Al-native customer operations for financial services
London, UK
- Tweets 227
- Following 126
- Followers 436
- Likes 112
We're at Fintech Meetup Europe in Lisbon from 6th to 8th October. 🇵🇹 Our CEO @dimitrimasin joins the day 1 panel "Replace or Reshape? Designing the Workforce of the Future” to talk about what actually changes for customer operations teams when AI agents take on the work.
We've also got meeting slots across the three days for banks, insurers, and fintechs looking at automating frontline and back-office work. Drop a comment or DM us to book one. 🤝
30
We spent last week in Lisbon for our first team offsite. 🙌
This was a great opportunity for our new joiners to meet the old crew, connect in person with the teams spread across time zones, and align on the exciting goals that will shape our strategy in 2027.
Highlights from the week included:
🦾 A hackathon, where small groups brainstormed innovative new capabilities for our agents or automated a process dragging down the day to day.
🏄 Surf class, tile painting, and a morning workout led by our very own Ashley Rosenthal.
📈 Long-term vision alignment sessions from leadership, paired with practical workshops to help teams hone in on their goals and priority projects.
🌊 Seaside walks, bonfires, and yoga.
We’re back in our regular offices feeling more connected as a team and fired up with new ideas to take us to 2027 and beyond. 🚀
1
2
289
Last week we hosted our first Gradient Collabs session in London with @yondercard!
Antony Atkins, Senior Escalations Manager at Yonder, sat down with Yulia Yan from our AI Delivery team to talk about Yonder’s two year journey running our AI agents in production, most recently to automate disputes.
A few things that stayed with us:
🧱 A safe rollout is a team sport. Ops, risk, and engineering had to agree what "good" looked like before anything went near a customer.
🧠 The mindset change is harder than the technology. Teams move from clearing a queue to supervising the work, and that takes time.
🤝 Start with the process you already know is broken. Not the one that's easiest to automate.
Then everyone sat down to eat, and the conversation carried on for another two hours. Thank you Antony and the Yonder team, and thank you to everyone who joined us around the table. 🧡
1
1
56
You can now find Gradient Labs listed as a supplier member of the Credit Union Executives Society (CUES). 🎉
Our team came together through relationships built over a decade in UK fintech, so we understand the importance of community and building something for people who trust you. 🤝
We also know our AI agents make the biggest difference to lean teams carrying more volume than headcount. At Zego, our AI agent scores a 77% customer satisfaction rating against 61% for their human agents, and we hold 90% CSAT on disputes at Yonder.
That's why we're confident we can help credit union member experience teams move faster, without losing the human touch that members actually notice.
If you're in the community, come say hi. 👋
1
4
79
Last week we had a full table at @DanielBoulud's Le Pavillon in New York. 🗽 We brought together operations, product, and technology leaders from banks, credit unions, lenders, and fintechs on the evening of @dimitrimasin's session at @finovate Fall.
The conversation: what does it actually take to run AI inside a regulated support team?
More of these coming. If you run customer operations at a bank, lender, or fintech, drop a comment to join us at our next one. 🍽️
1
23
Our CTO @neal_lathia took the 'AI in Action' stage at @FintechWeekLDN last week, alongside Sean Hederman, CTO at Zilch, for a panel on moving from copilots to agents.
Three things that came up in discussion:
🎯 Scope beats ambition. Agents earn trust on narrow, well-defined work long before they get near end-to-end ownership.
🛡️ Guardrails have to be deterministic. You can't rely on the model to police itself.
👥 Humans must stay meaningfully involved, doing real judgment work rather than approving whatever the agent suggests.
That last point: this is the philosophy that’s allowed our disputes automation with Yonder to thrive. The agent runs the full case, 80% arrive fully evidenced, and decisions that used to take a week+ now take a day on average.
Thanks to @fintechfringe for having us.
231
We’ve been named the #1 most talent-dense scale up in the UK by Nova. 🚀🇬🇧
Their Talent Density Score ranks every person currently working at a company on experience, education, international background, and languages, then averages that across the whole team.
What a fitting recognition as we head into our first global offsite this week. 🤗
Also, we're hiring! ✨ Check the link in the thread. ⬇️
1
301
It’s been a busy year for us deploying AI agents at banks and fintechs around the world. The biggest gap we’re seeing between our customers and others trying to scale? Horizontal agents largely struggle to reach over 60% resolution rates in financial services.
They’re built for frontline questions and handle them well, which is also why horizontal agents are so effective in non-regulated industries like retail. But the complex, regulated cases that sit behind the operation fall outside their scope.
And even where a horizontal agent reaches, the audit trail and safety guardrails are a manual build the buyer owns, rather than something that comes with the product. This adds deployment time, eats team bandwidth, and piles up the engineering work… which cancels out the productivity gain from having an AI agent in the first place.
