AI agentic startup Gradient Labs has raised $13 million in Series A funding.
Founded in 2023, Gradient Labs was co-founded by Dimitri Masin, Neal Lathia, and Danai Antoniou — early employees of UK challenger bank Monzo.
Gradient Labs assists regulated businesses, such as those in the financial sector, by eliminating repetitive tasks from the entire customer operations cycle, from frontline support to back-office processes.
Utilising automation, its AI agent consistently resolves up to 90 per cent of queries with a quality assurance pass rate of 98 per cent, outperforming human agents and alternative chatbots while reducing costs by 75 per cent.
I spoke to Gradient Labs CEO Dimitri Masin to learn more.
From Monzo to machine intelligence
Masin was one of the first 30 employees of the neobank Monzo, where he worked for around seven years and served as the first lead of data science, AI, and machine learning.
Over time, he built out the AI organisation across the bank.
Much of that work focused on customer operations, support, and financial crime — areas where data science could have the biggest impact.
He recalled:
"We spent years trying to automate these processes using traditional machine learning, aiming for 10 per cent efficiency gains here and there.
But when GPT-4 arrived in early 2023, it was clear that a major shift had become possible."
Suddenly, instead of incremental improvements, the team could aim to automate 70 to 80 per cent of repetitive manual tasks.
According to Masin, the common denominator in all of these tasks was SOPs — standard operating procedures that agents follow repeatedly.
They realised that if they could build an intelligent agent capable of following those SOPs, they could meaningfully transform customer operations in large organisations.
Inspired by a wave of transformation
Even earlier in his career, Masin worked at Google as a product analyst. He recalled:
"Back then, the big shift was mobile. All traffic was moving from the web to apps. At the time, I didn't grasp the magnitude of that shift—it took years to realise that changes like that only happen a few times in a career.
The internet. Mobile. And now AI."
"I'd already built and scaled teams, launched innovative products, and had a great network of people. It felt like the perfect moment to contribute something meaningful to this new wave."
That's how Gradient Labs was born.
Regulated sectors need a different kind of AI
Gradient Labs focuses on customer support, building a support agent that could follow SOPs and handle fairly complex tasks, with better results than current human teams.
From his time in banking, Masin knew how challenging it was to get companies to adopt new technology.
According to Masin, for many years, financial service companies have faced a dilemma: technology has made customers more demanding – they expect immediate service at any given time – yet the sector is risk-averse and heavily regulated, which makes it difficult to implement the innovations they have grown used to when interacting with businesses in other sectors.
Despite businesses paying as much as $13.50 per customer service interaction, meeting expectations remains challenging. Today, 66 per cent of customers expect a response to an inquiry within minutes, while one in three will walk away from a brand after a single bad experience. Subsequently, 87 per cent of service agents report high levels of job-related strain.
Why most AI tools disappoint in regulated industries
Masin asserts that most AI products only scratch the surface by providing simple frontline support, which accounts for just 25 per cent of total customer operations costs. Likewise, those operating in heavily regulated industries remain largely unsupported.
He asserts:
"Most AI customer support tools work fine for simple, repetitive queries. "What's my account balance?" "How do I reset my password?" The easy stuff that any chatbot can handle.
But in regulated industries, customer questions get complex fast."
Compliance requirements, nuanced policy interpretations, and multi-step processes that require understanding context and making judgment calls. This is where most automation tools break down and hand everything off to human agents.
Gradient Labs built its platform specifically for these complex scenarios.
"You've got to convince governance, compliance, and sometimes regulators.
However, if you can demonstrate better customer satisfaction, improved compliance, and achieve this at a fraction of the cost, then everyone wins — the customer, the regulator, and the company. That's what we aim for."
Competing with in-house teams — and winning
The company's biggest competitors are trying to build this in-house.
Masin notes that this started happening about 12 to 18 months ago.
"But no one's really nailed it yet—not at a high level of quality. Klarna was the most public example — they announced they'd replace human customer support with AI, but they rushed it. It backfired. The communication was poor, and as soon as people hear about jobs being replaced, they resist.
