The conventional wisdom goes something like this: wait for the new system to land, let the dust settle, then figure out where AI fits. It is understandable logic. It is also wrong — and the organizations discovering that are gaining a meaningful structural advantage over the ones still in the queue.
The Migration Paradox
Every large-scale ERP migration carries a paradox at its center. The entire purpose of the project is to build a better operational foundation — cleaner data, more integrated systems, faster reporting, better decision-making capability. And yet, in the 18 to 36 months it takes to get there, the organization is effectively frozen on any improvement initiative that touches the systems being replaced.
The teams carrying the operational burden keep running manual processes. The spreadsheets multiply. The workarounds calcify. And by the time the new system goes live, the gap between what the business needs and what the technology delivers is larger than when the project started. This is not a technology problem. It is a sequencing problem.
What Parallel-Track Organizations Are Doing
A growing number of enterprises are running a deliberate parallel track during their ERP migrations: continuing to invest in targeted automation that operates independently of the core system replacement, while simultaneously using that automation to generate the clean, structured data their new platform will need on day one.
The first benefit is immediate efficiency. Workflows that are automated during the migration period deliver ROI now, not in two years. Invoice processing, vendor claim management, document ingestion, approval routing — these do not need to wait for D365 or SAP to go live.
The second is data readiness. The single most common cause of ERP go-live delays is data quality. Automation deployed during the migration period does not just save manual effort; it creates clean, structured, validated data as a byproduct. The intake engine becomes the data cleansing engine.
The third is organizational change management. Organizations that have been running AI-assisted workflows for twelve months before go-live have a workforce already comfortable with automation as a partner. The cognitive leap at cutover is smaller. Adoption is faster.
The Last 10% Problem
Anyone who has led a major systems implementation knows the final stretch. The first 90% of the migration moves on a schedule. The last 10% — data validation, exception handling, the long tail of edge cases — is a grind that can stretch for months and carries the real implementation risk.
The organizations that have been cleaning data in parallel throughout the migration arrive at that final 10% with a fundamentally different risk profile. Their exceptions are smaller. Their validation cycles are faster. The go-live confidence is higher.
A Different Way to Sequence the Work
Rather than a single sequential path — replace the system, then optimize — the more effective model is two parallel tracks. Track one is the system migration. Track two is operational intelligence: the targeted automation and data work that delivers value now and feeds into track one at cutover. These tracks are not in competition. They are complementary. The best time to start was when the migration kicked off. The second best time is now.