Prodoscore AI & MCP Privacy Policy

1. Scope and Our Role

This Notice describes how Prodoscore processes personal information in two AI-enabled parts of our Services:

• Prodoscore AI — the AI Insights and AI Chat features inside the Prodoscore application, which generate written summaries and answer questions about workforce analytics.
• The Prodoscore MCP Server — a read-only Model Context Protocol endpoint that allows an authenticated Prodoscore user to query their permitted workforce analytics from a third-party AI assistant chosen by their organization.

Our role. For the information described in this Notice, Prodoscore acts as a processor (under the GDPR and UK GDPR) and a service provider (under the CCPA/CPRA) on behalf of our Customer — the organization that subscribes to Prodoscore. The Customer is the controller and business, and determines what is collected, who may see it, and how it is used. Prodoscore processes it only on the Customer’s documented instructions and as described here and in the DPA.

Key terms used in this Notice

If you are an End User: your employer decides what data is collected about you, who can see it, and how it is used. Please direct questions about those decisions, and requests to exercise your privacy rights, to your employer in the first instance. See Section 10.

2. Personal Information We Process

2.1 Identity and organizational information

• Employee identifier
• Full Name
• Job Title or Role
• Department
• Manager
• Organizational Unit
• Time Zone.

Email addresses are not returned by the MCP Server. An End User’s email address is used only at the moment of sign-in to authenticate them against the Prodoscore identity provider. It is not included in any analytics result the MCP Server returns.

2.2 Activity and productivity information

All of the following are counts, durations, timestamps, and derived scores — never the content of your communications:

• The Prodoscore productivity score (0–100) and its classification.
• Active time, gap time, working hours, and the first and last recorded activity of a day.
• Days worked and not worked, scheduled working days, and shift or schedule information.
• Meeting counts and durations, split between internal and external meetings.
• Volumes of email, chat, and meeting activity, including a breakdown by whether the counterparty is a supervisor, peer, subordinate, or external to the organization.
• Application and product usage, including usage scores, adoption rates, and session start and end times.
• Where the Customer has deployed the optional Prodoscore Desktop Agent: per-application and per-website duration, and mouse-click counts.
• Where the Customer has deployed the optional Prodoscore browser extension: per-website duration.
• Employee-specific holiday and paid-time-off dates.
• A retention-risk score: a modelled probability and risk tier indicating likelihood of attrition based on historical employee attrition patterns.

2.3 AI interaction information

For Prodoscore AI only: the questions an End User asks the AI assistant, the assistant’s responses, the resulting conversation transcript and its automatically generated topic, any feedback given on a response, usage and token counts, and internal quality scores produced by our automated evaluation process.

2.4 Technical and security information

For both Services: employee identifier, organizational unit, the name of the analytics query run, the identity of the AI Client used, response timing, and — for authentication events — browser user agent presented at sign-in.

2.5 What we do not process in these Services

Neither Service reads, stores, or sends to any AI Provider:

• The content of emails, chat messages, or documents.
• Meeting titles, agendas, attendee lists, recordings, or transcripts.
• Screenshots, keystroke content, camera, or microphone data.
• Passwords or payment information.

3. Where the Information Comes From

The analytics described in Section 2.2 are derived from the business systems the Customer has chosen to connect to Prodoscore. These typically include Google Workspace, Microsoft 365, CRM systems such as Salesforce, calendar and chat tools, HR and time-tracking systems, and — where deployed — the Prodoscore Desktop Agent and browser extension. The Customer decides which integrations to enable; Prodoscore does not add sources on its own initiative.

The AI features described in Section 2.3 draw on the analytics above, plus what the End User types into the assistant.

4. How We Use the Information

We process the information described above only to:

1. Deliver the Services to the Customer — producing analytics, summaries, and answers to questions.
2. Authenticate End Users and enforce the access limits the Customer has configured.
3. Operate, secure, monitor, and troubleshoot the Services, including detecting abuse and maintaining an audit record of access.
4. Measure and improve the quality and accuracy of AI responses, using automated evaluation.

We do not use End User personal information to train or fine-tune our own models, and we do not use it for advertising or marketing. Our AI Providers’ commitments are described in Section 5.1.

5. Disclosure to AI Providers and AI Clients

5.1 In-product AI — Google Vertex AI

Prodoscore AI is powered by Google’s Gemini models, accessed through Google Cloud Vertex AI under Gemini Enterprise, running in a Prodoscore-controlled Google Cloud project. To generate a summary or answer a question, we send Vertex AI the End User’s question, relevant analytics results (which may include employee names, identifiers, and the metrics described in Section 2.2), and our own instructions to the model.

