Here’s an invented scenario to frame the topic. A fictional mid-size sponsor, unconnected to any real company, runs a Phase III study that needs 600 randomized participants from 120 sites. Its clinical operations lead has watched a demo of a platform that matches patients to eligibility criteria, ranks sites and sends personalized messages. The medical director asks the sensible questions: how often is the match wrong, who sees patient data, and who reviews the messages that reach participants? The demo can’t answer those. This post lays out what Salesforce documents, what independent research shows, and what a careful pilot should test.

Scope and caveats come first. I read 9 pages. Salesforce supplied 3 of them: a Trailhead unit on clinical operations, its pharma solution page and its Life Sciences Cloud pricing page. The rest were an R&D World article reporting a Salesforce product announcement, an IntuitionLabs product listing, a 2012 Applied Clinical Trials article by Ken Getz, a 2024 Nature Communications paper on automated trial matching, an HHS answer on research recruitment and an FDA information sheet. Vendor pages are marketing, and customer quotes are favorable by design. I found no independent evaluation of Salesforce’s recruitment features, so confirm every claim with Salesforce, and confirm regulatory questions with counsel.

TL;DR

  • The features are described, not measured. Salesforce’s Trailhead unit says its agent evaluates patient records against inclusion and exclusion criteria and returns Match, Not Match or Insufficient Data per criterion, and it describes site search, feasibility surveys and recruitment forecasts. The unit gives no accuracy, conversion or time-saved figures, and the closest it comes to site engagement is the feasibility survey.
  • Independent research supports caution and some hope. A 2024 NIH-funded paper on a different system reported 87.3% criterion-level accuracy and a 42.6% cut in screening time in a small pilot, using public synthetic cohorts. In my invented model, even 87.3% accuracy implies about 1.9 wrong calls per patient across 15 criteria, so human review stays essential.
  • Privacy and recruitment rules shape the design. HHS says researchers outside a covered entity can’t use the preparatory-to-research provision to contact prospects, and FDA says recruitment materials belong in IRB review. A sponsor-side platform that holds patient data, or sends personalized messages, needs counsel’s sign-off before launch.

What the platform claims to do in a trial

Salesforce describes trial recruitment as a chain of linked steps: choose sites, start them up, find eligible patients, contact them and keep them enrolled. Those functions sit inside the broader Salesforce Life Sciences environment for clinical, commercial, medical and patient operations. The table below lines up each claim with its source and with what I found as evidence. In every row, the evidence is a product description, a customer quote or a promise of outcomes, and none reports measured results. Use the table as a diligence checklist, and ask for a reference customer who can show numbers for each row that matters to you.

ClaimWhere it appearsWhat the source saysEvidence found
Site selectionTrailhead unit; pharma pageProposes filters from the study record, summarizes sites and investigatorsNone; the unit gives no accuracy or time saved
Study start-upTrailhead unitAdds sites to a study and sends feasibility surveys tracked in a portalNone
Patient matchingTrailhead unitEvaluates records against criteria using structured data and clinical notesNone; “high precision” has no data
Recruitment forecastingTrailhead unitForecasts recruitment readiness across selected sitesNone
Candidate conversionTrailhead unitSays the approach increases conversion ratesNone
RetentionPharma page; R&D WorldPersonalized messages to prevent attrition; a customer founder reports fewer dropoutsCustomer assertion only
Site performancePharma pageSays intelligent site selection measurably improves performanceNone; no metrics on the page

Why recruitment is a hard problem

Enrollment trouble is old and well documented, which is why the pitch lands. Ken Getz, writing in Applied Clinical Trials in May 2012, summarized a Tufts Center for the Study of Drug Development study of about 16,000 sites in 151 Phase II and III trials run from 2008 to 2010. It found 11% of ready-to-enroll sites enrolled no patients, 48% of all sites under-enrolled, and 53% of studies ran past their planned timelines. About 1 in 6 took more than twice as long as planned. The sample came from 10 pharma companies and 2 contract research organizations, and Getz cautions that the figures likely understate the problem.

Those figures are about 14 years old, and I found no current equivalent in this research. They show that site-level shortfalls were common, but they don’t show why, and they can’t tell you whether better matching or messaging would have helped. Getz also notes that an older benchmark, 20% of sites enrolling nobody, hardened into accepted fact from conference interviews rather than a systematic dataset. Treat any statistic in a vendor pitch the same way: ask for the source, the date, the sample and the definition of enrolled before you build a business case on it.

