Bazaarvoice AI is the collective term for the AI solutions powering our artificial intelligence capabilities across the Bazaarvoice platform. The Bazaarvoice AI engines harness the power of the world's largest dataset of authentic shopper content: over 10 billion pieces of content and counting. This continuously expanding dataset enables Bazaarvoice AI to evolve and become more intelligent over time.
We've dedicated over a decade to pioneering AI/ML technology. Our continued investment in this technology shows our commitment to delivering measurable business impact, such as increased productivity, revenue growth, and actionable insights. Bazaarvoice AI-powered solutions take care of the more repetitive, time-consuming tasks so you can focus on the strategic, creative work that drives business results.
Our AI-powered features also benefit consumers. They improve the shopping experience by using billions of data points to give consumers relevant user-generated content (UGC) when they need it to make purchase decisions confidently.
Core technologies
The following terms and definitions will help you better understand the types of AI that make up the Bazaarvoice AI ecosystem.
- Artificial intelligence (AI): Methods and software that enable machines to perceive their environment and use learning and intelligence to take actions that maximize their chances of achieving defined goals.
- Generative AI (GenAI): AI systems that imitate content—such as text, images, and audio—by sampling from distributions represented in training data. We use GenAI to generate authentic review summaries and automatically caption photos.
- Machine learning (ML): A subset of AI where rules of a system are learned from example data. We use ML to moderate content.
- Natural language processing (NLP): Technology that enables machines to process and understand human language. We use NLP to extract insights and sentiment from customer reviews and questions.
AI principles and standards
We believe transparency is vital when creating AI product features and services. In addition to staying current with regulatory requirements worldwide and giving input to regulatory bodies where appropriate, we’re committed to the following transparency standards:
- Authenticity: We maintain the integrity of UGC by ensuring all reviews and feedback come from real people with genuine customer experiences. We don’t use AI to write reviews on behalf of consumers.
- Objectivity: We design our systems and processes to collect and provide unbiased, neutral feedback that reflects genuine consumer opinions.
- Continuous improvement: We systematically monitor our AI systems and incorporate operational insights to enhance the quality and reliability of our solutions.
- Privacy and security: Bazaarvoice is ISO 27001 certified. We protect all data through stringent security measures within our infrastructure, ensuring client and consumer information remains secure and confidential.
- Responsible development: We follow established AI governance frameworks, ensuring our machine learning models are developed, deployed, and maintained with appropriate oversight and accountability.
Machine learning operations at Bazaarvoice
Our AI and ML development approach incorporates industry-leading practices to ensure reliability, security, and performance.
We employ a structured experimentation process led by our team of seasoned AI veterans and leaders with 100 years of combined AI and ML expertise. We test some consumer-facing features with our Influenster community to gather insights and refine the capabilities before piloting with clients. We pilot test other features directly with selected clients and their consumers to gather valuable data before making them generally available.
The following sections outline key aspects of our ML operations framework that we leverage across our AI initiatives as appropriate for each use case.
Model development and maintenance
- Proprietary and carefully selected third-party technologies
- Model performance monitoring and optimization
- Evaluation cycles to ensure high accuracy standards
- Systematic model updates based on performance metrics
- Version control for model artifacts, training data, and configuration
- Reproducible model retraining pipelines
- Continuous monitoring for drift detection
- Comprehensive model documentation and reproducibility protocols
Quality assurance
- Comprehensive automated testing suites
- Rigorous offline and online performance monitoring
-
Systematic guardrails against AI hallucinations
Occurs when an AI system generates incorrect, misleading, or nonsensical information that it presents as accurate or factual.
- Manual expert validation of model results
Data security and privacy
- All processing conducted within our secure infrastructure
- Strict adherence to data privacy standards and regulations
- Exclusion of personal information in model training
- Regular compliance reviews and audits
For more information about our general data security and privacy policies, visit Bazaarvoice trust 
.
FAQs
The following are frequently asked questions we receive about Bazaarvoice's use of AI technology.
AI system development
Where is user input data hosted and stored?
Input data is hosted and stored in our secure cloud environment. Refer to our Privacy Policy for detailed information about how we collect, manage, and store data.
