How Ace AI Protects Your Follow Up Boss Data

Why data security is a different question for real estate AI

Real estate agents and teams handle unusually sensitive data in their CRM. Follow Up Boss contacts include names, phone numbers, email addresses, home addresses, financial situation signals, and personal details shared during the buying or selling process. This is not generic business contact data — it's information that carries real privacy risk if mishandled.

When you add an AI layer to that CRM, you're doing something new: you're asking an AI model to read, analyze, and act on that contact data. The question "is this AI safe?" becomes a concrete technical question with specific answers. Here's what we actually do. (For the short version across all of safety, privacy, and compliance, see is Follow Up Ace safe?)

What happens to your FUB data before it reaches an AI model?

The most critical security boundary in an AI-connected CRM is the point where your data crosses into a large language model. LLMs are external services — they receive whatever text is sent to them as input. The question is: what form is that text in when it arrives?

PII tokenization at the LLM boundary

Follow Up Ace tokenizes personally identifiable information before it reaches any LLM (verified: chat-app/utils/piiAnonymizer.js). Email addresses and phone numbers are replaced with consistent placeholder tokens — for example, an email like [email protected] becomes [EMAIL_a3f8b2c1] before being sent to the model. The mapping between the token and the real value is maintained server-side and is never sent to the LLM.

This means: if the LLM provider's logs captured the input, they would see tokens, not your clients' real contact information. The model responds using the tokens, the server deanonymizes the response, and only your interface sees the real data.

This is the technical implementation of the claim "PII is tokenized at the LLM boundary" — it's in code, not just a policy statement.

No model training on your data

Your FUB contacts, conversation history, call notes, and pipeline data do not train any model — ours or any provider's. The AI models used by Follow Up Ace (OpenAI's API) operate under API terms that do not use API inputs for model training by default. This is OpenAI's published policy and is separate from the consumer ChatGPT product.

We do not train any proprietary model on your account data. We use your data at inference time to generate responses for you, not at training time to improve models for others.

Encryption: what's protected and how

Encryption at rest

Sensitive fields stored in Firestore and related databases are encrypted at rest using AES-256-GCM (verified: chat-app/utils/encryption.js). AES-256-GCM is an authenticated encryption scheme — it provides both confidentiality (the data is scrambled without the key) and integrity (any tampering with the ciphertext is detectable). The authentication tag prevents silent corruption or modification of encrypted data.

Encryption keys are derived from environment secrets and are required in production — the code explicitly refuses to store plaintext sensitive values if no key is configured.

Encryption in transit

All communication between your browser, the Follow Up Ace servers, Follow Up Boss, Firestore, and LLM providers happens over TLS (HTTPS). This covers the entire data path: FUB API calls, webhook events, AI model requests, and the embedded interface in your browser. The compliance page summarizes this as "256-bit at rest and in transit across all FUB / Firestore / LLM paths" (verified: chat-app/public/compliance/index.html:260).

Fair Housing compliance: code-enforced, not policy-only

A critical but often overlooked security and legal dimension for real estate AI is Fair Housing compliance. AI models can — and without safeguards, will — generate text that makes illegal distinctions based on protected characteristics including race, color, national origin, religion, sex, familial status, and disability.

Follow Up Ace includes a code-level compliance scan function, scanForComplianceViolations(), that runs on AI-generated text before it's presented to the agent (verified: chat-app/utils/complianceGuard.js:293). This function checks against Fair Housing rules and licensing boundary rules.

The compliance layer is also injected into the LLM system prompt — the model receives an explicit compliance directive at session initialization that instructs it to refuse, rephrase, or flag responses that would violate Fair Housing or professional licensing boundaries.

This is two layers of protection: a system prompt directive that reduces the likelihood of a violation at generation time, and a code-level scan that catches violations before the output reaches the agent. Learn more on the compliance overview page.

SOC 2 Type II: where we are honestly

SOC 2 Type II is an audited security framework from the AICPA that evaluates a company's controls around security, availability, processing integrity, confidentiality, and privacy over a period of time (typically 6–12 months). It's a meaningful signal of organizational security maturity.

Follow Up Ace's SOC 2 Type II audit is currently in progress. It is not yet complete. We will publish the report when the audit is finished — not before, and not in summary or draft form (verified: chat-app/public/compliance/index.html:252 — "Audit underway. We'll publish the report when it's complete — not before.").

We are not claiming SOC 2 compliance today. What we're telling you is: the audit is underway, the technical controls described in this article are the things being audited, and we expect to complete the process. We will update this page and the compliance page when the report is available.

If SOC 2 Type II completion is a hard requirement for your brokerage's vendor approval process, the honest answer is that we're not there yet. The controls are in place; the audit attestation is not.

GDPR-aligned data handling

Follow Up Ace handles data in a GDPR-aligned manner even for US-based users where GDPR doesn't directly apply — because the practices are good data hygiene regardless of jurisdiction. Specifically:

• Data Subject Access: Users can request a copy of their data
• Right to delete: Accounts can request deletion of their data
• 90-day purge on cancellation: When you cancel your Follow Up Ace account, your FUB data stops being processed immediately. Your FUB data stays in FUB (which is separate). We retain your account record for 90 days, then purge it.

Your client data in Follow Up Boss itself remains in Follow Up Boss — it's owned by your FUB account, not by Follow Up Ace. When you disconnect or cancel, our access to it ends.

FUB Marketplace certification

Follow Up Ace is listed in the official Follow Up Boss integration marketplace. FUB Marketplace certification requires that an integration meet Follow Up Boss's standards for API usage, security practices, and integration quality. This is an active listing that can be revoked — it's not a one-time badge.

You can verify this listing directly in your Follow Up Boss account settings under Integrations.

Authentication: how Ace verifies it's talking to you

The embedded Follow Up Ace assistant authenticates through your Follow Up Boss session context. When the embed loads inside FUB, it receives a base64-encoded, HMAC-signed context payload from Follow Up Boss that contains your identity and account information. The server validates the HMAC signature on every request — requests without a valid signature are rejected.

Admin users authenticate separately through Firebase Auth with email and password. Admin sessions are scoped to the admin interface and do not share session state with the agent-facing embed.

What about the AI models themselves — are they secure?

The AI models used by Follow Up Ace are accessed through provider APIs (OpenAI). The security of those providers is subject to their own security programs. OpenAI maintains its own security posture documentation at openai.com/security. We use API access, not consumer products, which carry different data handling terms.

The PII tokenization described above means that even in the event of a provider-side security issue, your clients' real personal information would not be exposed from the LLM side — it was never sent in plaintext form.

The security posture: honest summary

For the complete compliance picture including the full Fair Housing compliance framework, see the compliance overview. For questions about your specific security requirements, contact us before signing up — we'd rather answer your questions honestly upfront than have them surface after you're onboarded.