How patient matching is described

Salesforce’s Trailhead participant management unit gives the most detail I found about matching. It describes an agent that reads a patient record and tests it against a trial’s criteria, then reports the result per criterion with supporting evidence. The list below summarizes the documented behavior and what each item means for a team that must defend the process to an ethics committee or an auditor. All of it comes from a vendor training page, so it describes intent, not measured performance.

  • Per-criterion results. The unit says results appear as Match, Not Match or Insufficient Data for each inclusion and exclusion criterion, with supporting evidence. That structure is useful because a coordinator can see why a patient was flagged. It also creates work, since every Insufficient Data result is a decision about whether to chase more records or move on. Agree with coordinators up front on who resolves each Insufficient Data result and how fast.
  • Structured data and clinical notes. The unit says the agent uses both structured fields and free-text notes. Notes hold much of the detail that eligibility depends on, but they’re also where abbreviations, copy-forward text and contradictions live. Ask Salesforce which note types the agent reads and how it handles conflicts between a note and a coded field. The page also says the agent calculates age from a birth date and flags missing information, which are useful checks to test on real records.
  • Single patients and lists. Matching can run on a single candidate or on lists drawn from records systems, referrals or registries, according to the unit. List-level matching raises the volume of protected health information that moves into the platform, so a governed layer such as Data 360 for Healthcare & Life Sciences may become part of the architecture when records come from several sources. Define which sources feed it, who approved each feed and whether the feed carries only what screening requires. The unit names EMRs, provider referrals and a sponsor’s patient registry as list sources, and each brings its own consent and data-sharing terms.
  • A precision claim without data. The unit refers to high precision but reports no accuracy figure, no test set and no comparison to manual review. Ask Salesforce for validation results on a trial with criteria like yours. If none exist, plan your own validation with coordinators scoring a sample of matches blind, and set an acceptance threshold before you see results. Without a baseline, any quoted gain can’t be traced to the tool itself.

What independent research says about automated matching

The best independent evidence I found concerns a different system. A 2024 Nature Communications study of TrialGPT from NIH researchers describes a language-model pipeline that retrieves, matches and ranks trials for a patient. A panel of 3 physicians annotated 1,015 patient-criterion pairs from 105 patient-trial pairs, and the paper reports criterion-level accuracy of 87.3% with explanations, close to the expert range of 0.887 to 0.900. It also reports that the retrieval step can recall over 90% of relevant trials using under 6% of the initial collection.

The authors report that a small pilot found screening time fell by 42.6%, with 6 trials, 6 vignettes and 2 annotators, and they call the sample limited. Accuracy with the tool was 97.2% versus 91.7% without, a difference they describe as not statistically significant. The evaluation used public datasets of 183 synthetic patients, relied on closed models, and didn’t test longitudinal notes, labs, imaging or full record data. So the study shows promise on benchmark data, not performance in a live site, and it says nothing about Salesforce’s implementation. It’s also the best available yardstick for what to demand in a validation.

Where privacy rules shape the design

Preparatory-to-research use. HHS says the HIPAA preparatory-to-research provision permits a covered entity to use or disclose protected health information for purposes such as aiding study recruitment. A researcher who is part of the covered entity’s workforce can use it to identify and contact prospective participants and seek authorization, and providers can discuss trial enrollment with patients. The provision doesn’t let the researcher remove the information from the covered entity’s site. A platform that copies patient data to a sponsor’s cloud org needs a different legal basis than a tool that stays inside the site.

Outside researchers and sponsors. HHS says an outside researcher can’t use the preparatory provision to contact prospective subjects. Contact information may come through a partial waiver of individual authorization from an IRB or Privacy Board, and HHS notes such a body may grant the partial waiver for recruitment contact even when it doesn’t waive authorization for the study itself. For a sponsor, this means a recruitment database that holds identifiable patient data is a regulatory design question before it’s a software question. This is a summary of a single HHS answer, so confirm the specifics with counsel.

Recruitment materials and ethics review. FDA’s information sheet on recruiting study subjects says the IRB should review the methods and material investigators propose for recruiting, including advertisements and receptionist scripts that gather personal information. It treats direct advertising as the start of informed consent and subject selection. Ads shouldn’t emphasize payment or claim a product is safe or effective. Salesforce’s page promises personalized messages that prevent attrition, so ask how message templates, generated text and approvals are controlled, and who keeps the records.