Is your generative AI product a shared or stand-alone instance?
Our generative AI services operate on shared infrastructure powered by a large language model in our secure cloud environment kept within AWS Bedrock. Our comprehensive data validation pipelines and versioning systems properly segregate and protect all client data.
How do you ensure that only the minimum necessary data is collected for AI processing?
Our data science team works closely with business leaders to identify the minimum viable data needed for accurate AI results. Collecting extraneous data creates unnecessary costs without improving model performance. We optimize efficiency and expenses by keeping datasets as small as possible while still delivering robust functionality.
Will my data be used to train AI models? If so, are these AI models specific to our contract with Bazaarvoice and our organization?
Bazaarvoice trains models with product usage data to improve client and consumer experiences. Moderation models, for example, are trained by UGC with multiple clients to improve the Bazaarvoice global moderation system. These models are not specific to your contract with Bazaarvoice or your organization. Data used in models that generate content, such as Review Summaries, are specific to your organization.
Will you use our data to train AI models for features and functionality unrelated to the products/services we have purchased or that we don't benefit from?
We use data from every client to train our authenticity models, which benefits all clients equally. The more data we use, the more accurately we can identify inauthentic content.
How do you ensure data integrity throughout an AI system lifecycle?
We maintain comprehensive data validation pipelines, versioning systems, and audit trails throughout the lifecycle of our AI systems.
How do you ensure the quality and reliability of your AI systems?
Our development process includes multiple review stages, automated testing, and continuous performance monitoring. All changes undergo thorough validation before deployment to ensure system reliability and output quality.
Who owns the generated output from the AI tools you use?
Bazaarvoice owns the output generated by our AI tools.
How will Bazaarvoice inform me about AI updates, improvements, and revisions?
We communicate updates and improvements to our AI systems through our standard product release and communication channels. We continuously monitor our AI systems' performance and implement updates following our established validation and deployment processes. Any significant changes to the service will be communicated through your account team and our product documentation.
Quality assurance
How do you monitor AI system performance?
We monitor all production systems with automated alerts for issues. AI performance is no exception; all production models are continuously monitored for issues such as drift and hallucinations.
What human interventions/checks are used to ensure the model functions effectively and as expected?
To ensure the accuracy and quality of production models, we conduct both automated and human evaluations, performed offline and online, and tailored to each model’s specific use case.
Have your models been assessed for the potential for bias?
Assessing models for bias requires detailed demographic information about consumers. We don't collect this information because it would violate our principle of least privilege and consumers' expectations of privacy. However, we do deploy guardrails that continuously monitor generative models' output for toxic, inappropriate, or harmful output.
Data security and privacy
What is the data classification of the information used as input to the generative AI product/service?
Our GenAI features, specifically review summaries, process UGC in the form of product reviews and associated metadata. This content is classified as non-personal business data and handled in accordance with our data protection policies and ISO 27001-certified security practices.
How is client data used and secured within your AI solution?
Refer to our Privacy Policy for detailed information about how we collect, manage, and store data.
How is data encrypted at rest and in transit when used by your AI systems?
Data is encrypted at rest using the Advanced Encryption Standard (AES) and in transit using Transport Layer Security (TLS) in alignment with leading practices and the ISO 27001 standard.
What access control mechanisms are in place to protect data used by AI systems?
In accordance with Bazaarvoice policies and standards, as well as the ISO 27001 standard, access to data used by AI systems is done using the principle of least privilege and leveraging advanced security solutions, such as single sign-on (SSO) and multi-factor authentication (MFA).
Is a login or input to the generative AI required?
Access to our AI features is integrated into our standard Bazaarvoice platform authentication system. A login is required for client access to the Bazaarvoice Portal. No separate login is required for the AI functionalities because they are seamlessly integrated into the products you've licensed.
What are your policies for data retention and deletion, specifically regarding data used for AI processing?
Client data, including that used for AI processing, is retained for the duration of the client’s relationship with Bazaarvoice. Shortly after a relationship ends, the client’s data is returned to them or securely deleted.
