AI Security FAQ's

1. What type of AI technologies does RevSure use?

RevSure utilizes a combination of Predictive AI/ML and Generative AI:

• Predictive AI/ML: Built using advanced machine learning models trained on customer-specific Go-To-Market (GTM) data. These models provide granular predictions at the lead, account, opportunity, campaign, channel, and overall pipeline/bookings levels.
• Generative AI: Powered by Google Gemini LLMs, these models are used solely for summarization and recommendations. RevSure does not train or fine-tune any LLMs.

2. What data is used to train the AI models?

Predictive AI/ML:

• Trained using structured, non-PII data from a customer’s GTM systems (e.g., Salesforce, Hubspot, MAPs, ABM tools).
• Data includes lead/opportunity/account attributes, activity metrics, campaign engagement, and behavioral patterns.
• Creates a unified GTM Data Graph to power predictive insights.
• Each model is trained uniquely for each customer — no model re-use across tenants.

Generative AI:

• RevSure uses no training data to build or modify LLMs.
• Utilizes aggregate data, structured metadata, and non-PII behavior patterns from customer GTM systems.
• Governed under Google Gemini’s Data Protection & Compliance Framework.

3. Are third-party foundational models used?

Yes. RevSure exclusively uses Google Gemini LLMs to support its Generative AI capabilities.

4. Does RevSure integrate with external platforms?

Yes. RevSure integrates with the following platforms depending on customer setup:

• CRM: Salesforce, HubSpot
• MAP: Marketo, Pardot, etc.
• Sales Automation: Outreach, Salesloft
• Paid Ad Platforms: Google Ads, LinkedIn Ads
• ABM Tools
• Website Analytics Systems

5. Is customer or user data used to train LLMs?

No. RevSure:

• Does not use any customer prompts, data, or outputs to train LLMs.
• Does not train or fine-tune LLMs.
• Leverages Gemini LLMs strictly for runtime tasks like summarization and recommendations.

6. Is PII used for model training or inference?

• Predictive AI: Uses non-PII data only. PII is not part of the model training process.
• Generative AI: May receive minimal masked identifiers (e.g., lead name, work email, company) strictly for context; however, customers can disable this. All such data is redacted/masked before submission to LLMs.

7. Does RevSure prohibit re-identification of de-identified data?

Yes. RevSure:

• Does not use de-identified data for its own purposes.
• Prohibits re-identification and enforces this via internal data access, masking, and encryption policies.
• Ensures redaction of any PII prior to Generative AI interactions.

8. What AI governance or model review policies are in place?

RevSure maintains strong governance and model review processes:

Bias & Accuracy Review: All AI-generated outputs are validated for accuracy and bias.

Guardrails:

• No retraining of Gemini LLMs using customer data.
• Google Gemini’s prompt logging is disabled by default.
• PII masking is enforced.

Review Cadence: Regular monitoring and QA reviews based on customer use cases, with metrics around accuracy, impact, and feedback loops.

9. What subprocessors does RevSure use and what data do they access?

Google Cloud Platform (GCP):

• Hosts all customer data.
• Deployment regions: us-central1 (Iowa) or europe-west1 (Belgium) (customer selection).
• Full encryption and isolation enforced.

Twilio (Optional):

• Used for sending reports/dashboards via email.
• Access is limited to message content when the feature is enabled by customers.

10. What regulations does RevSure comply with?

RevSure complies with the following:

• General Data Protection Regulation (GDPR)
• SOC 2 Type II
• ISO/IEC 27001:2022

11. Does RevSure maintain insurance coverage related to AI?

Yes. RevSure maintains cybersecurity, errors, and omissions insurance relevant to AI operations.
A copy of the Cyber & E&O Policy (Effective until 2024-09-21) is available upon request.

12. Does RevSure indemnify customers for claims related to AI outputs or training data?

Yes. RevSure’s contractual agreements, including the DPA (Data Processing Agreement), provide for indemnity covering claims arising from AI model training or output generation.

Let me know if you’d like this exported into a PDF, DOCX, or ready-to-share Notion format. I can also tailor this version for procurement, IT, or legal audiences.