How Gradient Labs Is Building Autonomous AI Customer Support for Regulated Industries

Interview with Dimitri Masin, CEO & Co-Founder at Gradient Labs

Dimitri Masin leads Gradient Labs, an AI startup focused on building autonomous customer support agents for regulated industries like financial services. Before founding Gradient Labs in 2023, he held senior roles at Monzo Bank and Google. Under his guidance, Gradient Labs reached £1 million in annual recurring revenue within just five months of launch. His goal is to create AI systems that deliver high performance while strictly adhering to regulatory compliance, enabling safe and scalable automation of complex customer operations.

What inspired you to start Gradient Labs after your success at Monzo?

At Monzo, customer support automation aimed for modest efficiency gains of around 10%. But in early 2023, the release of GPT-4 changed everything. It became possible to automate 70-80% of repetitive manual work autonomously. This shift motivated us to launch Gradient Labs. I've witnessed two major tech waves in my career: the mobile revolution and now AI. When you see a transformation this significant, you have to act. Our team recognized this moment as the right time.

What lessons from Monzo’s hypergrowth are you applying at Gradient Labs?

• Balance autonomy with direction: True autonomy means giving clear goals while allowing freedom in how to solve problems, not just “do whatever you want.”
• Top talent requires top pay: To keep top performers, compensation needs to match their value; otherwise, larger companies will recruit them away.
• Don’t reinvent foundational practices: Many companies try radical changes in structures or titles but often revert to traditional models. Focus energy where it matters instead.

How did you build an AI agent like Otto for regulated industries?

We took a different path by delaying release and spending 14 months perfecting Otto before launch. Regulated sectors require trust and high quality from day one. We weren’t building assistants but fully autonomous customer support agents. With our financial services background, we had clear benchmarks for quality, letting us iterate internally without relying on live customer feedback. This approach let us pivot quickly and deliver a superior product at launch.

How does Otto handle complex, multi-step, or high-risk workflows?

Otto operates based on SOPs (Standard Operating Procedures) written in plain English, much like instructions for human agents. Two core design choices help manage complexity:

• Limited tool exposure: Instead of overwhelming Otto with all available tools, each procedure exposes only a few relevant ones. For example, in a card replacement, Otto might access just 1-2 tools instead of 30. This sharpens accuracy by reducing choices.
• Chain-of-thought reasoning: Our system supports multiple processing steps between input and output, enabling deeper reasoning rather than just applying procedures directly.

What does “superhuman quality” mean in customer support, and how do you measure it?

Superhuman quality means outperforming humans in key areas:

• Comprehensive knowledge: AI can access all company information instantly, avoiding the common human issue of limited knowledge and constant escalation.
• Proactive information gathering: Unlike humans who ask clarifying questions first, AI reviews account info and flags before responding, speeding resolution.
• Consistent patience and quality: AI maintains high-quality, patient responses without pressure to rush, unlike human agents.

We primarily track this through customer satisfaction scores, regularly hitting 80%-90%, often exceeding human teams.

Why did Gradient Labs avoid relying on a single large language model provider?

Flexibility is key. Our biggest gains come from switching to the best available models as providers like OpenAI or Anthropic release improvements. This lets us optimize quality and cost dynamically. Our architecture supports multiple models, and eventually private open-source LLMs hosted on clients’ infrastructure, which is crucial for banks with strict deployment rules.

What are the main challenges in automating back-office processes with AI?

There are two categories:

• Simple processes: The technology exists, but integration is tough due to numerous unique backend systems in financial institutions.
• Complex processes: Tasks like fraud investigations require deep domain expertise that humans train for months. Transferring this knowledge to AI remains a difficult problem.

How does Gradient Labs balance AI speed with regulatory compliance?

Our AI takes more time to think through conversations carefully, checking if it understands the customer, provides accurate answers, detects vulnerability, or identifies complaints. This increases response times to about 15-20 seconds median, which is acceptable for financial firms given the improved quality and compliance. It’s still faster than human replies.

Do you see AI agents trusted for higher-stakes decision-making in finance?

Traditional AI has long been used for high-stakes decisions. The new opportunity is orchestration—coordinating the process rather than making the final call. For instance, AI can route documents, validate them, and trigger follow-ups. However, for high-risk decisions, explainability, bias prevention, and regulatory approvals remain major hurdles for large language models.

How will AI change customer experience in regulated sectors over the next 3–5 years?

• True omni-channel interaction: Seamlessly switch between chat, voice, and calls with the same AI agent.
• Adaptive UIs: Customers will use natural language commands to complete actions without navigating complex menus.
• Improved unit economics: Lower operational costs let banks serve more customers or reduce fees.
• Exceptional support at scale: AI enables personalized, high-quality support even as companies grow.
• Support becomes a valuable service: Customer support will shift from a costly necessity to an efficient, helpful experience.

For those interested in AI-driven customer support and automation, exploring relevant courses can accelerate your skills. Check out the latest AI courses on Complete AI Training.

To learn more about Gradient Labs and their work, visit their official site.