Version 2.2
This document outlines how Bluebird’s optional AI assistant, Aubrey, uses third-party or client-provided AI services, what data may be transmitted, how that data is handled, and the current safeguards and limitations within the platform.
Overview
Aubrey is Bluebird’s optional AI assistant designed to support and improve human-to-human sales interactions. Aubrey can provide conversational assistance, email generation, lead enrichment, lead analysis, and other language-based workflows.
Aubrey requires a compatible AI engine to provide these capabilities. Bluebird supports several deployment options depending on a client’s security, privacy, procurement, and technical requirements.
AI Provider and Deployment Options
Clients may choose from the following AI deployment models:
1. Bluebird-Managed OpenAI — Default
By default, Aubrey uses OpenAI services through API credentials managed by Bluebird.
Under this configuration:
- Bluebird manages the OpenAI integration and API credentials.
- Data required to perform enabled Aubrey functions may be transmitted to OpenAI.
- Bluebird configures its OpenAI services with data sharing and model training disabled.
- Customer data submitted through Aubrey is not made available for public model training.
- Bluebird remains responsible for managing the integration between Aubrey and OpenAI.
This is the standard Aubrey configuration and requires no additional AI infrastructure or integration work from the client.
2. Client-Managed OpenAI
Clients may elect to use their own OpenAI account and API credentials.
Under this configuration:
- Aubrey connects to OpenAI using API credentials supplied or controlled by the client.
- AI processing occurs within the client’s OpenAI account and is subject to the client’s OpenAI agreement, configuration, policies, and data controls.
- The client is responsible for its OpenAI account, usage, billing, and applicable OpenAI configuration.
- Bluebird continues to provide the integration between Aubrey and the OpenAI APIs.
Configuration requirements may vary depending on the client’s OpenAI environment.
3. Client-Selected AI Engine
Where required, Aubrey may be integrated with another client-selected AI engine or model provider.
Examples could include alternative commercial AI providers, privately hosted models, or other compatible AI infrastructure.
Because capabilities, APIs, security models, and functionality vary between AI engines:
- Integration requirements will be assessed on a case-by-case basis.
- Development work will be scoped and quoted as required.
- Aubrey functionality may vary depending on the capabilities of the selected AI engine.
- Data handling will be governed by the architecture and policies associated with the selected provider.
4. Client Internal / Enterprise AI Environment
Aubrey may also be integrated with a client-controlled internal or enterprise AI environment, where technically feasible. This could include environments based on Microsoft enterprise AI services, Azure-hosted models, Microsoft Copilot-related infrastructure, or other internally managed AI platforms.
Under this model:
- AI processing may occur within infrastructure or services controlled by the client.
- Authentication, access controls, data retention, model configuration, and other policies may be governed by the client’s enterprise environment.
- Integration requirements will depend on the APIs and capabilities made available by the client.
- Development and integration work will be scoped and quoted as required.
- Some Aubrey functionality may need to be adapted depending on the capabilities and restrictions of the client’s AI environment.
5. No AI/LLM integration
Aubrey and other usage of LLM/AI in the Bluebird line of products is optional, it does not have to be used. There are some usage scenarios where it is off by default.
Summary
These options allow organizations to select an AI architecture that aligns with their internal security, privacy, compliance, procurement, and technology requirements.
OpenAI Integration
Where either the Bluebird-Managed OpenAI or Client-Managed OpenAI option is selected, Aubrey uses OpenAI APIs to provide AI-powered functionality throughout the platform.
These features may support conversational AI, email generation, lead enrichment, and lead analysis workflows.
The specific OpenAI model used may change over time as models are updated, deprecated, or selected for particular workloads.
OpenAI APIs Used
The platform may use two primary OpenAI API families.
Chat API
OpenAI Chat APIs may be used for structured, single-response generation tasks.
These features include:
- Temptations
- Lead Enrichment
- Email Generation
- Parsed single-message interactions
Responses API
The OpenAI Responses API may be used for conversational AI interactions involving back-and-forth exchanges.
This API powers:
- Aubrey’s conversational chat experiences
- Ongoing lead-based conversations and AI interactions
Data Storage
Responses sent to or returned from Aubrey may be stored within Bluebird’s database systems.
Examples of stored content may include:
- AI-generated conversations
- Lead enrichment outputs
- AI-generated summaries
- Suggested responses
Additionally:
- Conversations are currently scoped to a single lead.
