
The CIAM Migration Diary · Part 1 of 6
The CIAM Migration Diary, Part 1. Two months into replacing a critical customer authentication platform at a Tier 1 bank, written while it is still happening.
This series is probably only useful if you are responsible for CIAM and you are frustrated with what you have today.
Over the last few years I have had versions of the same conversation with several organisations. They were unhappy with their existing CIAM platform, but for different reasons. Sometimes it was cost, and renewal conversations were becoming increasingly difficult to justify. Sometimes it was support or product direction. In other cases the concern was more fundamental: they had become heavily dependent on a single vendor for infrastructure that sits behind almost every digital customer interaction.
Eventually most of those conversations reached the same point. Even if there is a credible alternative, can you actually move?
That is a reasonable concern, because CIAM is not an isolated application. Over time, mobile applications, websites, APIs, gateways, fraud systems, authentication journeys and other services develop dependencies on it. Some are deliberately designed. Others simply evolve through years of implementation and change. After a platform has been running for long enough, I am not convinced anyone holds a complete picture of all of them. That is probably one of the main reasons organisations stay where they are. The business case for changing platform may exist, but the perceived migration risk overwhelms it.
What I find more interesting is what changes that calculation. The assumption is that staying is the safe option and moving is the risky one. That holds only while the incumbent remains a reasonable place to stand. At some point cost, strategic dependency and operational constraint start accumulating on the other side of the ledger, and the risk of staying stops being zero. Every organisation I have spoken to that eventually moved crossed that line at a different point, and none of them crossed it because of technology.
A few years ago we started looking seriously at whether that assumption still held. We evaluated the alternatives and concluded that Red Hat's build of Keycloak was credible enough for us to pursue for the large regulated organisations we work with. Enterprise support around an open source platform, extensibility and a commercial model very different from traditional proprietary CIAM all mattered.
But selecting an alternative on paper proves very little. The real question was whether we could replace a mature CIAM platform in a live banking estate without requiring everything around it to change as well.
This year we got the opportunity to find out.
We are now roughly two (and a half) months into a formal programme to replace a critical customer authentication platform at a Tier 1 bank. The incumbent is a global, enterprise grade, on premise IAM platform that has supported critical customer authentication there for years. I am not going to name it anywhere in this series. Almost nothing I have learned so far is specific to that vendor, and the useful conversation is about migration rather than about who made the thing being migrated from. Our target is roughly six months from formal programme start to production migration.
We are not finished.
I want to be explicit about that, because I would rather write this while we are doing it than reconstruct the story afterwards, when every decision looks more obvious and every problem appears to have had an orderly resolution. Some of the conclusions I reach over the next few months may change as the programme progresses.
From the beginning we imposed one major constraint on ourselves: changing the CIAM platform should not require the bank to change the systems that consume CIAM. If replacing CIAM means dozens of application teams have to modify their applications, reprioritise their roadmaps and coordinate releases around the migration, the problem stops being an identity programme and becomes something much larger. Our approach has therefore been to make the replacement platform adapt to the existing estate wherever reasonably possible, rather than asking the estate to adapt to the replacement platform.
Two months in, that principle has already taught me something about where migration complexity actually sits.
One example is the gateway layer. In this programme we are not only changing the CIAM platform, we are also moving parts of the surrounding infrastructure between cloud platforms. That makes this harder than a like for like CIAM replacement, and I will try to be clear throughout this series about which difficulties belong to which change. Gateways on different cloud platforms do not necessarily handle traffic in exactly the same way. Something as apparently small as different treatment of a header can alter an interface contract that an application has depended on for years. The identity platform may be behaving correctly and the application may be unchanged, yet the end to end journey still behaves differently, because something in between has changed. That is the kind of dependency that is easy to miss when you treat a migration as a product replacement rather than as a set of customer journeys.
We have also made mistakes ourselves.
Our initial migration approach during testing was essentially to move traffic to the new platform unless something had been explicitly identified as needing to remain on the incumbent. In hindsight that was the wrong default. Our knowledge of everything that should not yet move was incomplete, and the result was disruption in the test environments that we could have avoided. We should have started from the opposite position: keep traffic on the incumbent unless it has been explicitly proven ready to move. It sounds obvious now. It was not obvious enough when we designed the approach, and it is one of the changes I would make immediately if we started this programme again tomorrow.
The surprise in the other direction has been functional testing. I had budgeted for a difficult SIT. So far it has not been one, and the defect count is a fraction of what I would have predicted at the start, across a meaningful number of consuming applications and APIs. Much of that comes down to work done before formal integration testing began. During development we exercised the interfaces extensively at API level using Postman and proved the expected behaviour before handing journeys into wider testing. That is a statement about the waves we have completed, not about the ones ahead, and I will report it honestly if the pattern does not hold.
That distinction is becoming important to me. The difficult issues have rarely been "can the new CIAM platform authenticate the customer?" They have usually been "does the entire existing journey still behave exactly as everything around it expects?" Those are very different questions, and only one of them is a product question.
There have also been lessons with little to do with identity technology. The hardest part of this programme so far has not been the platform. It has been securing the time and attention of the teams that own the systems around it. An organisation of this scale runs many concurrent changes, and every consuming application team has delivery commitments that existed long before our programme did. The constraint has been availability rather than willingness. I have spent more of my own time on that than on anything technical, which is not what I expected when we started. We have held the plan, and that is largely down to active C level sponsorship and a steering committee that makes decisions rather than noting them, so that when something blocks delivery it gets escalated and resolved rather than acknowledged.
If you are planning a migration like this, that is the line item I would budget for first. Not the identity engineering. The consuming teams' time, agreed and protected before the programme starts.
So, two months into a six month target, I am not ready to tell anyone that replacing critical CIAM is easy. I am also less convinced than I was that the difficulty sits where most people assume it does.
What I cannot yet tell you is whether that holds as we move towards migration, cutover and production operation. Those are the stages where I expect our current assumptions to be tested properly, and where my view may change.
If you have been through a migration like this, I would like to know what surprised you, particularly the dependencies nobody had identified until you tried to move them.
This is Part 1 of six, published weekly while the programme runs. Part 2 is about the constraint we set ourselves and the places where it turned out to be imperfect. Subscribe if you want the rest as they are published.
The CIAM Migration Diary is Ajit Gupta's newsletter on LinkedIn. Subscribe on LinkedIn ↗

