How CovioIQ's Scoring and AI Work
Insurance is a regulated, judgment-driven industry. Here is how CovioIQ scores leads and renewals, where that happens, and what our AI assistant features send to an AI provider.
Our Principles
Scores are advisory. Brokers make every client-facing decision.
We never auto-contact your end clients without your explicit approval.
Your database is hosted in Canada, and lead and churn-risk scores are calculated there. The optional AI features (Alex, email categorization and the website chatbot) are processed by Anthropic in the United States, and send only what each answer needs.
Every score comes with the reasons behind it, so you can see why a lead or renewal was flagged.
You can recalculate or ignore any score at any time.
By Feature
Each scoring and AI feature in CovioIQ, what it does, and what it doesn't.
Lead Scoring
Purpose
Surface which inbound leads to call first.
Output
1–99 score with the reasons behind it and a suggested next step. Calculated with fixed rules, not an AI model.
Signals Used
- •Lead source (referral, partner broker or existing client score higher; cold outreach scores lower)
- •Activities logged in the last 30 days
- •Contact details on file (email and phone)
- •Status (lead / prospect / client / inactive)
- •Number of policies the contact holds
- •No client names, notes or email content are used
Broker Control
Advisory only. The broker decides who to call. The score is recalculated when you ask for it; it does not update on its own.
Churn-Risk Score
Purpose
Flag at-risk renewals before they lapse.
Output
0–100 risk score with the factors behind it and suggested next steps. Calculated with fixed rules, not an AI model.
Signals Used
- •Premium increase at renewal (larger increases add more risk)
- •Client tenure (clients under 12 months add risk)
- •Number of policies held (one policy adds risk; three or more reduces it)
- •Time since a call, email, meeting or SMS was last logged (longer gaps add risk; contact inside 30 days reduces it)
- •Starts from a baseline of 25 and stays between 5 and 98
- •Recalculated daily, and whenever you refresh a renewal
- •No client names, notes or email content are used
Broker Control
Scores of 70 or more create an in-app insight for you to review. The score is an estimate, not a prediction, and broker judgment is required. We never contact end clients on your behalf.
Email Categorization
Purpose
Suggest a category, priority and short summary for an email logged in CovioIQ.
Output
A suggested category (for example quote request, claim or renewal question), a priority and a one-to-two sentence summary. Generated by an AI model (Anthropic).
Signals Used
- •Runs only when you click AI Categorize on one email
- •Sends that email's subject and body to Anthropic, processed in the United States; nothing else from your book
- •The Emails section itself is still in development, so this is not reachable in the app yet
Broker Control
A suggestion only. You can change or ignore the category, and nothing is sent or filed automatically.
Alex Assistant
Purpose
Answer plain-language questions about your book of business. Team and Agency plans.
Output
A written answer generated by an AI model (Anthropic), built from your own CovioIQ records. It can be wrong; check anything you act on.
Signals Used
- •Sent with each question: client names, contact status and lead score; policy number, type, carrier, annual premium, status and expiry; open renewals with churn score and reasons; open opportunities; open tasks
- •Not sent: email addresses, phone numbers, street addresses, dates of birth, notes or documents
- •Processed by Anthropic in the United States. Anthropic does not use API data to train its models
- •Alex cannot send emails, change records or contact clients. Conversations are stored in CovioIQ in Canada
Broker Control
Advisory only, and it never gives coverage advice. You decide what to do with an answer, and you can stop using Alex at any time.
Where Processing Happens
Scoring & Region
Lead and churn-risk scores are calculated with fixed rules in CovioIQ's Supabase database and edge functions in ca-central-1 (Montréal, Canada), with no AI model involved. We do not host or fine-tune AI models.
AI Model Provider
Alex, email categorization and the public website chatbot use Anthropic's API (Claude models), processed in the United States. Each request carries only what that feature needs, as listed above. Anthropic is listed on our Subprocessors page.
Logging
Scoring functions log errors and request metadata such as timestamps. They do not log client names, notes or email content. Alex saves your conversation in CovioIQ so you can pick it up again, and keeps an audit record of each request (time, page and message length, not the text).
Rate Limits & Failure Modes
Scoring requests are rate-limited per user and return a 429 with a message when the limit is reached. If a score can't be calculated, you see an error and no score is saved.
About Accuracy Claims
CovioIQ is a pre-launch / beta product. The lead and churn-risk rules are simple and fully described above, and we have not yet measured how well they predict outcomes.
Treat any percentage in marketing copy as a directional target, not a measured outcome. Beta brokers can question any score at any time by emailing admin@covioiq.com.
If we publish accuracy figures, the method and data behind them will be linked from this page.
Questions about how CovioIQ handles your data?
Email our team. We're a small Canadian company — a real human will reply.