Practical notes from the engineers who build agents, copilots and AI products.
Measure a modernisation on four layers: business outcomes it was funded for, operational load on the people using it, system health against stated service levels, and output quality for any AI in the loop. Capture every baseline before the first module moves, because you cannot reconstruct it afterwards.
An enterprise platform implementation runs in five phases: discovery that produces artefacts, configuration against a locked scope, integration and data migration rehearsed in parallel, a phased cutover, and hypercare. The order is not negotiable, because each phase produces the input the next one cannot start without.
Enterprise platform implementation services projects fail for five repeatable reasons: scope written as a wish list, data migrated before it was cleansed, integrations discovered late, a big-bang cutover with no rollback, and nobody owning adoption after go-live. Each is a decision, not bad luck.
Measure an enterprise platform implementation on three things: whether people use it instead of the workarounds, whether the process it supports got faster or more accurate, and whether it runs without unplanned intervention. Go-live date, modules delivered and training attendance measure the project, not the platform.
Under India's DPDP Act you remain the data fiduciary for every record in the platform, and your implementation partner is only a processor. Compliance is therefore architectural: residency you control, retention enforced as fields, masked migration data and an audit trail that survives cutover.
You implement custom enterprise software development in six phases: discovery and domain modelling, architecture and interface contracts, a vertical-slice build, data migration with reconciliation, phased cutover with rollback, and hypercare into a support contract. Each phase ends with something inspectable.
Custom enterprise software development fails in five recognisable ways: a scope that was never designed, integrations discovered late, a migration treated as a tail-end task, a cutover with no rehearsed rollback, and a system nobody internal owns. Each has an early symptom and a specific preventive decision.
You measure custom enterprise software development on four layers: delivery health, adoption, business outcome and running cost. Pick two or three numbers per layer before the first sprint, baseline them against the process the software replaces, then review them monthly. A build with no baseline can only be defended.
Compliance is a property of what you build, not of the word custom. A bespoke system meets India's DPDP Act only if consent, purpose limitation, retention, erasure, access control and breach reporting are designed in. The advantage of building it yourself is that each control sits where you put it.