Platform coverage. I didn’t read Salesforce’s business associate addendum for this post, and none of the 9 pages I read says whether Life Sciences Cloud or its Agentforce add-on is covered. A current comparison of Salesforce Life Sciences Cloud vs Health Cloud can help separate the product families before agreement coverage is checked feature by feature. Ask Salesforce to confirm in writing which services and features are covered under your signed agreement for the recruitment features you plan to use, and have counsel read it. Coverage may differ by product and edition, so a general statement about the platform isn’t enough for a regulated data flow.

An invented screening workload model

Everything in this section is invented, and I recomputed it in code. Assume a site network reviews 20,000 patient-trial pairs a year, at 25 minutes each and a loaded coordinator cost of $45 an hour. That’s 8,333 hours and $375,000 a year. The reductions below are scenarios. The 42.6% figure comes from the small pilot in the NIH paper on a different system, and the lower rows are my assumptions. No source shows that Salesforce’s matching delivers any of them.

ScenarioTime reductionHours savedInvented value at $45 an hour
Cautious10%833$37,500
Moderate25%2,083$93,750
NIH pilot figure42.6%3,550$159,750

Accuracy matters as much as speed. If criterion-level accuracy were 87.3% and a trial had 15 criteria, the expected number of wrong criterion calls would be about 1.9 per patient-trial pair, and about 38,100 a year across the network. If errors were independent, only 13% of pairs would be right on every criterion. Real errors correlate and software confidence varies, so treat these as an illustration of why human review stays, not as a forecast. The design question is where coordinators spend their review time, which should be on Insufficient Data and borderline cases.

Site engagement beyond a feasibility survey

Site engagement is the weakest-documented area. Salesforce’s Trailhead unit describes site search with filters such as past enrollment performance, short summaries of sites and investigators, an Add Site to Study step that creates linked records, and feasibility surveys that are prepopulated and tracked in a secure portal. A Salesforce Community Cloud setup is one possible internal reference when evaluating how external site users could access Salesforce-based workflows. That’s a start-up workflow. It doesn’t describe the day-to-day contact that keeps a site enrolling, so a team should decide what else it needs and whether the platform supplies it.

  • Start-up tracking. The unit describes linked site records and survey tracking as a single source of truth for the team. That helps if your contracts, regulatory documents and activation milestones also live there. If they sit in a clinical trial management system, ask how the 2 systems stay in sync, and compare that answer with the scope of Salesforce Integration Services before assuming the connection is standard. R&D World paraphrases a Centricity Research executive as saying the platform integrates with such systems for scheduling and reporting. That’s a real convenience for busy start-up teams.
  • Performance data. Site selection depends on past enrollment performance, so the quality of the history matters. If your records are incomplete or inconsistent across studies, the filters will rank sites on bad data. Audit the history before you trust a ranking, and keep a manual check on the top and bottom of any list the tool produces. Ask Salesforce which fields feed that list, how often they refresh and whether sites can correct them.
  • Ongoing contact. Nothing I read describes how coordinators receive enrollment status, protocol amendments, patient referrals or reminders. Define those workflows yourself, with named owners, response times and escalation paths, and then test whether the platform supports them or whether another tool must. The Trailhead unit stops at start-up steps and the pricing page names no site features, so plan for site communication outside the platform, or budget for another tool alongside it until a pilot shows otherwise.

Retention claims and what would prove them

What’s claimed. Salesforce’s pharma page says AI-powered matching and personalized messages prevent attrition. R&D World paraphrases Paresh Shah, the founder of a customer called CliniLink, as describing use of the platform to reduce patient dropouts through predictive analytics and automated multi-channel outreach, and it gives no direct quote. R&D World doesn’t say whether any claim was independently verified, and the descriptions and testimonials appear to be vendor-provided. Neither source reports a figure for dropouts, a baseline or a comparison group.

What’s missing? A retention claim needs a baseline dropout rate, a comparison group, a defined period and a definition of dropout. I found no such data from Salesforce. I also found no independent benchmark of dropout rates in the sources I read, so I won’t quote a number. Rates vary by condition, trial length and burden, which is why the baseline must come from your own prior studies or from a source matched to your indication.

