Hosting, sub-processors and AI Act readiness, in plain language
You're evaluating AI that touches employee performance data. Here's exactly where it's hosted, who else sees it, what the AI does at each automation level, and where we stand on the EU AI Act - checked against the product, not written as boilerplate.
The whole product - database, file storage and application servers - runs on AWS in the eu-west-1 (Ireland) region. There is no other region to opt into or out of: it's the only deployment that exists, so every customer's data stays in the EU by default, not as a paid add-on.
Avatar images are the one thing stored with public-read access (an S3 bucket with public access enabled, specifically for that), because they need to load without authentication in the app. Everything else - feedback text, ratings, improvement points, company and user records - sits behind the application and database, reachable only through the API.
Who processes your data
Three sub-processors, and we'll say so if that ever changes:
AWS (eu-west-1) - hosting, the database and file storage.
Anthropic (the Claude API) - feedback summaries, AI-suggested ratings, generated improvement points and their evaluation. Only what a cycle's automation level actually calls for is sent; a Classic-automation cycle sends nothing automatically. The one exception is the "Generate summary" button on a review, which calls the API only when a manager clicks it.
Voyage AI (embeddings) - only the wording of a generated improvement point, with the person's and company's names stripped first, so similar points can be grouped for the team view and follow-up cycles. It never receives feedback text. Voyage's terms let it train its models on what it receives unless the customer opts out; we have opted out, which under those terms means what we send is deleted once Voyage has processed it.
What the AI actually does, at each level
Automation is a dial, not an on/off switch, and it's set per review cycle, not once for the whole company. Classic runs no AI automatically - no feedback or rating is sent to an AI model unless a manager clicks the "Generate summary" button on a review. Light adds AI summaries only; a human still writes every rating. Enhanced adds AI-suggested ratings that a manager can always override before a review is compiled. Full Auto goes furthest - see the confirmation step below for what stays human even there.
The panel below shows exactly what changes at each level. It's the same panel shown on the pricing page and in the app when a company picks its automation level - nothing is simplified or left out for this page.
The engine compiles the review end to end, then a manager confirms it with one click before it submits.
Compiling reviews
Summarizes every feedback submission before it reaches the compile screen
Estimates each item's rating from feedback text, for the manager to review
Compiles the review automatically - a manager still confirms it with one click before it submits
Improvement points
Suggests actionable and trackable new improvement points each cycle
Tracks improvement points that aren't tied to a category
Continuity add-on
Drafts a follow-up cycle to re-check unresolved improvement points
Runs those follow-up cycles on autopilot after a 3-day grace period
Improvement points tied to a category always track automatically from the rating trend, at every level.
A human always signs off
GDPR Article 22 restricts decisions based solely on automated processing when they have a significant effect on someone - a rating that feeds pay or promotion clearly qualifies. That's the setting we designed around, not an afterthought bolted onto Full Auto after the fact.
Below Full Auto, a manager compiles the review themselves, sees every AI-suggested rating and summary inline, and can edit or override anything before submitting. At Full Auto, the system compiles the review on its own, but - when the cycle has manual approval turned on - the review stops at "awaiting confirmation" and is assigned to the same person who'd have compiled it manually. They see the finished review and confirm it with one click before it submits. There is no automation level in review.center where a review reaches an employee's record without a person having had the chance to see and stop it.
Our EU AI Act roadmap
AI systems used to monitor and evaluate workers' performance and behaviour are classified high-risk under Annex III of the EU AI Act. Following the 2026 Digital Omnibus, those obligations are reported to take effect from 2 December 2027. As the provider, that means risk management, data governance, technical documentation, logging, human oversight and a conformity assessment on our side; as the deployer, your company has its own obligations too, including telling your people the tool is in use.
Where we stand today: the human-confirmation step above is built and live for every Full Auto cycle with manual approval enabled - not gated by region, since Romania-and-EU-first is effectively our whole customer base anyway. Risk management documentation, logging for audit purposes and a formal conformity assessment are not yet built; they're scoped ahead of the December 2027 deadline, not after it.
We deliberately don't market features like burnout detection, attrition prediction or emotion analysis of employees, even though they'd be easy to promise. review.center only summarises and follows up on what reviewers actually wrote.
Your rights over your data
A Data Processing Agreement is available on request - email office@review.center before sending us real employee data and we'll get one signed. We don't yet have a self-serve download for it, so that happens directly rather than through a signup flow.
Career-portable profiles are the clearest place GDPR data portability (Article 20) shows up in the product: with the person's explicit consent, their numeric rating trend and improvement-point outcomes - never the underlying feedback text - can move with them to their next review.center employer. Nobody is enrolled in this without agreeing to it themselves.
We don't yet have a self-serve export or delete tool for company data. Until we do, email office@review.center and we'll handle either directly.
FAQ
Questions about data, AI and trust
Where is our data hosted?
On AWS in the eu-west-1 (Ireland) region - the database, file storage and the application servers all sit there. We don't have a US or other-region deployment today, so EU data residency isn't a configuration choice, it's the only deployment that exists.
Does feedback text train Claude or any other AI model?
No. Feedback text is sent to Anthropic's Claude API to generate summaries, ratings and improvement points, under Anthropic's commercial API terms, which state that API inputs and outputs are not used to train their models. Voyage AI, which only ever sees the anonymised wording of an improvement point and never feedback text, may train on what it receives under its default terms; we have opted out of that. We don't separately use your data to train anything ourselves either.
Can we get a Data Processing Agreement?
Yes - email us at office@review.center and we'll get one in place before you send any real employee data. We don't yet have a self-serve DPA download, so this happens by email rather than through a signup flow.
Who are your sub-processors?
AWS (hosting, eu-west-1), Anthropic (the Claude API, for summaries, ratings and improvement points) and Voyage AI (an embedding model that turns the wording of an improvement point into numbers so similar points can be grouped; it never receives feedback text, and the person's and company's names are removed from that wording first; its terms allow training on received content unless you opt out, and we have). That's the full list - there's no analytics or support vendor with access to review content today.
Can an AI decide someone's rating with no human involved?
No. Even at the highest automation level (Full Auto), a review that would otherwise submit automatically instead stops at "awaiting confirmation" and needs one click from the person who'd have compiled it manually, before it submits. Below Full Auto, a manager compiles or approves the review directly. There's no automation level where nobody looks at it.
How long do you keep our data, and can we delete it?
We don't yet have a published retention schedule or a self-serve export/delete tool - both are on our roadmap alongside self-serve signup. Until then, email office@review.center for an export or deletion of your company's data and we'll handle it directly.
Is review.center affected by the EU AI Act?
Yes. AI systems used to monitor or evaluate worker performance are classified high-risk under Annex III, with obligations reported to apply from 2 December 2027. As the provider, we owe risk management, data governance, technical documentation, logging, human oversight and a conformity assessment; your company, as the deployer, has its own obligations too, including telling your people the tool is in use. See the roadmap section below for where we stand today.
Still have a question?
Email office@review.center - a DPA, a question for your security review, or a request to export or delete your company's data all go to the same address, and a person answers.