Bond Health
Bond Health runs clinical trial recruitment from chart screening and advertising through to booked visits and consent support, for research sites, site networks, health systems, sponsors and CROs. Identify reads clinical notes and structured records against a study protocol and gives coordinators a ranked list of candidates with the evidence for each criterion; Bond says it reads more than 10,000 charts an hour and cuts coordinator chart review by half or more. Engage reaches patients by voice and text in English, Spanish, Mandarin and other languages, switching mid call when needed, and Bond also builds and runs Meta and Google ad campaigns for each study. Book puts visits on the site's calendar with reminders, and Consent helps patients prepare for the consent discussion. The investigator decides eligibility and the site obtains consent.
Matching combines language models with ClinTextGraph, a medical terminology graph spanning 18 vocabularies, in separate relevance, criterion and ranking stages. Identify connects to Epic, Oracle Health, MEDITECH, athenahealth and other EHRs, with full integration in about 48 hours, and a two way CRIO link sits alongside Advarra, RealTime and Veeva SiteVault connections.
Bond, based in Boston, cites one site that randomized 193 patients in two weeks. It charges a fee per screened patient plus a share of each randomized patient's milestone payments, with ad spend separate.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
Chart interpretation and patient conversations are the AI modules inside a broader recruitment service. Identify reads narrative records against eligibility criteria, and Engage holds voice and text conversations that gather preliminary screening answers and book visits. Without the models those functions go, while campaign management, ad buying, appointment coordination, dashboards and staff support remain.
Deterministic rules rank candidates after the language model assesses them, so a patient's place on a list reflects both the chart evidence the model read and the site's configured requirements, not a free model decision about enrollment.
Patient contact runs within local calling windows and a site set attempt limit, such as three or five tries. An opt out ends outreach, and patients can ask for a coordinator through a live transfer or callback. Questions outside the approved script, including care, risk or whether to join a study, go to staff.
Identify supplies chart evidence without deciding eligibility, which stays with the investigator and coordinator, and the site obtains consent. Claims that every lead gets follow up describe workflow coverage within those limits, and script approval, stop conditions and escalation to people bound both screening and outreach.
ClinTextGraph grounds a three stage matching process: candidate relevance, criterion by criterion assessment, and deterministic ranking with eligibility gates. Bond's technical summary puts the graph at 3,270,078 nodes across 18 vocabularies, including SNOMED CT, RxNorm, LOINC and ICD-10-CM, while its homepage cites 1.6 million. Evidence packets tie concepts, synonyms, code mappings, relationships and provenance back to the chart.
That is a specific retrieval and reasoning design. The foundation models, their versions and how they are replaced are not named, and the August 2026 internal report behind the benchmark is not available to reproduce the pipeline.
Language models read chart text and carry patient conversations, but the model providers are not named. Agents switch languages mid call, which adds speech recognition and voice synthesis steps whose providers also go unnamed. Startup program ties to cloud and voice companies, including Google Cloud, AWS, NVIDIA and ElevenLabs grants, do not show which companies process a given patient's data.
The privacy terms describe infrastructure, telephony and monitoring providers bound against independent use without listing the inference chain, and website analytics and ad tools play separate roles. That leaves open who receives clinical prompts, conversation content and generated assessments, and what retention and training terms each applies.
An unnamed research site in the Dallas and Fort Worth area reportedly randomized 193 healthy adults into a Phase 3 vaccine trial in two weeks, twice its own concurrent recruitment result. The account gives no site name, dates, outreach counts or allocation method, so the comparison cannot show cause.
A benchmark summary reports a micro F1 of 0.9312 on held out n2c2 2018 data, and the homepage claims 90% or better matching accuracy, but the internal preprint behind them is not available. RECTIFIER and TrialGPT results describe other systems. None of this validates Bond's current recruitment performance across sites or protocols.
SMS histories and consent records can be kept for up to seven years. Account, patient and trial information lasts through the account or trial plus any research or legal obligation, and rolling logs and backups have no stated expiry. Later deletion or anonymization does not say whether clinical content trains Bond's models.
Service providers are barred from independent use, and mobile consent data is barred from outside marketing. Optional ad tags on selected survey links exclude answers and contact fields but send page events and browser details. Model level retention, training and handling of a wrong generated assessment remain undefined.
Bond acts as a business associate under an operational privacy policy that covers patient contact details, health information, workspace activity and consent records, and it signs a business associate agreement with each participating site or health system before handling its protected information. The security material describes subcontractor duties and return or destruction of PHI where feasible, and the public trust center maps 73 HIPAA Security Rule controls.
The full agreement is shared during contracting, so its scope, exceptions and fit with each module cannot be read in advance. Site recruitment approvals and IRB decisions stay separate from the vendor's role.
Bond holds a SOC 2 Type I report and ISO/IEC 27001:2022 certification for its own security program, with a SOC 2 Type II audit underway. The ISO certificate, ISMS-BO-101926, was issued by Advantage Partners on October 19, 2026 and runs to October 19, 2029 subject to surveillance, covering web applications, APIs, cloud integration and supporting operations.
