SECURITY
You're trusting us with your book and your carrier relationships. This page says exactly what we touch, where it goes, who can see it, and what happens when we're done.
THE SHORT VERSION
Each agency's data lives in its own database. Our shared code contains no client data.
Your data is not used to train a shared model, ours or Anthropic's.
A person reviews the work before it reaches your team. Nothing reaches a client without your approval.
You decide what is returned and what is deleted.
ACCESS AND STORAGE
WHAT WE ACCESS
Read-only exports or access to the systems you approve, scoped in writing at kickoff.
WHERE IT LIVES
A Supabase project of its own: one database and file store per agency, in the United States. No table is ever shared between agencies.
WHO CAN SEE IT
Ben Thomas. Anyone else is named in your agreement, under written confidentiality terms, before they see your data.
HOW IT'S PROTECTED
Encrypted in transit and at rest. Your team signs in to Workbench by magic link, and access is limited to your agency's own project. API keys live in server-side environment variables, never in the browser.
RETENTION
Kept for the length of the engagement plus 30 days.
WHEN IT ENDS
We return a full export on request, then delete your project within 30 days and confirm it in writing.
WHO PROCESSES IT
| PROCESSOR | PURPOSE | DATA | LOCATION |
|---|---|---|---|
| Supabase | Stores each agency's data in its own isolated database and file store | Client and prospect records and documents you share | United States |
| Anthropic (Claude API) | Reads documents and drafts text, through one audited wrapper | Only the excerpts each task needs | United States |
| Netlify | Hosts this website, your Workbench, and the request form | Form entries and Workbench sessions | United States |
| GitHub | Holds our code and configuration | No client data, ever | United States |
We'll update this list before adding any new processor, and tell active clients first.
HOW WE USE AI
Language models help us read documents, draft proposals, and surface patterns in your book. They work on the excerpts each task needs, not your whole database. Under Anthropic's commercial terms, API inputs are not used to train their models.
Improvements to our methods carry from one engagement to the next. Your data, and the analysis built on it, do not.
If something goes wrong, we tell you within 72 hours of knowing, along with what happened and what we're doing about it. Questions: ben@bellwethergrowth.com.
ONE AGENCY PER MARKET
We work with one agency per market. Tell us where you are and what you're trying to solve. Every request is read personally, and you'll get an honest answer on fit and availability.