How to measure it. Pick a single study with a stable protocol and compare retention across sites or cohorts using the platform against sites using your usual process. Track the date, channel and content of each contact, plus reasons for withdrawal. Health and life sciences teams can use Salesforce analytics for health and life sciences as a reference when defining the reporting layer for recruitment and retention measures. Have the ethics committee approve message content and frequency in advance, since reminders can feel like pressure. A fair test also records the burden placed on site staff, because a gain in retention that costs hours of extra coordinator work may not be worth it.

Pricing and cost signals

Salesforce does publish a Life Sciences Cloud pricing page, and its figures differ from a third-party listing. The page shows annual-billing prices per user per month: Core at $350, Advanced at $525, Service Max and Sales Max at $700 each, and Sales & Service Max at $725. Its metadata shows a modification date of Sept 23, 2026, and the page says pricing is for information only and subject to change. An undated IntuitionLabs listing, which says it isn’t affiliated with Salesforce, states a starting price of $325, so either that listing is stale or the 2 count different things.

The pricing page matters for recruitment in 3 specific ways. It lists a Participant Recruitment and Enrollment add-on at $10,000 per org per year, in increments of 100 patients referred, and the wording doesn’t make clear whether each increment costs $10,000 or whether that’s a flat annual fee. It lists Agentforce for Life Sciences Cloud at $150 per user per month as an add-on, available for purchase on Core and Advanced, and the Trailhead matching unit describes Agentforce features. It lists no site-specific features, and it doesn’t show which editions include patient features.

Usage and extras also count. Flex Credits are listed at 500K per org per year for Core, 1M for Advanced and 2.75M for the Max editions, and the page doesn’t explain how matching or messaging would consume them. Budget for validation, ethics review, integration with site systems, training and change management as well, since software seats are often the smaller line in a recruitment technology project. The IntuitionLabs listing also names a high starting price among its cons.

An invented return model for a single trial

This model was invented, too, and I recomputed it in code. Assume 30 users on the Core edition at $350 per user per month, which is $126,000 a year, plus the Agentforce add-on at $150 per user per month, which is $54,000, plus an invented $250,000 of first-year implementation and validation. For the recruitment add-on, I show 2 readings of the page: a flat $10,000, or an invented 20 increments at $10,000 each for $200,000. First-year cost is then $440,000 or $630,000. Assume each day of enrollment delay avoided is worth an invented $20,000 in extended trial costs.

Days of delay avoidedInvented value at $20,000 a dayNet with flat 10,000add-on(440,000 cost)Net with 20 increments ($630,000 cost)
10$200,000Negative $240,000Negative $430,000
20$400,000Negative $40,000Negative $230,000
30$600,000$160,000Negative $30,000
60$1,200,000$760,000$570,000

Break-even is 22.0 days of delay avoided under the flat reading and 31.5 days under the increment reading. The table leaves out benefits such as better data quality and costs such as ethics review and site burden. Value depends far more on whether the tool changes enrollment speed than on its price. No source I found shows that any matching or messaging tool shortens a Phase III timeline, so the days-avoided input is the assumption to test first. Use a pilot to estimate it from your own data, and rerun the model with the quote you actually receive.

Data quality and bias in matching

Notes versus structured fields. Matching that reads notes can find criteria that coded fields miss, but notes vary by clinician, system and specialty. Salesforce’s unit says the agent uses both sources and shows evidence. Ask how it cites the evidence, how it treats negation and history, such as a condition ruled out years ago, and what it does when notes contradict codes. A coordinator should be able to click from a result to the sentence that produced it.

Diversity claims. R&D World reports that Salesforce describes diverse participant matching as a capability, with no performance metrics. Broader work on generative AI in healthcare and life sciences raises the same need to validate outputs against real clinical and operational data instead of relying on product claims alone. Diversity goals are legitimate, and they’re also easy to claim and hard to measure. Ask what the feature does, which attributes it uses, how it avoids using protected characteristics inappropriately, and how its effect is measured. Counsel should review any feature that touches demographics, since fairness and anti-discrimination rules differ by jurisdiction and by use.

Selection effects. Matching can only find patients whose records are visible to the platform. Patients seen at sites that don’t share data, or whose records are sparse, will look ineligible or insufficiently documented. That bias can narrow the pool in ways nobody intended. Compare the demographic and geographic profile of matched candidates with the target population, and review Insufficient Data cases by group, so you can spot who the tool tends to miss.

