Customers have learned to expect that service should work like a conversation: describe the problem in plain language, get it resolved. Tools like ChatGPT and Gemini set that bar, and most contact centers fall short of it. The obvious fix is to add AI, yet MIT's NANDA initiative found that around 95 percent of enterprise generative AI pilots never reach measurable value. So the question worth asking is not whether to put AI into customer service. It is why so many attempts stall and what the few that work do differently. That challenge is sharpest in regulated industries, which is why building an AI contact center for financial services is one of the hardest versions of this problem to get right and one of the most instructive.
That was the question behind a live session I hosted with Tyler Pichach, who leads banking and payments strategy at Microsoft and Tom Arbuthnot, an independent Microsoft Teams specialist, Microsoft MVP and co-founder of Empowering.Cloud. We built the webinar for CX leaders in financial services, because that is where the stakes run highest: regulated advice, fraud, compliance obligations, and customers who expect a private bank's level of attention over the phone. But most of what we landed on holds well beyond banking and insurance. Here is what came out of the conversation, and what I keep hearing from CX leaders across the UK, Benelux and North America.
In the interest of being straight with you: I work for Luware, and Luware Nimbus is our platform built for this. I will get to how it works, but the argument does not rest on us. The reasons AI projects succeed or stall are much the same whoever you buy from, and they are worth understanding on their own terms.
Most AI pilots fail to scale because the technology is layered onto workflows that were already broken, not because the models themselves fall short. MIT's NANDA initiative, in its 2025 report “The GenAI Divide: State of AI in Business,” found that around 95 percent of enterprise generative AI pilots never delivered measurable business value, and only about 5 percent reached real production impact. MIT described the cause as organizational rather than technical: a learning gap, meaning the inability to fit AI into daily operations.
That finding lands harder in financial services than almost anywhere else. MIT also found that buying from specialized vendors and building partnerships succeeded far more often than internal builds, and it flagged this as especially relevant in regulated sectors, where many institutions default to building their own systems. Tyler made the same point from the banking side. As he put it during the session:
His advice was a deliberate progression rather than a leap. Prove value internally first, with employee assistants and back-office automation, which in banking means workflows like anti-money-laundering routing and document gathering, building the governance and organizational readiness that customer-facing AI depends on. Then introduce customer-facing assistants on owned channels such as the website and app, starting with a narrow set of journeys that prove measurable results, before extending further. The banks that have taken this path show what it produces: Commerzbank's AI assistant reportedly resolves around 75 percent of customer conversations on its own, and ABN AMRO's automates more than half of its millions of interactions a year.
The honest version of this is not “AI fixes customer service.” It is closer to this: AI delivers value in a bank or insurer only when it is embedded in daily operations, governed properly, and connected to the tools people already work in. Get that foundation wrong and you join the 95 percent.
The biggest change is that customer service stops being a separate island, which in financial services is also a governance problem. Most institutions already run on Microsoft Teams, with more than a million organizations using it, their data sits in the Microsoft cloud, and their people are hybrid. Yet many still run customer experience on infrastructure built for a different era. The result is familiar: agents switching between disconnected systems, customers repeating themselves across channels, and customer data scattered across vendors that each need their own controls and audits.
Luware Nimbus addresses this by turning Microsoft Teams into a single customer experience hub. It connects the contact center, reception, branch offices, and the internal experts a query often needs to reach, all in one place, so an agent works in the same interface they use all day. Because Luware Nimbus is a Microsoft-certified, Teams-native platform built on the Microsoft Extend model and hosted in Microsoft Azure, there is no separate console to learn and no external hardware to maintain. For a regulated firm, consolidating onto Teams also means fewer places where sensitive data lives, which is a quieter benefit that risk and compliance teams notice quickly.
This is also where cost and effort come down: one platform, one place to work, and one set of reporting and KPIs across voice, chat, and email.
Virtual User is Luware's AI self-service layer, and it replaces static IVR menus with an assistant that understands what a caller wants and acts on it. Instead of pressing 1 for balances and 2 for card services, a customer describes their issue in their own words and gets a resolution or a route to the right specialist with full context already attached. In a bank or insurer, that means a suspected-fraud call reaches a fraud-trained agent and a claims question reaches claims, without the customer working through a menu tree first.
This is the practical meaning of moving from the classic IVR to an AI foundation built on Microsoft Azure, which Microsoft now delivers through Azure AI Foundry. The phone tree was a blunt instrument; an intent-aware assistant is not. In my experience, the difference shows up quickly in two numbers leaders care about: containment, meaning the share of interactions resolved without an agent, and customer satisfaction, because people are no longer fighting a menu to get help with money matters that are often urgent.
