Birth Model

The predictions are the differentiator and there is a real non predictive product around them. Delivery timing forecasting and obstetric risk stratification have no conventional substitute, they are the reason the platform exists, and the company holds granted patents on the method. Against that, a meaningful part of what a unit buys is the consolidated view itself: a single source of truth for the labour and delivery floor, populated automatically from EHR data, which would retain operational value as pure aggregation even without a model behind it. Graded B rather than A on that basis, and the split matters commercially because a buyer should establish how much of the reported benefit comes from better prediction and how much from simply having the unit's state visible in one place for the first time.

The stated posture is advisory and the company frames it deliberately, describing its approach as balancing AI with clinical judgment and reducing alarm fatigue rather than adding alerts, which is a considered position in a specialty where alert burden is a real safety issue. Clinicians decide; predictions inform. What is not published is any performance boundary: no accuracy figure for delivery timing prediction, no discrimination statistic for the risk models, no threshold at which a risk flag is raised, and no description of what the system does when confidence is low. The failure mode specific to this product deserves naming, and it is not a clinical alert failure. Delivery timing predictions drive STAFFING decisions, so a systematically optimistic forecast produces a unit staffed for fewer simultaneous deliveries than actually arrive, and the harm surfaces as thin coverage during a complication rather than as a visibly wrong prediction.

Better disclosed than most small vendors because of an unusual route: the method is covered by granted United States patents, 11,664,100 and 12,133,741, and the company publishes the numbers with links under a virtual patent marking notice rather than referring vaguely to proprietary technology. Granted patents are public documents, so an evaluator can read the claimed approach, which is a genuine transparency artifact and the same property this index credits in Unlearn. Held at B rather than A on two counts. A patent describes a claimed method, not necessarily the deployed model, and the two can diverge substantially over time. And no performance disclosure accompanies it: no accuracy or error distribution for delivery timing, no area under the curve or calibration for the postpartum haemorrhage, preeclampsia and shoulder dystocia risk models, and no model card.

Credible institutional association, no published results. The backing is real and checkable: Johns Hopkins, Techstars through its Baltimore AI Health programme, and CareFirst BlueCross BlueShield, the last being notable because a payer investing in a product that claims to reduce primary caesarean rates is an alignment worth recording. The advisory board is named with verifiable credentials and includes senior obstetric and health technology figures. What is missing is the evidence itself. No hospital customer is named anywhere public, no study is published, and the outcome claims made by the founder, covering staffing cost savings, increased labour and delivery revenue, reduced burnout, improved patient satisfaction and lowered primary caesarean rates, carry no site, denominator, baseline period or methodology. Lowering primary caesarean rate is a serious and measurable clinical claim that would be straightforward to evidence at a single named site, and it is the one a buyer should ask to see substantiated first. Graded C: association with respected institutions is not the same as evidence of effect.

A real and verifiable disclosure surface rather than an assertion. The company runs a public trust centre hosted on Vanta, linked directly from the site footer and openable without contacting sales, and maintains a distinct privacy policy and terms of service. That is materially more than most vendors of this size offer and it gives a buyer somewhere concrete to start. What was not located in public materials: any specific commitment on whether customer clinical data is used to train or improve models, what retention applies, and how obstetric data, which is among the most sensitive categories a health system holds, is segregated. Held at B pending those specifics.

HIPAA compliance is asserted directly in the site footer alongside the SOC 2 mark. No business associate agreement is published and no terms are described, placing Birth Model on the middle rung of this index's HIPAA ladder where compliance is claimed and the instrument is unpublished. The trust centre may hold BAA information behind its request flow; nothing was readable from the public page.

