Migrating from a legacy RPA platform to Primo — in weeks instead of quarters
How a major retail chain replaced its existing RPA platform on Primo, layered in agentic AI for exception handling, and kept 24/7 store operations running through the migration.
Retail · Emerging markets · Orchestrator · Studio · AI Server · 10-min readAbout the customer
A leading retail group in Eastern Europe operating a chain of supermarkets — several hundred stores, 30,000+ employees, multi-billion-dollar annual revenue, and an operations footprint that runs 24/7 across stores, distribution centres, and e-commerce.
The group's RPA programme had been in production on an incumbent RPA platform for several years, with 46 automated processes covering store back-office operations, supply chain, finance, and customer service workflows. By 2024, the programme had reached a point where the platform choice needed revisiting: workflows that should have scaled were taking too long to redesign, exception handling was eating developer time, and the licence economics were trending in the wrong direction relative to the value being produced.
What made the case for a new platform
Three problems compounded.
First, scalability. The existing platform had been adequate for the first 20 processes; by process 40, the team was spending more time working around platform limitations than building new automations. Specific workflows — particularly those involving high-frequency exceptions in supply chain — had been on the "we will come back to this" list for over a year.
Second, licence economics. As the programme scaled past 40 processes, the per-process and per-robot licence cost was growing faster than the value being produced. The CFO's office had asked the CoE to model alternative platforms.
The non-negotiables: zero disruption to store operations during migration, no loss of any of the 46 production automations, and a credible path to expanded AI use cases on the new platform.
How we ran the migration
Track 1 — Process migration
Each of the 46 processes was inventoried, dependency-mapped, and converted to Primo Studio. Primo's automated converter — a code-translation tool that takes the existing workflow definitions and produces Primo Studio equivalents — handled approximately 70% of the conversion automatically. The remaining 30% was manual remediation, mostly on custom platform-specific components and exception handling.
Migrated processes ran in shadow mode against the incumbent platform instance for 24–48 hours per process — same input, both platforms producing output, output compared. Where the comparison passed, the process was cut over. Where it failed, the team triaged and either patched the conversion or escalated to a manual rebuild.
By week 4, all 46 processes had been migrated and validated. By week 6, the incumbent platform instance was decommissioned.
Track 2 — AI Server introduction
While Track 1 was running, a second team introduced Primo AI Server alongside the new RPA estate. The first AI use case targeted supply chain document exceptions — a category of exceptions that had been on the deferred list for over a year.
The AI Server use case went live in the same week that the platform migration completed. Within two weeks of go-live, AI was handling exceptions that had previously been routed to human review, with measured accuracy above the team's pre-defined acceptance threshold.
The migration every other vendor had quoted at six months took six weeks. The real surprise wasn't the speed — it was that we ended the migration with more capability than we started with, not less.
The numbers
What changed for the team
The RPA development team stopped spending half its time working around platform limitations. Velocity on new automations doubled in the first quarter post-migration, even as the team itself did not grow.
The exception-handling team — humans who had been triaging 1,500+ supply chain document exceptions per week — saw 60% of that volume absorbed by AI Server within the first two months. The team was redirected to genuinely complex exceptions and to root-cause analysis on recurring patterns, which had previously been backlogged.
Retail platform migration — deployment architecture
What's next for this customer
The retail group is now expanding the AI Server footprint into customer-service inbound triage and returns categorisation — both processes that had been deferred under the previous platform. The CoE is also using the migration playbook (automated converter + shadow mode + 6-week sprint) as a template for two other business units in the broader retail group that are still on legacy RPA platforms.