Gradient Labs Pushes AI Customer Service Beyond Basic Queries
Gradient Labs has automated 75% of customer service workflows at a European bank with 10 million customers-far beyond the typical 15-25% of basic inquiries that most financial services companies handle with AI.
The company's co-founder and CEO Dimitri Masin said the difference comes down to ambition. Most AI customer service implementations stop at simple account inquiries. Gradient Labs built systems that handle specialist support and back-office processes: investment questions, claims, dispute resolution, and compliance checks.
The result: customer satisfaction scores higher than human support teams, with faster resolution and greater accuracy.
Why most AI implementations fall short
Masin worked at Monzo as vice president of Data, Data Science, Financial Crime, and Fraud before co-founding Gradient Labs with Danai Antoniou and Neal Lathia. He said the problem isn't that AI can't handle complex customer service-it's that most companies don't try.
Basic queries are straightforward to automate. A customer asks about their account balance or transaction history, and the system retrieves the answer. These interactions represent at most 25% of customer operations.
The remaining 75% involves multiple steps, competing interpretations, and context. A dispute resolution case might generate a dozen different questions asked in different ways. Each requires reviewing customer history, related transactions, and documented procedures.
Masin said the foundation of handling these cases is "clearly defining intent." This is where customers sense whether a system truly understands their question. Skip this step, and even sophisticated AI fails.
How Gradient Labs built a system that works at scale
The European bank started with a goal: move from handling 10% of support requests with AI to 75%.
Gradient Labs began by ingesting the bank's internal knowledge base of more than 1,200 articles covering different service aspects. Banking staff added additional notes not in those articles. The company then reviewed thousands of recorded customer interactions, extracting 700 reference points that provided real-world context.
The system needed to handle the bank's full product portfolio-savings, investments, pensions, current accounts, business accounts, and subscription tiers-from day one, without months of training or gradual ramp-up.
To ensure compliance, Gradient Labs applied finance-specific guardrails covering prompt injection detection, financial advice detection, sensitive information handling, and complaint flagging. Every response was screened in real-time, both before and after generation.
The bank required 95% accuracy. The system scored 98%.
The disclosure gap
Masin noted that 50% of companies deploying AI customer service don't tell customers they're talking to an AI agent. At nearly all of those companies, the AI agent has higher customer satisfaction scores than the best human support staff.
Older demographics remain skeptical of agentic customer service. Masin expects that resistance to fade as the technology improves and proves itself through faster, more accurate responses.
For customer support professionals, the shift is practical: AI for Customer Support and AI Agents & Automation are moving beyond handling routine questions. They're now handling the work that previously required specialist training and significant headcount.
Your membership also unlocks:
