The lessons from failed transformations are well documented. The practice that acts on them is not.
A public institution I worked with completed a post-mortem on a failed implementation. The main finding was accurate: invest earlier in change, while it is still cheap to make. The finding was filed. The next major initiative launched without it.
I have seen versions of this across the organizations I work with. The lesson gets written. The practice that would make it matter on the next initiative does not get built.
Authorization services make this pattern visible in its clearest form.
Licensing, permitting, certification, accreditation, approval, permission: these are the services built around a single question. Does this person, organization, or product meet the standard required to proceed? We call this category authorization services, and the experience of navigating one an authorization journey. A nurse applying to practice in a new province. A construction firm working through environmental permits. What they share is structure: an applicant working to understand, demonstrate, and meet a standard, and an organization working to uphold that standard without making the path harder than it has to be.
These services are being redesigned at scale, and AI has widened what institutions think is possible when human and digital actors work together: shorter journeys, better guidance, faster decisions, lower cost. Whether or not AI ends up as a significant element of a given solution, it has changed the scope of what is being attempted.
The redesign typically works through a new digital product that connects the actors in the service — applicants, assessors, reviewers, administrators — in more effective ways. That product does not replace those actors. It changes how they work and how they connect. Redesigning an authorization service this way means many parts of a complex organization changing how they do things they have always done: how applications are received and assessed, how eligibility is determined, how decisions are recorded and communicated.
Each of those parts has a workstream. Each workstream has a deliverable. What no workstream has is a deliverable for how its work connects to the work next to it.
This is where many service transformations fail. Not because any one team falls short of its deliverable, but because the experience the redesign is meant to create depends on many changed elements coming together in a way that no single team's plan guarantees. The handoff between teams is where one part's changing work has to meet another part's changing work, on shared assumptions that were often never made explicit. When those assumptions differ, the redesign does not come together as designed. It fails at the seams.
The seam has no owner. Each team can show what they built and demonstrate it meets spec. Nobody is accountable for the space between the specs. This is why failure here is invisible until it is expensive: every measure the organization tracks is a destination metric, read from launch forward. The window during the build — after the future-state design is complete and before the service goes live — is when seam problems are still cheap to fix. That window rarely gets an instrument pointed at it.
This is what the post-mortem lesson is actually about. Invest early, while change is cheap: not more investment in the workstreams, which are already funded and staffed, but in seeing what the workstreams cannot see from inside themselves.
The approach we built for this is called Implementation Coherence. It runs as a loop through the build, on a recurring cycle short enough to leave time to act on what it finds, and it is grounded in implementation science — the body of research, developed primarily in health systems, on what actually moves evidence-based interventions into consistent practice.¹
Readiness diagnostic. We conduct structured conversations at the handoffs between workstreams, with the people doing the actual building on each side. We are not asking for status updates. We are asking each side to describe the connection from their position, and surfacing where their accounts differ. In one authorization service transformation, fifteen interviews across ten workstreams mapped thirty handoffs. Eight carried live mismatches: two sides building toward incompatible assumptions about what the shared element was. Three handoffs had no owner on either side at all. The diagnostic makes the invisible seam visible.
Targeted action. Each finding from the diagnostic points to a specific intervention — not a list of recommendations, but an action per finding, with a named person responsible for it. The action is designed for what the finding actually shows: a dependency that needs both sides at the table, an ownership gap that needs to be assigned, an assumption that needs to be surfaced and agreed before the build moves further.
Early measurement. Not satisfaction, which tells you what happened after launch of the new service. Implementation outcomes: how consistently the new ways of working are being adopted across teams, how closely the solution in progress reflects the intent of the design, how well it fits the real conditions it will meet. These are available during the build, before the destination metric fires. Tracked on the same recurring cycle as the diagnostic, they reopen it: a finding in the outcomes step sends the conversation back to the seams.
De-risking transformation
What the diagnostic and action steps have produced, in the engagement above, is concrete: sponsor approval to restructure the programme's direction within four weeks of the synthesis, and a shared schematic of the initiative's dependencies adopted by the whole team as its orientation artifact.
AI has raised the ambition of authorization service transformations. More human and digital actors working in new configurations means more elements that need to change, and more seams between them. The convergence problem grows with the ambition. What AI has not changed is the nature of that problem. A transformation with more AI in it still requires every changed element to meet the others as intended. The tool does not hold the seam.
The pattern holds wherever a redesign requires many parts of a complex organization to each change something they own — wherever the changed elements have to come together as intended for the service to work.
Invest early, while change is cheap. Versions of that finding are in post-mortems at organizations attempting authorization service transformations right now. The lesson is not the problem. The missing piece is the practice that makes it matter — pointed at the seams, running during the build, before the destination metric tells you what it cost.
Mark Kuznicki
Mark is the kind of person you want around when things feel messy. He has a knack for bringing clarity to complexity, helping teams, leaders, and organizations navigate change in a way that makes them more sustainable and ready for the future.