You Will Live in the Transition State Longer Than the To-Be. Budget the Middle.
Photo by Tsuyoshi Kozu on Unsplash
Every enterprise architecture deck has two columns. On the left, the as-is: a spaghetti diagram that everyone recognises and nobody defends. On the right, the to-be: clean domains, event backbone, composable front ends, a clean core. Between them is an arrow, usually labelled "18–24 months." That arrow is where the money goes, where the agility goes, and where the programme quietly dies. The transition state is not a gap in the roadmap. It is the architecture you will operate — often for three to five years — and it deserves to be designed, costed, and governed like a destination in its own right.
Coexistence is a funded architecture, whether you planned it or not
The economics are unambiguous. Research summarised in several 2025 modernization cost studies puts dual-run overspend at roughly 14% annually during migration transitions — duplicate hosting, overlapping licences, split support rotations, and two sets of integration contracts. A recent CIO.com piece on the cost of running two architectures makes the sharper point: if the legacy estate stays fully funded, the cost base never shifts, so the transformation never creates the capacity it was supposed to fund. You end up paying for agility twice and receiving it zero times.
The failure is not that coexistence happens — it always does. The failure is that it is treated as an accounting footnote rather than a design artefact. Nobody owns the interim integration layer. Nobody names the duplicated master data. Nobody sets a budget ceiling on the overlap. So the overlap becomes permanent, and the to-be becomes a slide that gets re-dated each planning cycle.
Design the middle as a first-class deliverable
A transition architecture that survives contact with delivery has a handful of non-negotiable properties:
- A named system of record per entity, per phase. Not "we'll sync both ways." One writer, one reader path, and an explicit date when the other side becomes read-only.
- A declared overlap budget. The run cost of carrying both estates, approved as a line item with an expiry. If the overlap outlives its budget, that is an escalation, not a variance.
- Reversibility boundaries. Which cutovers can be rolled back in an afternoon, and which are one-way doors that need rehearsal, freeze windows, and a named decision-maker.
- Interim contracts that are disposable by design. Transition interfaces should be built to be thrown away — versioned, documented as temporary, and excluded from the long-term reference architecture so they never acquire dependents.
- Telemetry on the seam. Reconciliation counts, latency between estates, and error volumes on the bridge. The seam is where incidents cluster; it should be the best-instrumented part of the landscape.
Two-speed IT is what happens when the middle is undesigned
The classic two-speed split — fast digital front end, slow core — is usually a retrofit explanation for an undesigned transition. Teams discover they cannot change the core at the cadence the business demands, so they draw a line around it and call the line a strategy. The more useful framing is that different components have different change cadences, and the transition architecture's job is to make those cadences safely decoupled rather than organisationally separated. Decoupling is an interface problem. Separation is a politics problem, and it compounds.
This is where architecture practice earns its budget. Gartner's recent work on minimum viable architecture pushes the same instinct: define the smallest set of constraints that let delivery teams move, then let the rest emerge. Applied to roadmaps, that means a to-be defined as a short list of invariants — ownership, contracts, retirement dates — rather than a fully specified end-state diagram that will be obsolete before the second release.
Make the roadmap a retirement schedule
The most honest reframe available to an enterprise architect is to stop presenting the roadmap as a build plan and present it as a decommissioning plan. Each wave is defined by what switches off, not by what switches on. Build plans are infinitely extensible; retirement schedules are not. They force prioritisation, they are auditable, and they align the finance conversation with the architecture conversation — because a system that is never retired keeps appearing in next year's run budget regardless of what the diagram says.
In large SAP and commerce estates especially, this discipline separates programmes that deliver capacity from those that deliver complexity. Our SAP services work almost always starts by mapping which legacy capabilities have a credible shutdown owner, and our broader architecture practice treats the overlap budget as a governance artefact rather than a finance surprise.
The next planning cycle will not reward the prettiest to-be. It will reward the team that can say, precisely, what the landscape looks like in month nine, what it costs to run, and what has already been turned off. Design the middle, and the destination mostly takes care of itself.
Frequently Asked Questions
How long should a transition state realistically last?
Long enough to retire something meaningful, short enough that dual-run costs do not consume the business case — typically 9 to 18 months per wave rather than one multi-year overlap. The practical test is whether each wave has a dated shutdown for at least one legacy capability. If it does not, the transition state has no natural end.
Who should own the transition architecture?
A named enterprise architect with authority over interim contracts and seam telemetry, paired with a finance owner for the overlap budget. Splitting the technical and cost ownership is what allows coexistence to drift unnoticed. The pairing creates a single escalation path when the overlap outlives its approved window.
Does this approach conflict with agile delivery?
No — it supports it. Agile teams move fastest when ownership, contracts, and retirement dates are unambiguous, because they can change their component without negotiating across the seam every sprint. The transition architecture supplies those constraints; the delivery teams choose everything inside them.

Comments (0)
No comments yet. Be the first to comment!