      Interactive page.

      Web developer 2 weeks

    3. 3

      Method

      Sources.

      Head of Research, DataLab 1 week

    4. 4

      Embed

      Share code.

      Web developer 2 days

    5. 5

      Promote

      Social and newsletter.

      Technical Product & Content Marketing Lead 2 weeks

    6. 6

      Measure

      Links and visits.

      Technical Product & Content Marketing Lead monthly

    First result
    1 month
    Cost
    Developer time
    Owner
    Technical Product & Content Marketing Lead

    Smallest test

    Publish the static chart as its own page.

    Why it works

    Make it an interactive page: pick a benchmark, see how much context it gives versus a real record, with the method and sources. Make it easy to embed and share.

    A shareable, visual argument spreads further than a static chart and links back to the evals page.

    Risks and how to handle them

    • Benchmarks update

      Date the figures.

    Evidence

    • Your websiteHealthcare Evals page (/model-builders/healthcare-evaluation, checked October 22, 2026): "Model makers should not grade their own homework."; independent, double-blind evaluations "on uncontaminated real-world patient data"; "De-identified records on ~40M patients. The data never leaves your private sandbox."; "At most two runs on the held-out test set"; "Receive score, report, attestation. Publish it as a double-blind evaluation by Protege — or keep it private"; "A credible go-to-market signal: Expand CPT codes, win health systems and payers". The three published results (Xlue AI, Standard Model Bio, Disease Odyssey) are individual blog posts, and both "SEE HOW YOUR MODEL SCORES" buttons lead to a general contact form at /contact-healthcare-evals. The page has tabs for Audio & Speech and Spatial & Physical next to Healthcare Evaluations.Source
    • HiringHead of Research, DataLab (remote, posted October 21, 2026): DataLab publishes "on the questions that matter: what agentic AI should actually be trained to do, how to quality-control large-scale corpora, and how to build evaluation datasets that reflect the real world rather than the leaderboard"; the head will "Serve as a customer-facing research partner to GTM" and "Represent Protege externally".Source
  • 22

    The launch sets out six pillars for trustworthy healthcare evaluation (real-world data, new data, real workflows, clinician-verified answers, independence, speed). A standard is stronger when others sign it.

    Medium impactMedium effort First result in 2 quarters Head of Research

    First action

    Invite five clinicians from past webinars and partners.

    How to measure it

    Signatories

    Now
    –
    Target
    25
    By
    3 quarters

    Steps

    1. 1

      Invite

      Five council members.

      Head of Research, DataLab 1 month

    2. 2

      Review

      Principles.

      Head of Research, DataLab 1 month

    3. 3

      Publish

      Open standard.

      Technical Product & Content Marketing Lead 2 weeks

    4. 4

      Signatories

      Collect.

      Head of Community 1 quarter

    5. 5

      Meet

      Twice a year.

      Head of Research, DataLab biannually

    6. 6

      Update

      Principles.

      Head of Research, DataLab yearly

    First result
    2 quarters
    Cost
    Staff time
    Owner
    Head of Research, DataLab

    Smallest test

    Invite the first five members.

    Why it works

    Invite a small council of health system AI leaders and clinicians to review the principles, publish them as an open standard with signatories, and meet twice a year to update them.

    Co-signed standards make Protege the convener of the conversation and bring buyers closer.

    Risks and how to handle them

    • Seen as marketing

      Keep the standard vendor-neutral.

    Evidence

    • Your websiteLaunch post "Announcing Independent Evaluations by Protege" (published October 21, 2026): "91% of healthcare organizations cannot verify AI governance"; compares the role to "JD Power", "the FICO score" and "Gartner"; Protege's partners compile "multi-modal data covering 300 million Americans"; HealthSAW "outperformed leading frontier models on 14 of the 16 disease-risk prediction workflows"; on the Disease Odyssey benchmark "Today's leading models get the diagnosis right only 51-67% of the time" for conditions such as sleep apnea and anemia; and "evaluations need to be refreshed as often as the models they are testing".Source
    • HiringHead of Research, DataLab (remote, posted October 21, 2026): DataLab publishes "on the questions that matter: what agentic AI should actually be trained to do, how to quality-control large-scale corpora, and how to build evaluation datasets that reflect the real world rather than the leaderboard"; the head will "Serve as a customer-facing research partner to GTM" and "Represent Protege externally".Source

03

How Programmable can help

Where an idea in this playbook is something Programmable builds or runs, it's listed here with what we'd do. Everything else is yours to run.