The best security disclosure on this sourcing list, and the smallest company on it, which is itself the finding. Birth Model states SOC 2 TYPE II specifically rather than the unqualified SOC 2 that this index flags wherever it appears, and it runs a live public trust centre on Vanta linked from the footer and openable without a sales conversation. Naming the type answers the question this index asks of every SOC 2 claim, since Type I describes controls at a point in time while Type II tests their operation over a period, and the trust centre gives a buyer a self serve route to the underlying documentation. HIPAA compliance is asserted alongside it. Graded A on the combination of correct specificity and a verifiable artifact. Worth stating plainly for calibration: several substantially larger and better funded vendors assessed alongside this one publish no attestation at all, and a seed stage company with a handful of named staff is out disclosing them.

No clearance, no submission and no published regulatory position. The two halves of the product sit differently against the boundary and a buyer should hold them apart. Delivery timing prediction driving staffing is an operational forecast and raises no device question. Risk stratification for postpartum haemorrhage, preeclampsia and shoulder dystocia is patient specific clinical risk prediction, which sits much closer to the software functions the 21st Century Cures Act exclusions were written to delimit, and whether it falls inside turns on whether the clinician can independently review the basis for the risk score. Nothing published states the basis, which weakens that argument rather than strengthening it. Ask for the documented analysis for the clinical modules specifically.

No subgroup or fairness performance data was published, and there is no clinical domain in United States medicine where that absence matters more. Maternal mortality carries among the starkest documented racial disparities in American healthcare, with Black women dying from pregnancy related causes at roughly three times the rate of white women, and those disparities are driven substantially by differential recognition and response to warning signs rather than by differential biology. A risk model trained on historical obstetric data learns from a record in which some patients' deteriorations were recognised late, and can encode that as lower predicted risk. The mechanism here is concrete rather than theoretical: predictions inform both clinical attention and staffing allocation, so a systematically low risk estimate for a subgroup translates into less attention for exactly the patients already receiving too little. The company's own materials reference addressing disparities, which indicates awareness, but awareness is not measurement and no calibration or performance breakdown by race, ethnicity, age, parity or payer was located. Publishing it would be the single highest value disclosure this vendor could make.

Genuinely deep in one system and undocumented outside it. Birth Model is listed in the Epic Showroom, describes automatic population from EHR data, and employs an Epic certified clinical systems lead holding EpicCare and STORK credentials, Stork being Epic's obstetrics module. That last detail is more meaningful than a generic integration claim, because obstetric data lives in a specialty module with its own flowsheets and documentation conventions and a vendor with certified expertise in it is not integrating superficially. The company has separately described achieving zero click Epic integration through SMART on FHIR. Mobile applications on both stores extend reach to clinicians away from a workstation. Held at B because Epic is the only EHR named anywhere; a labour and delivery unit on Oracle Health, MEDITECH or another platform should establish availability before evaluating anything else.

Cloud delivered through a provider web platform with companion iOS and Android applications. No hosting provider, region, tenancy model or data residency commitment is published on the public site, though the Vanta trust centre may carry some of this detail behind its request flow. Mobile access to obstetric risk information raises device management and authentication questions that no public material addresses.

No price, tier or unit is published and every commercial path routes through a scheduled demo. The company does publish a return on investment calculator, which is more structure than most vendors of this size provide, but it shares the asymmetry this index has now recorded across several records in this category: the calculator models the benefit side while the cost side stays behind the demo, leaving a buyer estimating half a ratio. The unit question is worth settling early, since per bed, per delivery, per unit and per health system produce very different economics for a service line whose volume is seasonal and demographically driven rather than discretionary.

The narrowest scope on this sourcing list and deliberately so: a single inpatient service line, labour and delivery, in a single country, on a single electronic health record. Within that scope the product covers both operational and clinical workflows and reaches clinicians on mobile as well as at the workstation. The limiting factor for this grade is not the narrowness, which is a coherent strategy, but the absence of any named deployment: no hospital, health system or birth centre is identified publicly, so there is no evidence of how the product performs across different unit sizes, acuity mixes or staffing models. A high volume tertiary centre and a rural unit delivering a few hundred babies a year are very different environments for a delivery timing model, and nothing indicates which have been served.