Impact
Faster analysis of exceptions
Senior consultants search less, explain more by drastically reducing manual analysis work.
From symptom to cause
Practical AI evidence
Keylane shows that AI adds value in a business-critical implementation process.
More growth capacity
Keylane can grow faster without scarcity of expertise growing with it.
About
Being ahead of the curve with AI sounds good. Until you have to prove it in an industry where one forgotten exception can later hit someone’s pension. Keylane is the backbone of the insurance and pension industry in Europe. They deliver robust, configurable SaaS platforms that optimize business processes and keep customers agile.
“If you’re not an AI adopter now, the market is overtaking you,” says Wouter Boeser, Principal Director of Consultants at Keylane. But Wouter is averse to AI for the stage. His ambition was clear: “I wanted to demonstrate in practice that AI can make us grow more efficiently, without using the customer as a guinea pig.”
In Plexus, Keylane’s pension platform, the movements in the industry come together in very concrete terms. To people, pensions often seem like something that just keeps going on in the background. Until their administrator merges, their scheme changes or the WTP (Wet Toekomst Pensioenen) moves existing rights to a new contract. Consolidation means that more and more pension administrations, schemes and participant data need to be brought together on one future-proof platform. At the same time, the WTP requires a careful conversion to a new pension system. For Keylane and its customers, this makes migration not a technical sideshow, but a structural core task.
“With Team Rockstars IT, and Mathijs specifically, you don’t just bring in AI knowledge. You bring in seniority. And especially in a field where everyone is still discovering, that makes all the difference.”
TO MIGRATE IS TO SEARCH FOR TRACES
Each trial migration toward Plexus produces XML failure reports. On paper, it sounds simple: Issue discovered, cause determined, resolved, move on. Reality behaves differently.
An outage rarely points to itself; the real cause lies in a configuration choice from years ago, a missing field, code change that no one remembers why it was made at the time. Senior implementation consultants dig through XML files, GitLab commits and Jira tickets, looking for traces that don’t illuminate themselves. Sometimes that sleuthing takes an hour. Sometimes a week. Behind every record is a human being counting on his retirement benefits. That forces thoroughness. And thoroughness takes time.
The time and expertise used by consultants to do this is scarce and therefore costly. The problem is that for every new outage, you are performing the same detective work all over again, when the solution already exists somewhere. In a ticket. In a commit. In the mind of a colleague who saw exactly this last week.
A system that thinks for itself
“Keylane gave room to explore, but with a clear question: how do we set this up securely and scalably? That combination of trust and sharpness made the project interesting,” says Mathijs van Bree, Principal Consultant AI, at Team Rockstars IT
Keylane and Team Rockstars examined the migration process in its entirety: where were the bottlenecks, how could the process be accelerated? It quickly became clear that one language model could not get the job done. “Issues arise because of exceptions, which are always different,” says Mathijs. The solution required autonomy: deciding which data source was relevant to which report. For example, a disability report could be traced back to a missing Social Security wage. Consultants had come up with that too. Only considerably later.
Together, they developed an Agentic AI Proof of Concept for migration analysis. The system mimics the reasoning path of an experienced implementation consultant: understand the issue, determine which sources are relevant, consult them in the right order, present a probable cause with source references. Mathijs decided to hook up Rockstar and Cloud Engineer, Kris de Groot. Kris was already involved in Keylane’s cloud infrastructure, “Working with the cloud team allowed the PoC to technically land within the existing environment.”
The PoC runs in an isolated Azure environment. XML failure reports are uploaded; the system does not write anything back to source systems. Through a FastAPI backend, LangChain for agentic orchestration and semantic search in a vector database, it makes connections that a consultant would otherwise have to make on his own. The microservice architecture makes components easily replaceable later. “For an AI solution, that’s extra valuable because you’re still experimenting more often,” Mathijs says. The choices made are always recorded. When people build on later, they understand why certain choices were made.
What changes when knowledge no longer evaporates
The payoff is not in a spectacular claim that migrations are suddenly twice as fast. They are too big for that, too dependent on customer context, legislation and tiered data processing. The profit is more focused: where senior expertise is most expensive and scarce.
- Faster exception analysis: consultants are expected to get from error message to explanation 50% faster.
- Less costly search time: hours or days of detective work are reduced to focused analysis.
- Senior expertise better utilized: consultants spend less time reconstructing and more assessing.
- Historical knowledge remains available: insights from tickets, fixes and commits are less likely to disappear into systems or heads.
- More growth capacity and customer focus: Keylane can better serve migration requests, from customers and internally, without further burdening the same small group of specialists.
Equally important: Keylane now has internal proof that AI works in a serious, business-critical process. Not in customer service, where privacy, reputation and ethics are immediately on edge, but within a controlled implementation context with anonymized data.
Finally, about the impact, Wouter Boeser, Principal Director of Consultants at Keylane, says, “Our success here was also the success of Team Rockstars. Mathijs didn’t think in checkpoints, but in what was needed to make the whole thing successful.”
Collaboration
“Team Rockstars knew much more about the pragmatic application of AI than we did. Not the presentation version of AI, but the version that really puts you to work in a project,” said Wouter Boeser, Principal Director of Consultants at Keylane. “You bring in AI knowledge and expertise, but also seniority.” That seniority was in technical choices, but also in the ability to maintain momentum without overrunning the organization.
Wouter saw that especially in the way they work: “They are not on the hourly bill, but on contributing to the development I need in my team. That is a substantially different way of working together,” says Wouter. “Team Rockstars stands out for me head and shoulders above the rest in energy, benevolence and openness. There is a mentality that permeates the organization. That’s a pleasure to work with.”
The method of collaboration was also decisive for Mathijs: “I sat next to the domain team. People who could explain exactly why a pension fund is complicated in a certain way. Precisely because of that we were able to make the solution practical,” Mathijs says. “What is strong about Wouter is that he really wanted to be advised. Not to get his own idea confirmed, but to sharpen together where AI could actually deliver value.”
Keylane now knows that AI works here. The question is no longer whether it can. The question is which bottleneck they will address next.
“Your Rockstar helped us not only with the technology, but also with the question: when is this proof of concept actually successful? That focus in the plan of action helped us tremendously.”
Want to know more?
Please fill out the form below and a colleague will contact you as soon as possible.
Contact
"*" indicates required fields