The Luware Nimbus Companion gives contact center agents real-time AI assistance during live conversations, surfacing the right knowledge at the right moment: suggested responses, prompts during the call, and summaries afterward. In a regulated environment, day-to-day change matters because the agent stops switching between systems to find the right answer and gets consistent, current guidance instead.
Two effects come up consistently. New agents reach competence faster: Luware Nimbus customers report up to a 50 percent reduction in agent training time, in large part because people work in the familiar Teams interface rather than learning a separate tool. And quality becomes more consistent because every interaction draws on the same knowledge foundation rather than the individual memory of whoever happens to pick up. In financial services, where a wrong or inconsistent answer can carry regulatory weight, that consistency is not a nice-to-have.
The argument in a regulated bank is that the governance foundation for AI already exists: consent, identity, and audit trails are already in place. The shift is not building new controls; it is replacing static IVR with assistants that understand intent and act within those existing controls.
This was Tyler's strongest point for financial services, and it holds for insurance and wealth management too. Azure AI plugs into the governance a regulated firm already has rather than working around it. With Luware Nimbus, customer data stays inside the Microsoft environment and is transient by design, which is what makes the model viable for compliance-sensitive institutions. It is worth being precise: this does not remove a firm's compliance obligations. It means AI operates inside the same audited, controlled Microsoft 365 estate as the rest of the business, which is a very different proposition from sending data out to a separate AI system.
The trust signals back this up. Luware Nimbus holds Microsoft's Financial Services AI certification, which recognizes solutions that let regulated firms adopt current AI services while staying compliant, alongside SOC 2 Type II, ISO 27001, and Cloud Security Alliance STAR registration. On the recording and surveillance side that financial services depend on, Luware Recording provides compliance recording, archiving, and speech analytics across Teams and other channels, and Luware has been recognized repeatedly as Verint's EMEA Compliance Partner of the Year. The point is not the badge count. It is that the AI conversation and the compliance conversation can finally happen on one platform rather than two.
One of the most compelling applications of outbound AI in financial services is its ability to transform reactive customer service into proactive engagement. In the insurance sector, claim status inquiries represent a high volume of inbound calls, a repetitive, easily automated interaction that outbound AI virtual agents are well-placed to eliminate. Rather than waiting for a customer to call in, an AI virtual agent can proactively reach out the moment a claim progresses to a new stage, informing the customer directly, answering immediate questions, and even prompting next steps such as scheduling a callback with a human agent or flagging any outstanding documentation required. These outbound calls can be further enriched by automatically sending relevant links or supporting information via email or SMS in real time. The result is a clear win-win: customers are kept informed throughout their claims journey without having to chase updates, while inbound call volumes are meaningfully reduced. This frees contact center agents to focus their time and expertise on higher-value, more complex interactions, improving both operational efficiency and the overall customer experience.
Finance responds to containment rates and cost per interaction; risk responds to governance and auditability; operations respond to simplification. A business case in financial services has to speak to all three, because all three usually have to say yes.
On the finance side, the banks Tyler pointed to are resolving somewhere between 75 and 90 percent of contact center conversations autonomously, in line with the examples earlier. Those figures come from full agentic deployments, so I would treat the top of that range as an aspiration rather than a day-one assumption for any single rollout, since results depend heavily on interaction mix and how well the journey is designed. But the direction is real, and containment is the number a CFO sees first. Companion then adds agent productivity on top of it. For risk leadership, the argument is governance: data stays within the Microsoft environment and is auditable end to end.
The argument that lands with operations is simpler, and in my experience, it is often the deciding one: one platform inside Teams, one vendor, one governance framework. That consolidation removes a whole category of integration and vendor-management overhead that most CX and compliance teams have quietly learned to live with. As Tyler put it:
The operational side of this is visible with customers like SIX a global financial market infrastructure company, which standardized its customer service on Luware Nimbus inside Teams.
A move to a cloud contact center like Luware Nimbus starts with a “Redesigning the Customer Journey” workshop, a structured discovery session that maps the ideal end-to-end experience and produces a Customer Journey Blueprint. That blueprint becomes the foundation for a proof of concept.
The biggest blocker I hear is fear of disruption, and in financial services that caution is fair. The reassuring part is structural: because Luware Nimbus is native to Microsoft Teams and hosted in Azure, most of the infrastructure complexity that makes legacy migrations painful is simply not there. There are no third-party session border controllers to manage and no separate agent client to roll out. Institutions typically go live faster than they expect. I would still tell any leader to plan the journey design carefully, because that is where the value is won or lost, not in the technical cutover.
If you are weighing this up, the most useful first step is not a feature comparison. It is mapping your current customer journey honestly and deciding where AI would remove real friction rather than add a new layer, with your compliance requirements in the room from the start. That is exactly what the Redesigning the Customer Journey workshop is built to do.