Google processes this data as our subprocessor under the applicable Google Cloud terms. Customer data submitted to Vertex AI is not used to train or fine-tune Google’s foundation models.

The assistant may also consult general reference sources, including web search, for background information such as definitions or public facts. Employee personal information is not included in those lookups.

Prodoscore uses Google Cloud Vertex AI under Google’s applicable enterprise data protection and Minimal Data Retention commitments. Customer prompts and related data are not used by Google to train or fine-tune its foundation models.

5.2 Third-party AI Clients via the MCP Server

Where the Customer enables it, an End User may connect an AI assistant of the Customer’s choosing to the Prodoscore MCP Server. Supported assistants currently include Anthropic Claude, OpenAI ChatGPT, Microsoft Teams and Copilot Studio, and Google Gemini Enterprise; the supported set may change over time.

How a connection works, end to end. When the feature is enabled, and an End User connects an AI assistant and asks a question, the following occurs:

1. Enablement. The Customer has turned the feature on for the organization and for that individual End User; absent both, no connection is possible (Section 9).
2. Authentication. The End User authenticates individually to the Prodoscore MCP Server (OAuth 2.1 with PKCE). The AI assistant connects on that authenticated user’s behalf and inherits only that user’s permissions.
3. Query. The assistant issues a read-only analytics query. Prodoscore resolves it against only the data that the End User is already permitted to see in Prodoscore (Section 6), and returns the results.
4. Transmission to the provider. The assistant returns those results to the End User through the assistant’s provider. At that point, the analytics results have left Prodoscore and become subject to that provider’s own privacy policy, terms, and data retention practices.

Prodoscore does not control, and is not responsible for, what that provider does with the data once it is transmitted — including whether it is retained or used to improve that provider’s models.

Customers should note the following before enabling the feature:

• Read-only. An AI Client can query analytics; it cannot create, alter, or delete anything in Prodoscore.
• Per-user access. Each End User authenticates individually, and an AI Client can only retrieve data that the End User is already permitted to see in Prodoscore (Section 6).
• Off by default. The feature must be enabled by the Customer, both for the organization and for each user (Section 9).

What the MCP Server can and cannot return

The Prodoscore MCP Server exposes read-only analytics derived from activity metadata. Through an authenticated connection, an AI Client can retrieve the following, always limited to the data the End User is already permitted to see (Section 6):

What it can return

• Productivity scores — Prodoscore values for an individual, team, department, role, or organization.
• Active time and gap time — time active versus idle gaps on worked days.
• Work schedule and worked days — working patterns, worked/holiday days, and schedule information.
• Meeting information — meeting counts and durations at the individual, team, department, or organization level.
• Product usage — which applications and products were used and how much (by category, e.g., standard products, browser extension, desktop agent), including per-product sessions and usage rankings.
• Retention-risk indicators — model-derived risk signals for individuals the End User manages.
• Organizational structure — managers, direct/indirect reports, department and role membership, and directory lookups by name.
• Rankings — employees ranked by score, active time, gap time, or product usage, within the End User’s permitted scope.

What it cannot return

• No underlying content. The Server returns activity metadata only. It does not expose the content of emails, chat or messages, documents, files, calendar event details beyond meeting metadata, meeting or call transcripts or recordings, keystrokes, screenshots, or the specific web pages or URLs a person viewed.
• No data outside the End User’s access level. An AI Client inherits only the connecting End User’s permissions (SELF, TEAM, or ORGANIZATION) and cannot retrieve any individual’s data the End User could not already see in Prodoscore.
• No writes. The connection is read-only: an AI Client cannot create, modify, or delete scores, users, settings, or any other record in Prodoscore.
• No cross-provider or administrative actions. The Server exposes analytics queries only; it does not expose account administration, billing, configuration, or integration-management functions.

For the MCP Server specifically, our Services are designed to ensure that conversational inputs and analytics outputs are not retained once a request is finalized. Prodoscore does not receive the End User’s natural-language prompt, which is handled exclusively by the AI Client. The MCP Server receives only the structured analytics requests necessary to fulfill the AI Client’s request and return the permitted analytics results. Analytics data used to fulfill a request is maintained only in a temporary, session-specific cache to facilitate immediate repeat queries and is not committed to persistent databases. We retain only limited invocation metadata — including the user identity, access permissions, and request timestamps — for security auditing and service monitoring. This audit record is strictly limited to technical identifiers and does not include the specific query details or result contents.

5.3 Other recipients

Beyond the above, we disclose this information only to infrastructure subprocessors that host and operate the Services, and where required by law. We do not sell personal information, and we do not share it for cross-context behavioral advertising. The advertising and marketing technologies described in the Prodoscore Privacy Policy operate on our public websites and are not present in either Service covered by this Notice. Our current subprocessors are listed in the DPA.

