DarioHealth (DRIO) Trains Clinical AI Only on Consumer Data

On August 11, 2026, DarioHealth (DRIO) used the closing question of its FY2026 Q2 earnings call to explain that it trains its clinical AI models only on data from its direct-to-consumer business. Member data on the employer and health-plan side of the company cannot be used that way because of compliance rules [1]. No other vendor in the sector has stated the rule this plainly.

---

Member data belongs to the customer who pays, not to the vendor

US digital health companies generally sell in two ways. One is to sell chronic-condition management to employers or insurers, who pay for it while their employees or members use it — the industry calls this B2B2C. The other is to sell directly to individuals, who subscribe with their own money, which is B2C. The difference is not only who pays: in the first model the members are the customer's employees or plan members, and the health data they generate belongs legally to the customer's side.

The constraint therefore lands on whether that data can be used to build products. HIPAA and state health-information laws limit what customers may hand over to a vendor, and DarioHealth's own annual report carries this as a risk factor, stating that these laws restrict the ability of its customers and research collaborators to share health information with it [2]. What a vendor is free to use for model training is whatever first-party data sits on its own B2C side. The limit applies to every comparable company at the same time and has nothing to do with technical skill, so the vendor with the largest contracted membership is not necessarily the one with the most trainable data.

---

The same dataset was split in two between quarters

The explanation came in response to a question about data provenance. An analyst asked how much of the 13 billion data points behind Dario IQ, the company's AI product, is first-party data from its own devices versus third-party data from insurers and employers. Chief Executive Erez Raphael answered that the 13 billion spans both the B2C and the B2B side, but that R&D work such as model training is done purely on B2C, for compliance reasons [1]. The same materials list this as a risk: compliance rules restrict the use of B2B customer data for AI model training, creating a dependency on B2C first-party data, and AI innovation could slow if B2C data growth comes in below expectations [1].

Three months earlier the same dataset was presented as a single asset. On the May 13, 2026 call, management described FDA-cleared devices it designs itself generating data, that data flowing into its platform, and AI running on top — the whole chain owned in-house, with no B2C-versus-B2B distinction and no mention of what is trainable [3]. The commercial framing arrived this quarter too: the company estimates Dario IQ could add roughly 10% to 15% in recurring revenue from existing customers over time, against quarterly revenue of $5.2 million and a 62% gross margin [1].

Two enterprise-only peers discussed AI on calls in the same stretch, and they described something else. Hinge Health (HNGE) reported Q2 revenue of $213 million and an 87% gross margin, about four points above the 83% of a year earlier, attributing the improvement to care-team efficiency gains from AI and automation [4]. Omada Health (OMDA) said cost to serve per member fell more than 10% year over year, with AI applied to coach tooling and prediction of member demand [5]. Both are far larger than DarioHealth, and neither disclosure describes training a proprietary clinical model on member outcome data.

---

Trainable data becomes a separate asset, decoupled from enterprise scale

If DarioHealth's account of the rule holds, the basis for judging a digital health company's AI capability changes. Contracted members, contract value and renewals all sit on the enterprise side; whether a vendor can build its own clinical model depends instead on the size and growth of first-party data on the B2C side, and the two need not move together. The company with the largest enterprise book may not hold the most data it can legally train on.

The boundary is equally clear. This is one company's description of an industry rule, no peer has said the same thing, and not saying it is not the same as not facing it — Progyny (PGNY) did not raise the subject at all on its call in the same period [6]. DarioHealth presents the rule as cutting both ways: the advantage comes from also running a B2C business, and the risk comes from depending on B2C data growth [1]. Two things are worth tracking: whether the company begins disclosing growth in B2C members or data points, and whether enterprise-side peers start explaining what data their own model training runs on.

---

Companies exposed to this change:

• LifeMD (LFMD): Its core business is direct-to-consumer telehealth, and it is also expanding insurance coverage and multi-condition programs [7]. Under the rule DarioHealth describes, data from cash-pay business belongs to the company itself, while ownership shifts once members move to an insurance channel. LifeMD has disclosed nothing on this point.
• Teladoc Health (TDOC): The largest company in the sector, it is moving its previously cash-pay BetterHelp behavioral health business into insurance networks this quarter [8]. When a visit shifts from cash-pay to insured, both the payer and the data ownership change, though the company has not connected this to AI training.
• Progyny (PGNY): A fertility benefits manager whose customers are almost entirely employers, which places it on the most constrained side of this logic [6]. It has not discussed data training itself, so the connection rests only on its sales channel structure.

---

Sources

[1] Drillr · DarioHealth (DRIO) · 2026-08-11 · FY2026 Q2 earnings call

There are a lot of elements that are related to compliance on how the data can be utilized. And one of the big advantages that Dario have is that we are operating the entire B2C business. And when we are talking about data, the 13 billion is something that is between B2C and the B2B. But for most of what we do on the R&D side, training models and so on, we are doing it purely on the B2C because of compliance aspects.

[2] Drillr · DarioHealth (DRIO) · 2025-03-10 · FY2024 Form 10-K, risk factors

[3] Drillr · DarioHealth (DRIO) · 2026-05-13 · FY2026 Q1 earnings call

[4] Drillr · Hinge Health (HNGE) · 2026-08-04 · FY2026 Q2 earnings call

[5] Drillr · Omada Health (OMDA) · 2026-08-06 · FY2026 Q2 earnings call

[6] Drillr · Progyny (PGNY) · 2026-08-06 · FY2026 Q2 earnings call

[7] Drillr · LifeMD (LFMD) · 2026-08-05 · FY2026 Q2 earnings call

[8] Drillr · Teladoc Health (TDOC) · 2026-07-29 · FY2026 Q2 earnings call

This is here to surface industry changes and companies that may be overlooked - not a stock recommendation.