What customers write in feedback forms
Customer feedback is not abstract data. It is specific: "The billing page charged me twice for the same seat," "I had to reset my password three times and then gave up," "Your API rate limits are blocking us from migrating off a competitor." That level of specificity is what makes feedback valuable — and what makes it sensitive.
A single support ticket can contain a customer's name, their organization, the plan they are on, a complaint about a specific transaction, and a frustration that signals churn risk. Aggregated across thousands of users, a feedback database reveals which features are failing, which cohorts are unhappy, and where the product is losing people to competitors. That is commercially valuable — which means it is worth protecting.
What happens when you send feedback to a SaaS tool
The standard SaaS model is: you send feedback text to the tool via API or CSV, the tool processes it (often by sending subsets to a third-party LLM API), and you get back categorized results. The tool stores the raw text and the derived data in its own database, under its own data retention and access control policies.
In practice, this means at minimum two third parties have a copy of your customer feedback: the SaaS tool and whatever AI provider the tool uses for analysis. The SaaS tool's terms determine how long they keep it, whether they use it for model training, and what happens to it if the company is acquired. The AI provider's terms add a second set of retention and usage rules.
None of this is necessarily malicious. But it does mean you have accepted a set of terms governing your customers' data that your customers themselves did not consent to — and that you may need to disclose in your own privacy policy.
What self-hosting actually changes
When you self-host Rereflect, the feedback text stays in your PostgreSQL database, on your infrastructure, in whatever region you choose. The analysis that Rereflect runs — sentiment scoring, pain point detection, urgency flagging — runs against a model you configure, using credentials you own. Nothing flows to a Rereflect server.
If you use a hosted LLM key (OpenAI, Anthropic, etc.), the feedback text does flow to that provider when it is analyzed. If you use a local model via Ollama or another OpenAI-compatible runtime, even that is eliminated. The VADER fallback removes the LLM call entirely for sentiment analysis.
The practical result is that you can draw a clear, honest line in your privacy policy: "Customer feedback is stored on our own infrastructure and processed locally." That is a statement many SaaS tools cannot truthfully make.
Ownership is also about access and deletion
Data ownership is not only about where data is stored. It is also about whether you can access it, query it freely, and delete it completely. With a SaaS tool, deletion is typically "submit a request and we will process it within N days." Export is typically "download a CSV of what we decide to expose in our export UI."
With a self-hosted Rereflect, your PostgreSQL database is your database. You can run any query, export in any format, and delete individual records or the entire dataset whenever you want. There is no intermediary to ask permission from, and no retention window you have to wait out.
For teams operating under GDPR, CCPA, or contractual data processing agreements with enterprise customers, this is not a nice-to-have. It is the difference between being able to honour a deletion request and having to explain to a customer why their data is still sitting in a third-party system.
The honest trade-offs
Self-hosting means you are responsible for the infrastructure: backups, uptime, upgrades, and security configuration. A managed SaaS tool handles those things for you in exchange for custody of your data. That is a legitimate trade-off, and different teams will land in different places.
What matters is that the choice is explicit. If you have decided a managed tool is fine and you have read and understood its data handling terms, that is a reasonable decision. What is harder to justify is choosing a managed tool without having thought through what happens to the feedback data that flows through it.
Rereflect is open source and MIT-licensed. You can read every line of code that processes your feedback. If you decide to self-host, the full deployment tooling is in the repository. If you decide a managed setup fits your situation better, you have made that decision with eyes open.