A vertical AI agent carries that depth and expertise from day one and is uniquely equipped to fully automate case work that crosses into the back office, like disputes and underwriting. This is AI that actually clears the backlog. 🦾
1
1
31
Read more about horizontal vs vertical AI agents: gradient-labs.ai/guides/vert…
20
In his recent post for Weekly Fintech Digest, @linasbeliunas made the case that finance’s AI copilot era is over, and the future belongs to autonomous agents that handle entire workflows and hand off to each other like a team. 🤖
He’s right. And for our customers, this isn’t future state: it’s how they’ve already chosen to integrate AI agents, and it’s working.
The capability Linas flags as what’s next, agents that pass work to each other and improve like humans, is live for us today. Our frontline and back-office agents hold context and hand cases between themselves across disputes, collections, and KYC.
And a few weeks ago we launched Collaborate, where operators, engineers, and AI agents build, test, and improve the same agent as peers, with version control, evals, and continual learning built in.
Collaborate better empowers human teams and AI agents to work side by side. 🦾 Find out more: eu1.hubs.ly/H0y7mkd0
2
42
The team’s all here at Finovate Fall, ready to hear CEO @dimitrimasin speak on the main stage later this afternoon. 🙌
We’ll be in the city through the end of the week, so if you can’t make it to the event, feel free to reach out for a coffee or chat outside of the conference.
We’ll take all excuses to walk around Manhattan and see more of our ads on the street. 🤭 Have you spotted one yet?
42
📣 Gradient Labs is launching its first-ever Gradient Collabs series, where we sit down with one of our customers to hear their real-world insights from working with AI agents. 🤝 Our guest coming up next week is Antony Atkins at @yondercard, in conversation with our very own Yulia Yan. We’ll be hosting Antony and senior leaders across banking and fintech at an intimate, invite-only dinner in London, where we’ll dive into Yonder’s impressive work on disputes automation.
For a sneak peek on what they’ll talk about, check out Yonder’s case study: eu1.hubs.ly/H0xT0y90
Interested in joining us for this event? Leave us a comment ⬇️
1
249
🤔 Imagine you're chasing a refund. You message your bank to ask what might be causing a delay. The AI agent handling the chat tells you it's being processed, then closes the conversation. Three weeks later the money still hasn't landed, so you get back in touch, and your follow-up opens as a brand new ticket.
On the bank's dashboard, both of these conversations count as wins: two tickets left the support queue without a human getting involved. This is called deflection, and a lot of teams grade their AI agent’s effectiveness by measuring deflection rate. At Gradient Labs, we know it isn't a true indicator of success.
💡 Deflection tells you a chat ended, but it doesn't tell you the customer got what they needed. Worse, every time an unsolved problem comes back as a fresh ticket, the deflection rate looks better, but the customer gets more frustrated.
So how do you know if your AI agent is actually solving your customers' problems? You look at the resolution rate. 👈
Resolution is a much harder number, but it’s an honest one. Push to automate resolution rather than deflection, and you'll move past the 60–65% ceiling that stalls teams focused on deflection alone.
Want to understand the difference properly? We wrote a full breakdown at the link in the thread. ⬇️
1
54
When you dispute a charge on your card, the worst part is the wait. In most cases, it takes at least a week to find out if you’ll get your money back.
@yondercard, the UK fintech behind lifestyle reward cards, has ended that wait with Gradient Labs. Dispute cycles that used to take a week now close in 1 to 2 days, even during volume spikes.
Because our frontline and back-office agents share the full context of a case, 80% of disputes arrive for human review fully evidenced the first time. This empowers faster decisions for human teams who start with all the information up front, as opposed to having to chase context with a customer days later.
What do Yonder’s members think about the new system? Customer satisfaction ratings have jumped to 90%. 🧡
Read how Yonder did it: eu1.hubs.ly/H0xJxbd0
1
60
Our team spent a few days in Atlanta last week at Fintech South, where we also teed up for some Topgolf in between conversations with senior leaders in banking, payments, and fintech. ⛳
Our biggest swing of the conference? Explaining that our AI agents have already sailed far past the pilot stage and into deployment at scale for banks and fintechs worldwide. Disputes? Automating 150% faster. CSAT? Outpacing human teams. 🦾 All par for the course over here.
2
47
Last week, we launched Collaborate. Now, anyone can improve our AI agents like code, no matter where they work.
Dive deeper into the launch on our blog, co-authored by our CTO @neal_lathia and Head of Marketing @emmalynefurlong here: gradient-labs.ai/blog/introd…
1
2
419
See you real soon, NYC. 🗽 We’ll be at Finovate Fall, and our CEO @dimitrimasin will be on the mainstage for a panel with Royal Bank of Canada, Prudential, and the University of Michigan Credit Union.
The question on the table: as agentic AI moves past live pilots, where are financial institutions actually making or saving money? 💰
If you’re in NYC the week of September 9th, drop us a line! Our team on the ground would love to connect.
1
1
92