Why this matters now
Customers no longer benchmark their bank against other banks. They benchmark it against the apps they already use for everything else - parcel tracking, ride-hailing, food delivery - where answers are instant, available at 3 a.m., and don't require a phone call. A decade of digital conditioning, instant confirmations, clean interfaces, and predictable fulfilment has created a benchmark that crosses categories. Banking, travel, retail, healthcare and public services are now judged through the same expectation of ease.
For one of Europe's largest banking groups, operating across multiple Central and Eastern European markets and serving millions of retail and enterprise clients, this shift has a hard edge. Almost 20% of European consumers now use a direct bank or a neobank as their main financial service provider, and the migration is fastest in exactly the regions where the group competes hardest. A balance check that takes two minutes on a competitor's app and twelve minutes through a call center is not a service gap. It is an attrition signal.
The economics are equally direct. A traditional contact center handling balance inquiries, garnishment information, and basic loan questions burns expensive agent capacity on conversations that don't need a human at all. Meanwhile, on the corporate side, SME relationship managers spend hours explaining EU and national financing programs that are routinely updated, intricate to navigate, and where matching the right business to the right instrument is more about structured filtering than expert judgment.
The bank's goals were to:
- Build a true self-service digital hub for individuals: Give retail customers a single, always-on entry point for the high-frequency questions that don't require a banker
- Streamline customer onboarding: Remove friction from the first interaction, where drop-off is most expensive
- Surface EU and national financing programs for enterprises: Help SME clients find the right grant, guarantee, or loan instrument without a long advisory cycle
- Offload the contact center back office: Free human agents from repetitive request handling so they can focus on complex, high-value casework
The Challenge
The bank's contact center was operating under the kind of volume that makes incremental improvement insufficient. Routine requests - balance inquiries, garnishment status, basic loan information - were consuming a disproportionate share of agent time. The same questions, asked millions of times a year, across multiple channels and languages, with no automated first line to absorb them.
Three constraints made the problem harder than a simple "add a chatbot" framing would suggest:
- Retail volume at scale: With more than a million inbound conversations annually expected to land on any deployed solution, the system had to perform consistently from day one. A pilot that worked at 10,000 conversations but degraded at scale would have been worse than no pilot at all.
- Enterprise complexity on the funding side: Matching an SME to the right EU or national financing program - across Horizon Europe, COSME, EIB lines, and country-specific schemes - requires structured intake, not free-text search.
- Regulated environment, sensitive data: Account balances, card blocking, and garnishment information are not low-stakes use cases. The automation layer had to integrate cleanly with core banking and CRM systems, respect authentication boundaries, and produce audit-ready conversation logs - all without slowing down the customer flow.
The bank also faced the structural reality common to large incumbents. Banks in Europe and North America allocate up to 74% of their budgets to maintaining legacy systems, which naturally leaves very little room for innovation. Whatever was deployed had to sit cleanly on top of the existing stack, not require it to be rebuilt.
The Solution
The bank selected Druid AI to build ADA, a conversational AI agent serving both individuals and enterprises across web and messaging channels, available 24/7, integrated with the bank's core systems and CRM.
The deployment was structured around the bank's two distinct customer populations - retail and corporate - with a shared platform but differentiated skill sets. Rather than launching narrowly and expanding, the bank scoped ADA from the start to handle the full breadth of high-frequency requests on the retail side, while building the enterprise advisory capability in parallel.
The solution includes the following core capabilities:
- ADA for individuals: account, cards, and self-service
Handles account info, deposits, loans, card management, and card blocking - the highest-volume requests, resolved end-to-end with in-conversation authentication. - ADA for enterprises: financing program matching
Structured intake on company profile, sector, and project scope matches SMEs to the best-fit EU and national financing programs. Relationship managers engage only when it's time to structure the application. - Modular skill architecture: FAQ, MyAccount, Customer Onboarding
Reusable Druid skills can be improved or extended without rebuilding the agent, and the same components scale to new use cases as ADA expands. - Round-the-clock availability
24/7 coverage removes the office-hours constraint - critical for time-sensitive cases like card blocking, where every minute of delay is a liability.
Learnings
Volume justifies investment; specificity justifies design
The bank's deployment didn't succeed because conversational AI is a generic upgrade. It succeeded because the request mix was concentrated: three categories accounting for the vast majority of volume. Identifying that concentration up front - and designing ADA to handle those three categories exceptionally well rather than every category passably - is what produced the 89% resolution rate. Generic coverage would have produced generic results.
Self-service for individuals and advisory for enterprises are the same product, differently applied
The instinct in banking is to treat retail and corporate as separate technology problems. The ADA deployment shows they don't have to be. The same agent, the same skill framework, and the same integration backbone serve both populations - the difference is in the skills loaded and the data the agent has access to. For banks running parallel digitization tracks for retail and SME, this matters: one platform investment, two business cases.
Containment is the foundation; revenue enablement is the next layer
ADA's immediate value is defensive: 47 FTEs of saved effort, 89% containment, faster resolution. But the financing-program matching skill on the enterprise side points to where the real upside sits. Once an AI agent is the default entry point for high-volume conversations, it becomes the natural channel for proactive outreach, cross-sell, and program activation. The bank is positioned to convert ADA from a cost-reduction asset into a revenue-generation channel without a second platform investment.
The hardest part isn't the model; it's the integration
Conversational quality matters, but it isn't what made ADA work. What made it work was clean integration with core banking systems, CRM, authentication, and case management - so that when ADA answered a balance question, the answer was real-time and the audit trail was intact. Banks evaluating conversational AI should weight integration depth at least as heavily as model capability. A brilliant agent disconnected from the systems of record is a demo; a competent agent integrated end-to-end is a production system.