6. Access Controls and Data Minimization

Both Services enforce the access level the Customer has assigned to each End User:

Every query is additionally restricted to the End User’s own Customer organization; no query can return data belonging to another Customer. Requests for data outside a user’s permitted scope are refused, and the refusal is recorded.

Further minimization measures applied by design: the MCP Server is read-only; it returns no email addresses; results are capped in row count and date range; and the analytics it exposes are aggregates and derived metrics rather than communication content.

7. Automated Processing, Profiling, and the Limits of AI Output

Prodoscore is by design an automated analytics product. Productivity scores, activity classifications, and retention-risk scores are automated inferences about working patterns. They are estimates derived from activity signals, not measurements of an individual’s value, effort, or performance, and they can be wrong or incomplete.

AI-generated summaries and chat responses are produced by a large language model. They may be inaccurate, incomplete, or misleading, and should be verified against the underlying data before being relied upon.

Prodoscore does not make employment decisions. We do not use these outputs to make, and do not provide them for the purpose of making, any decision producing legal or similarly significant effects about an individual. Any decision — about performance, compensation, discipline, or continued employment — is made solely by the Customer, which is responsible for how it interprets and acts on the Services’ output, and for meeting any applicable workplace-monitoring notice, consultation, or works-council obligations in its jurisdiction.

Individuals in the EEA or UK who wish to object to automated processing or profiling should contact their employer, which as controller is responsible for responding.

8. Retention

Upon termination of a Customer’s subscription, data is deleted or returned upon the Customer’s written request, in accordance with the Prodoscore DPA.

9. Customer Administrator Controls

A Customer controls the AI features described in this Notice and can:

• Enable or disable MCP access for the entire organization, and separately for each individual user. Both must be enabled for a connection to succeed; disabling either takes effect promptly.
• Choose which AI Clients, if any, are permitted, and revoke a previously authorized client’s credentials.
• Choose which business systems are integrated as data sources, and whether to deploy the optional Desktop Agent or browser extension.
• Configure each End User’s access level, which determines the scope of everything both Services can return.

End Users can delete individual AI conversations from the Prodoscore interface.

10. Your Rights

If you are an End User (an employee or contractor of a Prodoscore Customer): Prodoscore processes your information on your employer’s behalf. Please contact your employer to exercise your rights of access, correction, deletion, restriction, portability, or objection — including objection to automated processing and profiling. Where Prodoscore receives such a request directly, we will refer it to the Customer and assist the Customer in responding, as required by the DPA.

California (CCPA/CPRA). The categories of personal information processed in these Services are identifiers, professional and employment-related information, internet and electronic network activity information, and inferences drawn from that activity. It is processed for the business purposes set out in Section 4. We do not sell, and do not share for cross-context behavioral advertising, the personal information described in this Notice. In these Services, Prodoscore acts as a service provider and does not retain, use, or disclose this information for any purpose other than performing the Services. Section 7 of the Prodoscore Privacy Policy describes California rights in full.

EEA and UK. Section 9 of the Prodoscore Privacy Policy describes these rights. For information processed as a processor on a Customer’s behalf, the Customer is the controller and is responsible for responding.

11. Security

We protect this information with encryption in transit and at rest; individual authentication using OAuth 2.1 with PKCE (S256) for AI Client connections; least-privilege service credentials; network isolation of internal services; and audit logging of access and authentication events.

AI Client connections authenticate against a single MCP resource scope; this scope grants connection, not data access. Data access is governed separately and enforced server-side on every query: each authenticated user is bound to one of three access levels — SELF (their own data), TEAM (their direct reports), or ORGANIZATION (their domain and any child domains) — derived from the user’s identity and role rather than from client-requested scopes. Requests for data outside a user’s access level are rejected. This role- and organization-scoped enforcement, applied uniformly across all tools, is the effective authorization boundary.

We further protect the platform with automated security and dependency scanning in our development pipeline.

Prodoscore maintains an Information Security Program aligned with AICPA standards and undergoes annual independent third-party audits to maintain SOC 2 Type II certification covering the Security, Confidentiality, and Availability Trust Services Criteria. Our current SOC 2 Type II report is available to Customers upon request.

Security incidents involving personal information processed in these Services are handled in accordance with the breach notification provisions of the DPA.

No system can be guaranteed secure. Section 13 of the Prodoscore Privacy Policy applies.

12. International Transfers

The Services are hosted in the United States, and the AI processing described in Section 5.1 takes place in Google Cloud infrastructure. Information may therefore be transferred to, and processed in, the United States and other countries where our subprocessors operate.

Prodoscore applies the applicable data protection and contractual safeguards governing international transfers, as described in the DPA and applicable subprocessor agreements.