Most manufacturers already own the data they need. It’s distributed across an ERP, a shop floor system, a warehouse system, a handful of spreadsheets, and occasionally a supplier portal nobody has logged into for a year. The problem isn’t absence of data. It’s that no single system can see enough of it to draw a useful conclusion.

Integration Doesn’t Mean Replacement

The instinct is to consolidate onto one platform. For most mid-sized manufacturers this is the most expensive available answer and rarely the right one — it means replacing systems that work, retraining everyone, and absorbing years of disruption to solve a data-flow problem.
The alternative is to leave the systems in place and connect them. Each keeps doing what it’s good at, and a layer above reads across all of them.

Decide Who Owns Each Field

This is the decision that determines whether integration succeeds. For every data element, one system is authoritative and the others defer to it.
Item master and supplier master usually belong in ERP. Actual production quantities and timings belong in the MES. Physical stock location and movement belong in the WMS. Forecast and plan belong in the planning layer. Where two systems both claim a field, you get reconciliation work forever and eventually a data set nobody trusts.
Write this down before any technical work starts. It looks like an administrative exercise and it’s the actual architecture.

Match the Method to The Need

Not everything needs to move in real time. Item master changes daily at most — a nightly batch is fine and far simpler to build and support. Stock movements and production status need to be current, because decisions depend on them within the hour.
Batch where you can, stream where you must. Building real-time integration for data that changes weekly adds cost and failure modes for no benefit.

Plan for Things Breaking

Integrations fail. A system goes down for maintenance, a network drops, someone changes a field format without telling anyone. What matters is whether the failure is visible.
Silent failure is the dangerous case: an interface stops running, nobody notices for three weeks, and planning has been working from stale data the whole time. Build monitoring that alerts on absence — if the expected transfer didn’t happen, somebody should know that morning.

Start Narrow

Connect the flows that unlock the highest-value decisions first. Usually that’s stock position, open purchase orders, and production status — enough to see material availability against demand, which is where most planning problems live.
Prove that works, then extend. Big-bang integration programmes fail for the same reason big-bang implementations do: too many things change at once and nobody can isolate what went wrong.
One practical note on scope: resist the temptation to clean up master data and build integrations simultaneously. Both are necessary, and doing them together makes it impossible to tell whether a discrepancy came from a mapping error or from data that was always wrong. Sequence them.
ticktick.ai reads from existing ERP, MES, and WMS systems rather than requiring consolidation, with defined ownership per data element and alerting on failed or stale transfers.