Have it built

ProgrammableAgency

1 of the 22 ideas are things Programmable builds and runs for its customers: pages, outreach, prospect lists, booking flows and monitors. Pick one, or send the whole list.

Prospect lists1The list compiled from public sources, verified, deduplicated and delivered to your CRM.

A 30-minute session with the engineer who would build it, not a sales call. You leave with a plan whether or not you hire us.

Book a working session

Usually live within days.

Tools for the rest

ProgrammableTalk

Fill the events in this plan

These ideas are events. Talk gets registrants for them and tells you who's coming and what their company does before the day.

Plan an event (opens programmable.talk in a new tab)
ProgrammableDigital

Show it, don't just describe it

These ideas need something to show: a demo, a video, an animation, graphics or a deck. Digital turns screenshots or the site into all of them, in the company's own brand.

Make a demo (opens programmable.digital in a new tab)
More example playbooks and reviews

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • DataDome growth strategyBot and AI agent trust management: real-time detection of bots, fraud and AI agents across websites, apps, APIs and MCP servers, with…20 growth ideas
  • Veritus growth strategyAn AI communications platform for financial services: compliant voice, SMS, email and chat agents for origination, servicing and…23 growth ideas
  • Cascade Space growth strategyBuilds a commercial alternative to NASA's Deep Space Network: a ground array of smaller dishes combined by digital beamforming, with…22 growth ideas
  • Talkspace growth strategyOnline therapy and psychiatry with licensed providers by messaging, video and voice, covered by most major insurance plans, plus Talkspace…21 growth ideas
  • Code3 growth strategyAn independent integrated marketing agency (owned by Graham Holdings) that runs paid social and media, Amazon and retail media, creative…22 growth ideas
  • Canonical growth strategyThe publisher of Ubuntu and a portfolio of open source infrastructure: Ubuntu Pro, OpenStack, Kubernetes, MAAS, Juju, LXD, data and AI…22 growth ideas
  • Blockchain.com growth strategyA crypto finance company: a consumer app for buying, trading, earning and borrowing, a self-custody wallet, the Blockchain Explorer, a…20 growth ideas
  • Thoughtly growth strategyAI voice and text agents that call new leads within seconds, answer questions, qualify them and book meetings, logged to the CRM; building…20 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on October 23, 2026, and cites them throughout. Protege did not commission, review or endorse it, and Programmable is not affiliated with Protege. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how Protege runs its business.

Something wrong, or do you represent Protege? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 307 ideas

    Evaluation registry and attestation pages, evaluation intake, webinar banner, date format, scale definitions, targeted builder outreach; start the hero speed fix.

  2. Days 31 to 6010 ideas

    Buyer guide, Disease Odyssey leaderboard, data-holder intake, partner directory, builder case studies, DataLab brief, dataset cards, vertical kits and media partner recruiting; the context explainer.

  3. Days 61 to 905 ideas

    Commit to continuous evaluation, administration benchmarks and the market report; scope audio evaluations and the principles council.

The research behind the plan

3 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value A partner that helps define the data they need rather than a catalogue. The most cited problem: finding the right non-public data is slow and expensive.

  • Multimodal data linked across sources
  • Saturated public benchmarks
  • In their words: “real-world data”

Market

Competitive landscape

5 competitors compared: Scale, Vals AI, Defined.ai, Truveta and 1 more.

  • Gap to exploit: Publicly checkable healthcare AI evaluations
  • Scale: Less focus on licensed real-world healthcare data
  • Vals AI: Relies on partners for healthcare data

Hiring signals

Protege is turning its data marketplace into a two-product company.

  • GTM Commercial Lead, Assessments: First assessments customers. Ideas 1, 2, 6.
  • Technical Product & Content Marketing Lead: Launches, buyer language, field assets. Ideas 5, 11, 15.
  • Head of Research, DataLab: Research partner to GTM. Ideas 8, 12.

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks firstNew accounts get 100,000 free enrichment credits with code BETA100K (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammableMediaAn AI note taker, interpreter and webinar host for your calls (opens programmable.media in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence