Author: Domagoj Markovina
Accenture’s 2026 banking research introduces a concept worth paying attention to: the “10× bank.” The idea is straightforward – small teams managing AI agents to deliver ten times the impact – something that previously required entire departments. Capacity no longer depends on headcount.
This is not yet another fascinating forecast – Goldman Sachs already has autonomous agents handling trade accounting and client onboarding. Lloyds Banking Group expects AI to add £100 million in value this year through automated fraud investigations and complaint handling. JPMorgan runs hundreds of AI models enterprise-wide, with roughly 150,000 employees using large language models every week.
The direction is clear. By the end of 2027, Gartner projects that multi-agent systems will independently drive 30% of all day-to-day banking decisions. And that is not about answering questions but making decisions such as approving loans, managing compliance workflows or executing customer requests.
And this is where it gets interesting for the rest of the industry.
The gap isn’t AI capability. It’s the layer underneath.
The banks I mentioned have something most mid-sized institutions don’t: massive engineering teams and budgets to build custom AI infrastructure from scratch. JPMorgan alone employs thousands. Goldman Sachs can afford to develop bespoke agent systems on top of Anthropic’s models.
Most banks can’t do that – and they don’t need to.
The real difference between banks that are scaling AI and those stuck in AI pilots is not in the quality of their AI models. It’s in the platform they’re running the models on.
Ninety-five per cent of generative AI implementations in financial services remain in pilot phases rather than running live. Only four of the fifty largest global banks reported actual ROI from AI use cases last year.
That’s a massive gap between ambition and execution. And the reason is almost always the same: banks are trying to bolt AI onto legacy infrastructure that wasn’t designed for it.
What the 10× bank actually requires?
If you strip away the marketing language, the “10× bank” model depends on three things working together. Miss any one of them and you’re back to running expensive pilots that never reach production.
A unified agentic platform. Not a collection of point solutions, but a single environment where you can build, deploy, orchestrate, and govern AI agents across your operations. Most banks today have a chatbot from one vendor, a fraud detection model from another, an ML scoring tool from a third, and none of them talk to each other. Each requires its own compliance review, its own data pipeline, its own maintenance. That fragmentation is exactly why pilots can’t easily scale.
CRM-connected agents that can actually act. An AI agent is only as useful as the data it can access and the actions it can take. An agent that can read your CRM but can’t update a customer record, trigger a workflow, or send a personalised offer isn’t an agent – it’s a smart search engine. The agents that deliver real value are the ones embedded in your operational systems, who work with live customer data and execute real business processes. When a customer contacts your bank about a mortgage renewal, the agent should be able to pull their full history, check their eligibility, generate a personalised offer, and route it for approval – all within the same conversation.
AI governance that scales with autonomy. This is the part that most of us underestimate. When you have one AI pilot running in a sandbox, governance is manageable. When you have dozens of agents operating across customer service, lending, compliance, and sales, you need governance built into the platform – not added on top as an afterthought.
That means:
- Human-in-the-loop controls where you define exactly how much autonomy each agent has
- Audit trails that log every decision, every action, every escalation
- PII protection that’s automatic, not reliant on someone remembering to configure it
- Quality thresholds that prevent an agent from sending a customer response unless it meets a confidence standard
- Budget controls that stop AI costs from spiralling without anyone noticing.
The EU AI Act makes this non-optional. While the Omnibus regulation shifted the high-risk compliance deadline to December 2027, the transparency obligations under Article 50 are already enforceable. Banks need to disclose when customers are interacting with AI. And the governance infrastructure – system inventories, risk classifications, decision traceability, conformity assessments – takes time to build properly.
The mid-sized bank advantage
There’s a bit of a catch here. The largest banks have the resources to build custom AI infrastructure, but they’re also weighed down by the complex legacy systems. A major bank might run hundreds of applications across dozens of platforms, some dating back to the 90s. Getting AI agents to work across that landscape is a rather tall order.
Mid-sized banks have less technical debt. They can adopt modern, AI-native platforms that already integrate CRM, workflow automation, and agent governance in a single environment. They can move faster precisely because they have less luggage to haul around..
The AI talent shortage actually reinforces this. Mid-sized banks can’t compete with big tech for AI engineers – and they don’t have to. Platforms that let business users build and deploy AI agents using natural language and no-code tools change the equation entirely. Your compliance team builds a KYC agent, your marketing team builds a personalised outreach agent, and your service team builds a complaint resolution agent — all without writing code or waiting for IT.
What to do about it?
If you’re running a mid-sized bank and the “10× bank” concept sounds appealing but unreachable, here’s the how to go about it:
You don’t need to replicate what Goldman Sachs is doing. You need a platform that gives you the same capabilities – agentic AI, CRM integration, multi-channel deployment, built-in governance – without requiring massive engineering team to build and maintain it.
The practical steps are:
- Audit your current AI landscape. How many pilots are running? How many are in production? What’s the total cost of maintaining them separately?
- Evaluate whether your existing CRM and workflow platform supports native AI agent deployment – or whether you’re bolting AI onto a system that wasn’t designed for it.
- Assess your governance readiness. Can you trace every AI-driven decision? Can you control agent autonomy per function? Are you ready for the EU AI Act requirements coming in 2027?
- And start with one high-value use case. Customer service auto-reply with quality thresholds. Personalised product recommendations triggered by real behavioural data. Automated lead qualification that hands sales teams qualified opportunities instead of raw lists.
The 10× bank isn’t about doing ten times more with AI. It’s about building the operating layer that makes AI useful, governed, and scalable – and then letting your teams multiply their impact from there.
Talk to Our Team