Failure modes to test in a pilot

A recruitment tool fails in ordinary ways, and most can be tested before it touches a live study. The sources I read report no failure rates, so this list is my own, informed by the documented behavior and the NIH paper’s limitations. Run each test with a validation set scored by coordinators who don’t know the tool’s answer, and record results with dates so the go or no-go decision rests on evidence.

  • Wrong eligibility call. The agent marks a criterion Match when notes say otherwise, or marks Not Match when the evidence is simply missing. Measure false positives and false negatives separately, because the first wastes coordinator time and the second silently loses candidates. Set a threshold for each before the pilot begins, and publish it to the team. Review a sample of each type every week during the pilot, and feed the patterns back to Salesforce or your integrator.
  • Insufficient Data pile-up. If many candidates land in Insufficient Data, coordinators inherit a new backlog. Track how many results fall in that bucket, how long each takes to resolve and what share finally qualifies. A tool that sorts only the easy cases may not save time on the hard ones. Also record whether coordinators trust the evidence snippets enough to act on them, since a result nobody believes only adds a second review step to the same patient.
  • Unreviewed outreach. A generated message reaches a patient without ethics approval or mentions payment too prominently. The FDA’s Recruiting Study Subjects guidance makes IRB review of recruitment methods and materials part of the process, so lock templates, require approval workflows and log every send. The Trailhead unit says nothing about whether humans review matches before outreach, so don’t assume a review step exists, and design it yourself with a named approver and a time limit.

Questions to ask Salesforce and your integrator

The questions below turn the gaps in this post into requests you can send before a demo. Ask for written answers, attach them to the quote and keep them in the project file. A Salesforce consulting team for health and life sciences should be able to answer the implementation, integration, validation and compliance questions alongside the Salesforce account team. If a rep can’t answer, record that, since it’s information about the product’s maturity and about how much of the work your team will carry. Send the list to your Salesforce account team and your integrator at the same time, and compare their answers.

TopicQuestionA satisfactory answer includes
ValidationWhat accuracy results exist for matching on trials like mine?A test set, method and results, plus a reference customer
Notes handlingWhich note types does the agent read, and how are conflicts handled?A documented list and a demonstration with messy notes
Evidence trailCan a coordinator trace each result to its source text?A live demo showing citations
Data scopeWhich data feeds are needed, and what stays at the site?A data map reviewed by counsel
Agreement coverageIs each feature I plan to use covered in my signed agreement?A feature-by-feature statement in writing
Outreach controlsHow are message templates approved and logged?A workflow demo and exportable logs
Site burdenWhat do sites see, and how many extra logins or steps?A site-user walkthrough
PricingHow do agent actions and data volumes affect cost?A rate card and a volume-based estimate

Where the evidence stops

The evidence has 3 limits. First, I found no published outcome data for Salesforce’s recruitment or site engagement features, only a training unit, solution pages, an announcement and customer quotes. Second, the independent research I found concerns other systems and other settings, and its best result comes from synthetic cohorts and a small pilot. Third, the enrollment statistics are old, so they describe a problem that still exists without measuring its present size or how a tool might change it.

Those limits don’t mean the platform fails. They mean the buyer carries the burden of proof, and a good pilot can supply it at modest cost. The pilot should have a baseline, a comparison group, predefined acceptance thresholds and an ethics review of all participant-facing content. It should measure coordinator time, accuracy of matches, enrollment speed and retention, and it should end on a date with a decision to scale, adjust or stop. Anything short of that gives you a demo, which is useful but proves little.

A pilot plan for a sponsor or site network

Weeks 1 to 4: scope and approvals. This timeline is my suggestion, not a sourced benchmark. Pick 1 study and 5 to 10 sites. Write the questions above into a vendor request, and get written answers. A team using Salesforce consulting for health and life sciences can use this phase to map data sources, validation responsibilities and integration boundaries before configuration begins. Have counsel review data flows against the HHS and FDA material, decide the legal basis for each data feed, and brief the ethics committee on any participant-facing content. Capture baseline numbers for screening time, match accuracy by manual review, enrollment rate and dropout.

Weeks 5 to 8: configure and validate. Build the matching workflow on a limited criteria set and run it against a scored validation sample. Have coordinators review results without seeing the tool’s answers first, then compare. Record false positives, false negatives and Insufficient Data rates. Adjust only after recording the first results, so you can show what changed and why. Stop here if accuracy falls below the threshold you set in advance.

