How Coverage Companion actually works
We believe you deserve to understand the tools you use. This page explains every component of Coverage Companion — from the AI chat to the cost simulation model to the scoring formula and the behavioral insights.
The Companion Chat
The chat assistant is powered by Claude by Anthropic, a large language model. It's designed for educational conversations about health insurance concepts — not to give medical, legal, or financial advice.
When you chat, Coverage Companion also runs preference extraction in the background. It pattern-matches your natural language for signals about your risk tolerance, expected usage, and specific care needs — then surfaces those as profile chips you can apply to the Confidence Engine.
Detected signals → risk: averse | balanced | seeking
usage: low | medium | high
budget: premium | balanced | oop
needs: [mental_health, prescriptions, specialist, ...]Constraints: The chat provides educational guidance only. It will not recommend a specific plan, diagnose conditions, or advise on medical treatment. Always verify coverage details with your insurer or HR.
The OOP Simulation Model
The engine simulates your estimated annual out-of-pocket costs under three utilization scenarios — low, medium, and high — using benchmark service quantities and nationally representative allowed amounts.
For each service in each scenario: deductible_applied = min(service_cost, remaining_deductible) coinsurance_owed = (service_cost − deductible_applied) × coinsurance% copay_owed = min(coinsurance_owed, copay × quantity) total = premium + Σ copay_owed, capped at OOP max
Utilization benchmarks are drawn from MEPS (Medical Expenditure Panel Survey) data and CMS average allowed amounts by service type:
The Scoring Model
Each plan receives a Confidence Score (0–100) composed of four sub-scores, each weighted by your profile. The weights shift based on your budget priority and risk tolerance.
| Score | Full name | Formula | What it captures |
|---|---|---|---|
| CES | Cost Efficiency Score | 100 × (worst − plan) / (worst − best) | How cost-efficient this plan is relative to the others you entered |
| RPS | Risk Protection Score | 100 × (1 − OOP_max / $9,450) | How much financial downside protection the plan offers |
| PS | Predictability Score | 100 × (1 − std_dev / $4,000) | How consistent costs are across the three usage scenarios |
| NMS | Needs Match Score | Σ service_points / max_points × 100 | How well the plan's copay/coverage structure fits your selected care needs |
Final score = w_CES × CES + w_RPS × RPS + w_PS × PS + w_NMS × NMS
Default weights (balanced profile):
CES: 0.35 RPS: 0.30 PS: 0.20 NMS: 0.15
Weights shift when budget = "low premium" → CES ↑
or risk = "averse" → RPS ↑, PS ↑Behavioral Insights
Health insurance decisions are vulnerable to predictable cognitive biases. The engine detects four common patterns and surfaces a nudge when it spots one:
| Bias | Detection | Nudge approach |
|---|---|---|
| Loss aversion | Low-risk profile + high OOP delta between plans | Quantify the worst-case scenario explicitly |
| Simplicity preference | User picks plan with fewest fields or lowest copay | Show total annual cost including premium, not just copay |
| Overconfidence in health | Low usage + low risk + HDHP scores best | Remind that medium-usage scenario is statistically most likely |
| Anchoring on premium | Budget = "low premium" but OOP delta is large | Show premium savings vs. expected OOP increase side by side |
⚠️ Educational simulation — not financial advice
Coverage Companion is an educational tool. The cost simulations use benchmark utilization data — not your actual claims history. Scores reflect relative plan performance under assumed scenarios and should not be the sole basis for a coverage decision. Actual costs will differ based on your providers, claims history, and plan-specific rules. Always verify details with your insurer or a licensed benefits advisor.