Artificial intelligence (AI) is advancing at an unprecedented pace, transforming industries and redefining the boundaries of what’s possible. This rapid evolution, while exciting, can also be overwhelming. The temptation for organizations to hastily adopt the latest AI technologies without fully considering essential safeguards is real and potentially risky.
However, it is prudent to take a step back and critically assess the AI solutions being evaluated. Is a vendor utilizing data responsibly during the training of their models? What measures are they implementing to prevent hallucinations and inaccuracies? Do these solutions keep the user at the center of their design philosophy?
At Bloomreach, we are deeply committed to building AI the right way. Our approach to AI development is grounded in responsibility, ethics, and a clear set of guiding principles. We understand that our customers need confidence in the solutions we provide, and we take that trust seriously. These are Bloomreach’s core AI development principles being used to guide our product and engineering teams:
AI is changing — quickly — and we’re using these principles to ensure we’re innovating in a way that’s responsible and ethical. Let’s take a closer look at our principles.
Transparency and Explainability
Modern generative models, especially large language models, often function as “black boxes” due to their complexity and the vast number of parameters they utilize. Essentially, tokens get put into a model, which then outputs a bunch of numbers. This opaqueness applies to all search solutions using generative AI, ours included. While it’s challenging to interpret every nuance within these models, we believe in providing as much transparency as possible even with these limitations in interpreting foundational models.
At Bloomreach, we are committed to explaining the processes behind our AI outputs to our users. We utilize data from a multitude of sources to determine search rankings and product recommendations. The encoder model — derived from complex mathematical computations — is just one component, and we’re able to show how that model’s output is used to make decisions. For example, we can show the relative weighting of the distance between query and product and the contributions of other signals like add-to-cart performance and query-dependent performance signals.
We make this information accessible to our users through tools like Ranking Diagnostics and AI Studio in our Discovery platform. These features offer insights into how our AI makes decisions, allowing users to understand and trust the results. Similarly, in our Engagement platform, the Autosegments feature provides full visibility into data clusters and attribute relationships. Users can see the composition of segments and adjust them based on their expertise and intuition.
Our principle is clear: be as transparent as possible, provide meaningful insights at levels that matter to our customers, and ultimately enhance the end-user experience.
Data Privacy and Security
Data privacy is a paramount concern in AI development, and we treat it with the utmost seriousness. At Bloomreach, we firmly believe that our customer’s data belongs to them. We protect this data through robust virtual walls, ensuring that information from one customer does not intermingle with another’s.
When we use customer data to train aggregate models, such as our embedding models, we anonymize it beforehand. This anonymized data contributes to improving our models without compromising individual privacy. Our generative AI applications, like Bloomreach Clarity, incorporate specialized guardrails that filter outputs to prevent the inclusion of any personally identifiable information (PII) before results are returned to end-users.
Compliance is integral to our operations. We adhere to regulations like GDPR and maintain stringent security standards, including SOC 2 compliance. Our advanced customer data solution features real-time tracking of individual customer consent for data storage and usage. This ensures that all data we handle is compliant from both user consent and regulatory perspectives.
Fairness and Non-Discrimination
We recognize that AI models can inadvertently perpetuate biases present in training data. For example, in Bloomreach Clarity, we’ve implemented rigorous measures to ensure fairness and non-discrimination in our AI outputs. Our models are trained to avoid biases related to gender, sexual orientation, socioeconomic status, and other sensitive attributes through a guardrails filtering layer.
User-Centric Design
User experience is at the core of our AI development philosophy. We design our solutions to empower users, providing interfaces and tools that are intuitive and adaptable to varying levels of expertise.
For example, our AI Studio allows merchandisers to harness powerful AI capabilities to optimize search rankings effortlessly. Users can leverage advanced AI without needing deep technical knowledge, yet the platform remains flexible enough for those who wish to customize signals or algorithms to align with specific goals and campaigns, as well as set relevance temperature ranges.
Our aim is to make AI accessible and actionable, enabling ecommerce professionals to achieve their objectives efficiently. We provide the tools to manipulate and guide the AI, ensuring that technology serves the user’s intent.
Continuous Improvement and Monitoring
AI technology is not static; it evolves rapidly, and so do we. At Bloomreach, we are committed to continuous improvement, regularly evaluating and integrating the latest advancements in AI to enhance our solutions as part of our development process.
We actively monitor the performance and relevance of our models, often utilizing AI tools to assess and refine our systems. Collaborations with industry leaders like Google and NVIDIA enable us to leverage cutting-edge technologies that we used to develop advanced capabilities like our hybrid vector search. We utilize our partner’s technologies in advanced foundational models, frameworks, and tools to continually maintain high quality and performance for our products.
Applying the AI Development Principles to What’s Next
We’re excited to be continually applying these AI development principles as we create and deliver more innovative features to our customers. The conversation around AI has transformed — even in the last year — and we’re always working toward developing AI responsibly and in a way that benefits both our customers and end users.
To stay on top of our latest AI features, be sure to watch our webinar, “Cutting Edge: Bloomreach’s AI Feature Showcase.”
