31%,
Moving a logistics group to AWS and rebuilding its demand forecasting
96 workloads migrated to AWS and right-sized for a logistics group.
The context
The group's two operating companies, a freight forwarder and a contract warehousing business, each ran their own hosting: one on a managed private cloud, the other on a half-finished public cloud move. Between them sat 96 workloads, from warehouse management and billing to a demand forecasting model planners had stopped trusting.
Hosting contracts were up for renewal in twelve months, which set a hard deadline. Peak season could not be touched, both engineering teams had to stay involved, and the board wanted the forecasting problem fixed as part of the same programme rather than as a separate project later.
What we did
- 1
Workload inventory
We catalogued all 96 workloads with their dependencies, utilisation data and owners, then assigned each a migration path: rehost, replatform or refactor. Twelve months of utilisation data showed most servers peaked below a third of their capacity.
- 2
Landing zone first
We built a shared AWS landing zone in Terraform with separate accounts per company, central logging and tagging by cost centre. Both engineering teams reviewed every module, so nothing in the platform is a black box to them.
- 3
Migrate and right-size
Workloads moved in waves of eight to twelve, outside peak season. Each was sized from measured utilisation rather than its old specification, moved to Graviton instances where possible and scheduled down out of hours if it served only office users.
- 4
Rebuild demand forecasting
With the data finally in one place, we rebuilt the forecasting model on Amazon SageMaker using shipment history, customer orders and seasonality from both companies. Planners now see forecasts with confidence ranges, retrained weekly through a monitored pipeline.
The result
31%.
Lower cloud spend
Combined cloud and hosting spend fell 31% against the twelve months before migration, measured from invoices and normalised for shipment volume. Right-sizing accounted for most of the saving, with reserved capacity and out-of-hours scheduling making up the rest.
The forecasting rebuild was the part planners noticed. Forecast error on weekly volumes dropped by about a quarter, and both companies now plan labour and vehicles from the same model, on a platform their own engineers run.
- workloads migrated to AWS
- 96
- operating companies on one landing zone
- 2
- lower weekly forecast error
- 24%
“Tarka moved us to AWS and rebuilt our demand forecasting on top. One team across both companies, clear scope, no black boxes.”
Have a number you need to move?
Tell us what it is. We will tell you what we would measure, and how long a discovery would take.