- Aubrey does not retain awareness of conversations across multiple leads.
- Conversations may be overridden if multiple AI processes operate on the same lead simultaneously.
Data Sent to the AI Provider
The specific information transmitted to the selected AI provider depends on the feature being used and the client’s chosen AI deployment model.
The descriptions below reflect Aubrey’s standard functionality. Where a client-selected or client-internal AI engine is used, implementation and data flows may differ based on the agreed integration.
Temptations
The Temptations feature sends cleaned agent text to the AI engine.
Before transmission:
- Content begins with agent_text.
- Any content removed through lead transformers is excluded.
- Phone numbers and email addresses matching configured regex patterns are removed before sending.
The platform currently removes phone numbers and email addresses using multiple regular expression filters, including:
- General phone number detection
- North American phone number formats
- Email address detection
However:
- Names are not currently removed.
- Physical addresses are not currently removed.
- Geolocation information is not currently scrubbed from the lead data.
The feature may also use regex-based extraction to isolate specific portions of the lead data before sending it to the AI engine.
Each Temptation includes its own instruction set that defines how the AI should process the data.
Lead Enrichment
Lead Enrichment may send the following information to the AI engine:
- Lead text
- Lead source
- Lead subject line
- Custom enrichment instructions
Unlike Temptations, Lead Enrichment uses the original lead text before transformer processing. This allows the enrichment output to later be added to agent-facing text.
Example enrichment tasks include:
- Profanity detection
- Lead scoring
- Routing rule adjustments
Email Generation
The email generation feature may send the following information to the AI engine:
- Agent text
- Custom instruction sets
Agents may manually edit the text before generation to remove sensitive information.
Aubrey Conversational AI
Aubrey’s conversational interactions may send the following information to the selected AI engine:
- Agent name
- Base instruction set
- Additional instructions
- Supplemental training or reference data, including RAG content
- Current date
- Full agent text
- Follow-up status
- Conversation history associated with the lead
Aubrey’s agent_text is currently unfiltered.
Conversation context is limited to a single lead record. Aubrey does not currently maintain memory across separate lead conversations.
Default Platform Behaviour
By default, Bluebird does not automatically send lead data to an external AI engine.
Specific AI functionality must be enabled or invoked before AI processing occurs.
Examples include:
- Enabling AI functionality within settings
- Creating AI summaries
- Using the robot icon in the lead footer to generate AI-powered email responses
The AI email generation feature is enabled for agents by default, though an administrator can disable it.
OpenAI Data Sharing Settings
When using Bluebird-Managed OpenAI, Bluebird configures its OpenAI services to disable data sharing and model training where supported by the applicable OpenAI service and agreement.
The current Bluebird configuration includes:
- OpenAI training data sharing disabled
- OpenAI conversation sharing disabled
- Default OpenAI tools enabled where applicable for Aubrey workflows
Customer data submitted through Aubrey using Bluebird-managed OpenAI credentials is therefore not shared with OpenAI for public model training.
When a client uses Client-Managed OpenAI, the client’s own OpenAI account configuration, contractual terms, retention policies, and data controls apply.
When a Client-Selected AI Engine or Client Internal / Enterprise AI Environment is used, data processing, retention, and model-training policies are determined by that environment and the agreed implementation.
Summary
Aubrey is an optional AI layer within Bluebird. Organizations can choose the AI architecture that best aligns with their technology, security, privacy, and compliance requirements.
Available approaches include:
- Bluebird-Managed OpenAI — the default and fully managed option.
- Client-Managed OpenAI — using the client’s own OpenAI credentials and account.
- Client-Selected AI Engine — integration with another AI engine, with development scoped as required.
- Client Internal / Enterprise AI Environment — integration with compatible client-controlled AI infrastructure, with development scoped as required.
Depending on the enabled feature and selected AI architecture, data processed by Aubrey may include lead text, agent text, conversation history, agent names, instruction sets, follow-up status, and supplemental reference or training data.
Some workflows currently perform partial scrubbing of phone numbers and email addresses, while other workflows may transmit unfiltered lead content.
Bluebird’s default OpenAI configuration disables data sharing and model training. Clients requiring additional control over AI processing may instead use their own OpenAI account or work with Bluebird to integrate Aubrey with an approved AI environment.