
Why core banking programmes fail, and what the survivors did differently
The failure pattern is remarkably consistent across every abandoned core replacement we have been asked to review. It is almost never the software.
Priya Venkatesh
Response-time service levels reward acknowledging incidents quickly. They are indifferent to whether the same incident happens again next month.
Daniel Okoye
1 min read

A hospital network came to us with a support contract that had been met in full, every month, for two years. Response times were green. Resolution times were green. Their clinicians were experiencing the same nine priority-one incidents every month.
The contract was working exactly as written. It measured how quickly someone responded to a problem. It said nothing at all about whether the problem should still exist.
A provider paid to respond to incidents, and penalised only for responding slowly, has no commercial reason to eliminate the underlying causes. Fixing them permanently reduces ticket volume, which in a per-ticket or per-seat model reduces revenue. The incentives point in precisely the wrong direction.
Monitoring was rebuilt first, because a third of incidents were being reported by clinicians rather than detected. One hundred and twenty-four bespoke integration interfaces were consolidated onto a standards-based layer, removing a large share of the incident volume at source.
Twelve months later, priority-one incidents were down sixty-four per cent and clinical system availability had reached 99.99 per cent. The service level targets in the new contract were, if anything, less aggressive than the old ones. They were simply measuring the right thing.
Daniel Okoye
Chief Operating Officer
Writes on delivery models, managed services and recovering programmes that have gone wrong.

The failure pattern is remarkably consistent across every abandoned core replacement we have been asked to review. It is almost never the software.
Priya Venkatesh

Retrieval-augmented generation gets you to a convincing demo. Everything between that demo and a system a regulator will accept is the real work.
Marcus Lindqvist

Most cloud business cases are built at datacentre level and validated never. Here is the workload-level model we use instead.
Aisha Al-Mansoori
Our architects run working sessions on exactly these problems.