A strong answer. “I would separate what has to be true in ninety days from the end state, because the failure mode here is a two-year consolidation programme that stalls at eighteen months with two of everything permanently.

Ninety days — make it survivable. Federate rather than migrate: establish trust between the two IdPs so people can reach what they need with their existing credential. That buys time without moving anyone's identity. In parallel, get visibility — one place that can answer who exists across both estates, even if it is read-only and reconciled nightly, because the auditor's question arrives before the migration finishes.

Then sequence by risk, not by convenience. Privileged accounts first, because that is where the acquisition risk actually sits and it is a small population. Then shared applications. Then the long tail, which is mostly the acquired company's own tools and can wait.

Pick one system of record early and hold it. The specific failure I would design against is ending up with two places where access is granted, because then neither can answer who has what, and that state is remarkably durable once it exists.

And decide explicitly what we are not consolidating. If they have a specialist application with its own identity model that serves fifty people, migrating it costs more than it saves. Naming that upfront is what keeps the programme finishable.

I would also expect to lose people during this, so I would want the leaver process unified early — an acquisition with two leaver processes reliably produces orphaned access.”