Dialpad Ai Principles

Ai Principles

Dialpad has been building AI features since 2018. At the core of our AI team are scientists who care deeply about building tools that are part of a world we want to live in. As a diverse team within a global company, we know that business works best when it works for everyone, and that philosophy is reflected in our process and our products.

In order to keep our customers first, several years ago we created a cross-functional ethics committee full of team members who are passionate about the topic. The team provides internal thought leadership and guidance on best practices and how to apply them through Dialpad’s AI Ethics Principles. We are continuously reviewing existing guidelines, industry best practices, and post-mortems on public incidents and making improvements to our recommendations.

Here’s what we’re thinking about when we’re building the AI-powered features that make Dialpad the best place to communicate with your customers.

Fairness and Inclusiveness

Dialpad Ai is developed and used by a diverse team, and we recognise that no matter how accurate AI models are, they’re based on data collected from the real world, which creates the risk of including or propagating biases.

The principles of Fairness and Inclusiveness mean that Ai should work consistently and fairly across languages and cultures, so we do our best to design models that handle real world input while outputting a service that works equitably and inclusively for every user.

What it looks like in practice:

• Taking cultural differences into account when evaluating call sentiment — we understand that different cultures react and express themselves differently in speech, from apologies to swearing.
• Recognising the many factors of identity that are relevant to communications — such as gender, sex, dialect, ability, culture, and language — and designing with them in mind..

User-focused Benefit

We don’t just build Dialpad, we use it every day, meaning we think like customers and users, too. User-focused Benefit means we consider the larger social and economic picture when developing Dialpad Ai’s capabilities and building tools that are part of the future we want to live in. Our business is providing our customers secure, high-quality communication services and tools — not profiling users or or selling customer data.

What it looks like in practice:

• Understanding that Dialpad Ai features and models rely on the business conversations happening on Dialpad’s platform, and on the trust our customers put in us as the place these conversations happen. We never sell your information or data to third parties.
• Building to facilitate human performance, not replace it. Instead of taking over the tasks of decision making or judgment of performance, Dialpad Ai provides tools and context to help humans make better decisions, faster.
• Being realistic. We describe what our service does, and don’t promise features or benefits that our system cannot deliver.

Accuracy and Objectivity

The principles of Explainability, Fairness, and User-focused Benefit depend on a fundamental confidence that Dialpad Ai is based on accurate, reliable data. We take substantial measures to ensure that the data we use is accurate so that the outputs are as objective as possible.

What it looks like in practice:

• Starting with an industry-leading speech and transcription model and encouraging customers to tailor it so it works best for them, from a Company Dictionary for industry-specific key terms to consulting your company’s directory to make sure everyone’s name is spelled right.
• Curating broad training data to develop AI systems that are as representative as possible of different domains, industries, dialects, etc.

Security and Safety

Because Dialpad Ai’s features and models are only as good as the real-world data they are based on, the security and safety of customer data is fundamental to ethical data use. Dialpad’s information security program is therefore a critical part of its ethical AI principles, to ensure that both inputs and outputs are secure.

What it looks like in practice:

• Securely encrypting data in transit and at rest.
• Undergoing annual 3rd party audits of our security and privacy program, including SOC2 and ISO 27001, 27017, and 27018.
• Real-time redaction for sensitive PI

Privacy and Control

Whereas Security and Safety involve the steps Dialpad takes to ensure that your data is protected and available, Privacy involves the steps we undertake to respect the basic premise that your data remains yours.

What it looks like in practice:

• Customisable data retention as a standard feature to ensure that your data is kept for the right amount of time.
• Granular control of what company data is used to improve Dialpad’s Ai model, from individual users to the entire company.
• Careful anonymization of data used to develop our generative Ai features.

Explainability

We believe that trust requires understanding. We design our technology to be transparent about how your data is used and how Dialpad Ai turns conversations into insights. Dialpad’s Privacy Policy, Trust Page, and Help Centre explain many of the most common questions about our data management practices, and we go further by providing Dialpad Ai-specific resources to understand how inputs connect to outputs.

What it looks like in practice:

• Articles on how Dialpad Ai works
• Help Centre resources on Data Security & AI Language Models

Accountability

We are ultimately responsible for ensuring that Dialpad Ai adheres to these principles. Accountability means making commitments and accepting the increased responsibility of meeting them while expanding and improving our offerings.

What it looks like in practice:

• Publicly sharing our principles
• Massively responsive to feedback
• Growth mindset: remaining static is falling behind