Security & Data Handling
Last updated:
Before you give anyone access to your business systems, you should know exactly what they will touch and how you take it back. This page explains how Cadence Ridge handles access, business and customer data, and AI models, in plain English. The written agreement for an engagement is what binds us; this page describes how we work by default.
The short version
- The free audit needs no access at all. Never send us passwords or logins.
- When a build needs access, your team creates it, in accounts your company owns, with the fewest permissions the work needs.
- A person approves anything that moves money or reaches a customer.
- Every run is logged, failures raise an alert, and you can pause the system yourself.
- When we finish, you remove our access and keep everything we built.
How access works
The audit needs no access
The free Automation Audit is a 30-minute call. We don't need access to any of your tools for it; screen-share is plenty. If you send numbers afterward, counts, dates, and screenshots are enough, and anonymized is preferred.
No shared passwords, ever
We never ask for your password or an owner or admin login, and we won't accept passwords or logins by email or on a call. If a build needs a credential such as an API key, it is passed through a password manager, not plain email, and kept there.
Your team grants access, with the minimum permissions
- Access is a staff or collaborator account, or an app or API token, that your team creates in accounts your company owns. You can see it, and remove it, at any time.
- We ask only for the permissions the agreed workflow needs. If we are offered more than that, we ask for less.
- Read-only first, where possible. Where a tool allows it, we start with read-only access and add write access only for the steps that need it.
- Where a tool has a sandbox or test environment, we use it. If it doesn't, we use a clearly labeled test area with made-up records, approved by you, and remove it afterward.
Your business and customer data
What we touch
Only the data the workflow in your written scope needs. For a lead follow-up or support workflow, that usually means the contact, account, and ticket or order records the step reads, and the customer details on them (name, email, company, history). It does not include payment card numbers; we don't need them.
Keeping it to the minimum
- For the audit and diagnostic we ask for the minimum: counts and dates rather than message contents or customer details.
- Testing uses anonymized examples or made-up cases. If an example contains sensitive details, we anonymize it or use made-up cases only.
- Anything you send us is kept in a folder for your account only. We don't paste your customer data into outside tools or chat assistants; if we need an example there, it is anonymized first.
Where it is processed
- In your own tools. Systems are designed to run in accounts your company owns: your CRM, helpdesk, email, accounting, store platform, and spreadsheets.
- Any added service is named first. If a workflow needs a tool you don't already use (for example an automation platform or an AI model provider), it is named in the scope, billed to you directly, and disclosed before you approve.
- Before you are a client, what you send through our website or by email is handled as described in the Privacy Notice: Cloudflare hosts the site and the form, Resend delivers the form to our inbox, and Google provides analytics.
How long we keep it
Audit requests and emails are kept up to 12 months after our last contact, unless we start working together or you ask us to delete them sooner. For a client engagement, the written agreement sets retention, and we delete our copies when the work ends. Cadence Ridge is run by one person, Sidney Rees, as a sole proprietor, so there is no wider team with access to your data.
AI models
When AI is used, and what it can do
Some steps use an AI model, for example to read a ticket or an email and sort it, pull out an account or order number, or draft a reply. The model works only from approved sources (your policies, your templates, the records the step needs) and with defined tools. When it isn't confident, or its output doesn't pass a check, the case goes to a person instead.
The AI never makes the final call on anything that moves money or reaches a customer. Drafted replies stay drafts until a person approves them, and refunds, payments, record changes, and purchase orders wait for approval. Fully autonomous replies with no approval are not part of our standard builds.
Do the AI providers train on your data?
Builds call AI models through the provider's business API (Anthropic's Claude or OpenAI's models, named in your scope), not through consumer chat apps. Per each provider's published policy:
- Anthropic does not use inputs or outputs from its commercial products, including the Anthropic API, to train its models by default. The exceptions it lists are feedback or bug reports a customer sends, and a customer choosing to allow it. Anthropic's policy
- OpenAI does not use data sent to its API to train or improve its models unless the customer explicitly opts in. By default it keeps abuse-monitoring logs for up to 30 days. OpenAI's policy
We don't opt in to sharing your data for training. These are the providers' policies, not ours, and they can change; check the links for the current wording.
Controls you keep
- Human approval. A person on your side approves anything that moves money or reaches a customer: refunds, payments, record changes, purchase orders, and customer messages. For the first week live, follow-ups are drafts or wait for approval.
- Tested before it goes live. Every build runs in a sandbox or test area first, against real (anonymized) and made-up cases, including the awkward ones. Nothing goes live until you have approved the test results in writing.
- Every run logged. Each run records what came in, what the system decided, what it did, and any errors. Records the system creates are labeled so they can be found and reversed.
- Alerts on failure. If a step fails, it retries, and if it still fails, the item waits in a visible queue and an alert goes to us and to your named contact. Nothing is dropped silently.
- A pause switch you control. Every system has a kill switch checked at the start of every run: on, review-only (everything goes to a person, nothing is written or sent), or off. You get a one-line instruction for using it, and you don't need us to press it.
When we finish
Whether the project ends or you move to someone else, leaving should be simple:
- Revoke our access. Remove our staff or collaborator accounts and revoke the tokens and API keys created for the work. We give you the list, so nothing is missed.
- Keep what we built. The workflows run in your accounts. You get the documentation, an export of each workflow, and the list of every connection and key it uses, so another team can run it.
- We delete our copies. Any exports, test data, or notes with your customer data that we hold are deleted, and we confirm by email when it is done.
For the general version of this, including how to leave any vendor, see the vendor exit checklist.
Certifications
Cadence Ridge is a one-person business and does not hold security certifications or audit reports such as SOC 2. If your business requires them from vendors, tell us on the audit call and we'll say plainly whether we are a fit.
Report a security issue
If you think you have found a security problem with this website or with a system we built, email [email protected] with the subject “Security”. Include what you found and how to reproduce it, and please don't access or change other people's data while checking. We read every report and will reply. The same contact is published in /.well-known/security.txt.
See also our Privacy Notice and Website Terms of Use.