Weeks 9 to 20: live pilot. Turn on matching and approved outreach at the pilot sites, and keep comparison sites on the usual process. If participant communication is part of the test, compare the workflow with a defined patient engagement software model so messaging, consent, handoff and reporting requirements are explicit. Track coordinator hours, candidate flow, enrollment and withdrawals, and ask sites for feedback every 2 weeks. Review message logs weekly with someone from compliance. Watch for data scope creep and access issues, and keep a running list of defects with owners and dates. Share each summary with the sites.

After the pilot: decide. Compare results with the baseline and the comparison group, then price the full rollout with the pilot’s actual effort, not the vendor’s estimate. If the days of delay avoided exceed the break-even in your model, roll out a study at a time. If they don’t, keep the pieces that helped, such as survey tracking, and drop the rest. Record the decision and its reasons so the next team doesn’t repeat the work.

FAQs

What does Salesforce offer for clinical trial recruitment?

Salesforce’s Trailhead unit describes an agent that filters and summarizes sites and investigators, sends and tracks feasibility surveys, matches patient records to inclusion and exclusion criteria, and forecasts recruitment readiness. Its pharma page promises matching and personalized messages that prevent attrition. None of these sources reports measured results. Ask for validation data on a trial like yours, and run your own pilot with a baseline before you scale. Treat the demo as a starting point for diligence.

How accurate is the patient matching?

I found no accuracy figure for Salesforce’s matching. The Trailhead unit refers to high precision without data. The best independent evidence concerns a different system: an NIH-funded paper reported 87.3% criterion-level accuracy against physician annotation on public synthetic cohorts, with a 42.6% screening time reduction in a small pilot. Those results don’t transfer automatically, so score a sample of your own matches with coordinators who can’t see the tool’s answers.

Can a sponsor use patient records to contact prospects?

It depends on the legal basis, and counsel should decide. HHS says the preparatory-to-research provision lets a covered entity’s own workforce use protected health information to identify and contact prospects, but an outside researcher can’t use it for contact. An IRB or Privacy Board may grant a partial waiver for recruitment contact. A sponsor-side platform that holds identifiable patient data therefore needs a specific design and approval before launch. Confirm the design with counsel before any data moves.

Do recruitment messages need ethics review?

FDA’s information sheet says the IRB should review the methods and materials investigators propose to recruit subjects, including advertisements and receptionist scripts that collect personal information. It also says ads shouldn’t emphasize payment or claim that a product is safe or effective. Personalized or generated messages raise the same questions, so lock templates, require approvals and log every send. Have your ethics committee confirm its expectations before outreach begins. Keep copies of every approved version.

Does the platform improve retention?

The evidence I found is a vendor promise and a customer founder’s statement that dropouts fell, as paraphrased by R&D World. No figures, baseline or comparison group accompany either claim, and R&D World doesn’t say whether any claim was independently verified. A fair test compares retention at pilot sites with comparison sites on the same protocol, with ethics-approved messages and a record of site staff burden. Judge the result on your own data.

How much does it cost?

Salesforce’s pricing page lists Life Sciences Cloud editions from $350 per user per month for Core to $725 for Sales & Service Max, billed annually, plus a Participant Recruitment and Enrollment add-on at $10,000 per org per year in increments of 100 patients referred. An undated third-party listing says $325, so the figures conflict. In my invented model, first-year cost reached $440,000 to $630,000 for 30 users, depending on how the add-on is read. Get a written quote that spells out agent usage, credits and integration.

Is it covered under our regulated-data agreement?

I can’t confirm. I didn’t read Salesforce’s business associate addendum for this post, and none of the 9 pages I read says whether Life Sciences Cloud or its Agentforce add-on is covered. Ask Salesforce to state in writing which services and features are covered for your use, and have counsel read your agreement. Coverage may also depend on the data flows you design, so ask about each feed, each feature and each edition you plan to buy.

What should a pilot measure?

Measure coordinator time per screened patient, match accuracy against blind manual review, rate of Insufficient Data results, time from referral to consent, enrollment rate per site, dropout and site staff burden. Agree on thresholds before the pilot starts, include comparison sites, and have compliance review all participant-facing content. End on a fixed date, then decide whether to scale, adjust or stop based on your numbers rather than the demo. Share the results with sites as well.

By

Leave a Reply

Your email address will not be published. Required fields are marked *