The Vanta portal shows control status for encryption, disposal, continuity and retention, along with role based access, SSO, audit logging and penetration testing. The certificate and audit report themselves are not posted there for inspection.
Recruitment and preliminary screening are the job, with eligibility decided by the study team and consent obtained by the site, and the company states plainly that its software is not FDA cleared or IRB approved. Identify ranks possible matches rather than diagnosing disease or choosing treatment, and Engage routes questions about medical care and study risk to a coordinator, limits that fit trial logistics.
No specific statutory device exclusion is cited for the whole offering. References to FDA guidance on recruitment or electronic systems describe the research setting rather than authorizing this software.
Protocol setup and script review are concrete steps before any patient is contacted. Sites and their IRBs approve the wording, translations are reviewed, and test calls check opt outs and handoffs before launch, with script changes going back through approval. When a protocol amendment touches criteria, configuration is redone and coordinators review a sample of decisions.
Offering several languages does not show recognition or matching works equally well in each, and no results separate languages, demographic groups or patients with incomplete records. The terminology benchmark measures matching mechanics, not fairness across the populations a study wants to enroll.
Criterion to chart traceability lets coordinators challenge a proposed match before anyone is enrolled. Patients can stop contact or reach a person, and the privacy policy accepts direct correction requests with a 30 day response, which covers personal information rather than a wrong eligibility assessment.
The general service terms disclaim accuracy and completeness, cap total liability at the greater of 12 months of fees or $100, and point the express indemnity toward Bond, though a customer agreement can override them. No standard duty to compensate, correct clinical conclusions or remedy harm from a wrong screening output exists.
Identify reads from Epic, Oracle Health, MEDITECH, athenahealth, eClinicalWorks, NextGen, Veradigm, OncoEMR and other major EHRs, through FHIR R4, HL7 or a partner connection, so the list does not mean an identical native connector for each, and full integration takes about 48 hours.
As a CRIO Certified Partner, Bond sends prescreened patients into CRIO and gets recruitment status back, and it also lists Advarra OnCore, Advarra Clinical Conductor, RealTime and Veeva SiteVault. That two way trial system exchange is separate from proven production read and write integration with a major EHR.
The service runs in a hosted recruitment workspace with role based access, SSO and logs of access and actions. EHR data reaches that environment for screening, and voice, text and optional survey measurement add further processing paths.
The general privacy terms allow processing in the United States or wherever providers operate, rather than tying the whole clinical workflow to one region. No customer managed deployment or full account of tenant isolation is given, and a site's control of its own EHR and ad account does not settle where model inference, call content, backups or support access sit.
Bond charges a volume fee for each patient called or texted through prescreening plus a percentage of the milestone payments each randomized patient generates, from randomization through later study visits. The screening fee applies even when a patient does not enroll, so the model is not payment for success alone.
EHR integration, configuration, ongoing support and all modules are included with no integration fee, and ad spend stays with the site. Neither the per patient amount nor the percentage is listed, and a pilot on an existing patient list needs a quote rather than a self serve purchase.
Independent research sites, site networks, health systems, sponsors and CROs are the buyers, with recruitment running from chart screening to outreach and visit scheduling. Named partner sites include Southwest Family Medicine Associates, Commonwealth Pain and Spine Clinic and MRCT at Brigham and Women's Hospital, and a coordinator at an academic medical center describes use on an oncology trial.
The detailed enrollment example covers healthy adults in a Phase 3 vaccine study. Protocol setup allows many eligibility definitions, but support for a specialty is not validated performance within it, and broader therapeutic and multilingual results are not reported.
Alternatives to Bond Health
Other products that do the same kind of work as Bond Health, and what each one does.
Alleviate Health provides conversational recruitment software for clinical research sites. Its agents contact potential participants, ask initial eligibility questions, and arrange an on-site visit or a call with a clinical research coordinator.
HealthKey helps research practices find potential clinical trial participants in their existing patient records. Teams supply a protocol's inclusion and exclusion criteria, and the software applies those criteria across structured EHR fields, clinical notes, and documents such as faxed PDFs.
Clinical trial platform that embeds patient identification, eSource and data capture inside provider organizations so trials can run in routine care, paired with its own recruitment coordinators and eligibility specialists.
Echo Booking is an AI receptionist for radiology groups, medical practices and dental offices, built by Echo Health Solutions.
CluePoints provides clinical trial data quality and oversight software for pharmaceutical sponsors and contract research organizations.
ConcertAI's Accelerated Clinical Trials platform helps pharmaceutical sponsors and contract research organizations design, plan and run oncology studies.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Custom quote
|
A fee per patient screened through calls or texts combines with a percentage of each randomized patient's milestone payments, including later study visits. The screening charge applies even when the patient does not enroll. | A business associate agreement accompanies each site or health system engagement; full terms are shared during contracting. | EHR integration, configuration, and ongoing support are included without a separate integration fee. Advertising spend is separate. | Vendor Published |
Per patient rates and milestone percentages are quoted. Identify, outreach, booking, consent support and the dashboard are included. A pilot on an existing patient list can begin before an EHR connection, but no free pilot price or enrollment guarantee is stated.