Ten systems that don't talk, and one we can't replace
These arrive together. The spreadsheet in the middle exists because two systems never met, and the system nobody will touch is the reason one of them never will.
Is this an integration problem or a modernization problem?
Because integration gets treated as a technical problem between systems, when the thing that actually needs connecting is the work. Two teams reconcile the same figures because neither trusts the other’s source. Someone rekeys an order because the handoff was never designed, only assumed.
Usually not, and the instinct to replace is what turns a manageable problem into a multi-year programme. The system that everyone complains about is often doing its job; what has decayed is everything bolted around it — the exports, the reconciliations, the spreadsheet that has quietly become production infrastructure.
In practice they are one problem seen from two ends. The manual bridge exists because the old system cannot reach the new one; the old system survives because replacing it would break every bridge built around it. Fixing either in isolation tends to entrench the other.
This is where two elephants live at once — Silo, the disconnected architecture and the blindness to how work really crosses it, and Shadow, the duct tape holding legacy together plus the ungoverned AI quietly filling the gaps.
What usually goes wrong here?
- The integration is drawn between the systems on the diagram, not between the people who actually hand work to each other.
- A platform is bought to solve it, and the manual step survives, because the manual step was never the platform’s fault.
- The replacement is scoped before anyone has measured what the current system actually costs to run.
- AI-accelerated delivery ships fast and leaves code the team cannot maintain, so velocity is borrowed, not earned.
How do we deal with both at once?
Map the real flow
An organizational network analysis shows where work crosses team lines, and where it stalls waiting for a person rather than a system.
Automate around it first
The manual work surrounding an old system is usually where the cost is, and it can be removed without touching the system itself.
Bring the shadow into the light
The AI your people already use, inventoried and given a governed path — so the convenience survives and the exposure does not.
Replace only what earns it
When something genuinely has to go, it goes as a scoped piece with a measured outcome — not as a programme with a ribbon-cutting.
Sound familiar?
Start with an assessment — people, process and technology, looked at together.