For insurers, employers & health platforms
Give every member a health assistant they actually trust.
Embed Iris into your product, WhatsApp or email. Lower claims, deepen engagement, and sell more of what you offer — without the risk or cost of building your own AI.
Why raw AI fails for health
ChatGPT, Claude and Gemini are powerful — but for health decisions they're a liability.
You see the opportunity in AI — but you also see the risk. Iris lets you move fast on health intelligence without the liability of raw consumer tools.
Over-confidence
The problem: ChatGPT and Claude present one possibility as if it's the answer. For health, certainty is dangerous.
Iris: Each response is reviewed by an independent AI system before it reaches the user. Unsupported claims and inconsistencies are caught and resolved at source.
Amnesia
The problem: Close the tab and everything is gone. Every conversation starts from zero — no continuity, no pattern detection.
Iris: Iris keeps a structured, editable memory of each member's full history. Context builds over weeks and months.
No structure
The problem: "I felt bad after lunch" isn't data. Raw AI can only guess without structured symptom, food and sleep logs.
Iris: Iris captures symptoms, food, sleep and labs as structured data — so AI can actually reason about it.
Unaccountable
The problem: Consumer AI tools aren't built for regulated environments. No audit trail, no explainability, no governance.
Iris: Every answer shows its reasoning and its reviewer. Auditable, explainable, safe for health platforms.
Safety & governance
Built to be safe, not just smart.
Partners need to know an AI health assistant won't overstep, hallucinate or leave them exposed. Iris is designed with clinical-grade guardrails from the ground up.
Stays in its lane
Iris is designed for information, guidance and monitoring — not diagnosis or prescription. Clear guardrails keep every response within safe boundaries, and members know when to consult a clinician.
Escalates when it should
Built-in triage logic flags symptoms and situations that need human review. Iris routes members to appropriate care rather than guessing at outcomes it can't assess.
Catches its own mistakes
Every answer is reviewed by a second independent model before it reaches the user. Over-confident or inconsistent responses are surfaced and resolved automatically — not after the fact.
Full audit trail
Every conversation, reasoning step and model decision is logged and reviewable. For compliance, quality assurance and incident investigation — you have complete visibility.
Partner benefits
Built for insurers, employers and health platforms.
Iris drops into the channels your members already use — lowering cost, strengthening engagement, and maintaining the safety standards regulated environments require.
Fewer unnecessary claims
Handle everyday health questions and early symptoms before they turn into avoidable GP visits, urgent care calls and specialist claims.
Operational efficiency
Deflect low-acuity queries away from call centres and clinicians. Free your team up for the cases that actually need them.
Cost down, quality up
Independent review on every answer means you cut cost without cutting corners on clinical quality or safety.
Sell more of what you offer
Turn insights into timely, relevant nudges toward blood tests, screenings, supplements or programs you already sell.
Ongoing relationship, not one-off
Members and customers open Iris every week, not once a year — deepening engagement and lifetime value.
Meets members where they are
Deploys inside WhatsApp, email or your existing app. Members adopt instantly and keep coming back — no barriers, no learning curve.
Why not build it yourself?
Most teams need 12–18 months to build something comparable. Iris is already live.
The eight components below are what a team would have to design, ship and stress-test in production before putting AI in front of members. Each one drops into your existing product as a capability layer.
Persistent, verifiable memory
A user-owned memory layer with version history, source citations, and single-source-of-truth consolidation. Every fact Iris knows is visible and editable — the retention infrastructure every platform is now trying to build.
Structured data from plain conversation
Users just talk — text, voice, meal photos, lab pictures, wearables. Iris turns it into schema-validated entries across 23 categories (meals, symptoms, sleep, meds…), so analysis runs against structured data, not inferred from free text.
Cross-model validation on every message
One provider writes the answer, a different one reviews it. Discrepancies and over-confidence are surfaced and resolved before anything reaches the user. No major consumer product ships this today — it's the audit trail regulated buyers need.
Autonomous follow-up and monitoring
Users define objectives — weekly sleep trend reviews, pre-appointment briefings, medication adherence checks. Iris schedules, runs and reports back — sustained engagement without notification fatigue, because the user set the goal.
Best model for each task, no lock-in
Anthropic for conversation, Google for structured extraction, OpenAI as backup. Composable with whatever AI vendor relationships you already have.
Cents per conversation
Real production costs on high-value conversations — appointment prep, multi-week pattern analysis — run $0.06–$0.16 in model spend. Adding this to an existing user base costs cents per active user.
Channel-native, no app required
Full functionality over WhatsApp — voice notes, meal photos, everything — plus web and push. One production family runs their child's entire multi-condition care log on WhatsApp alone.
Operational maturity from live deployment
Daily scored conversation reviews, weekly synthesis into prompt changes, a documented failure-mode taxonomy. This is the part you can't rebuild from a spec — and it's the difference between an AI demo and something safe to put in front of members.