We've shown it can be done well, where all parties benefit."
Gradient Labs did something different from most software startups. Instead of launching something early and iterating with users, it spent 14 months building quietly before going live.
Masin explained,
"That's because we knew the problem extremely well from our Monzo days — we didn't need feedback to define the problem. The goal was always clear: beat human-level quality in customer support."
He admits that the first version wasn't good enough, — just a basic GPT integration. They wanted a product that could deliver almost guaranteed quality, especially for financial services, where a single PR slip could be a disaster.
So the team built obsessively. According to Masin, when Gradient Labs finally launched, the agent already performed better than most human support teams.
In one case, it even outperformed the client's top agent in customer satisfaction.
"That kind of result wouldn't have been possible if we'd launched too early.
We'd have been distracted by feature requests or pivoted to simpler use cases, such as e-commerce, where quality expectations are lower. We wanted to go deep and do one thing really, really well."
Masin believes that from a technical perspective, many companies treat this like a back-end system problem, thinking,' We've built systems before; this is just another one.'
"But AI doesn't work that way. LLMs are non-deterministic. You're dealing with fuzzy, ambiguous inputs, like how people talk. Customers write in one word: 'Payment.' Or 'Dispute.'"
He asserts that the biggest difference in quality comes from having the right AI engineers — people who will read through thousands of conversations, build mental models, and iterate constantly.
"We always ask, 'What would a human do in this situation?" That's our North Star. It's a grind, but it's necessary.
You can't treat it like traditional system design—it's prompt engineering, AI modelling, and human behaviour modelling."
The company has taken a different approach from building a wrapper on ChatGPT or Claude, which largely results in pattern matching rather than understanding, leading to poor customer service outcomes.
When Gradient Labs onboard a new client, it reads through historical conversations and builds a knowledge graph — a structured internal model of the company's information.
It makes the knowledge more discrete and navigable, not just a black-box set of fuzzy associations.
So when a customer says something vague, like "payment," our system knows it needs more information and asks a clarifying question. "Is this a card payment or a bank transfer?" Because it understands the company's operational structure.
According to Masin, that's what makes the agent more human-like.
"It understands intent, navigates ambiguity, and reasons before answering.
Behind the scenes, each reply involves 10 to 15 separate models — working in parallel to understand the question, retrieve knowledge, formulate the answer, and run validations."
A $10,000 challenge to the competition
Gradient Labs recently issued a bold challenge, offering $10,000 to any company that tests its platform head-to-head against any other customer support automation solution — if the other solution performs better.
To qualify as a "loss," the competing system must come within five percentage points of Gradient Labs' customer satisfaction (CSAT) scores and match or exceed its automation rate.
It's a smart way to woo potential users as "win or lose, you get a fully-featured trial period, a complete testing framework, detailed analytics throughout the process, and hands-on support with automating your customer support environment.
The company is currently focused on geographic market expansion and will soon add voice capabilities.
In the medium term, it aims to expand beyond customer support into back-office operations.
Masin detailed:
"For example, if someone reports a suspended account, there's usually a process behind that where someone reviews the account for flags. We want to automate that too.
We started with support because it's the most homogenous part of customer ops. Most companies use the same tooling — Zendesk, Intercom, Salesforce — which makes it easier to build something scalable.
In contrast, back-office systems are much more fragmented. Every company's stack is different, and each process might only represent a small slice of time.
"So we started with the big, uniform surface area—customer support—and we're now expanding from there."
Redpoint Ventures led the funding with participation from Localglobe, Puzzle Ventures, Liquid 2 Ventures, and Exceptional Capital.
Alex Bard, Managing Director at Redpoint Ventures, asserts that the team has an "exceptional founder-market fit, having worked together for years at Monzo, giving them the deep expertise needed to build the best-in-class product for a complex, highly regulated market.
"Just three months post-launch, they secured nine customers – including one of Europe's largest banks—and have since continued to demonstrate strong traction and momentum."
The capital will enable Gradient Labs to expand its technology, marketing, sales, onboarding, and customer success teams, and increase its investment in research and development.
Would you like to write the first comment?
Login to post comments