On site, hosted or vendor-run: what deployment changes for a plant

The line keeps moving when the link drops, or it does not

Start by listing the functions that physically gate production: label printing, weighing, scanning a component before assembly, releasing a batch, allocating a serial identifier. Any of those depending on a round trip to a remote data centre will stop the line during an outage, and site connectivity is not as dependable as urban office experience suggests. The answer is usually architectural rather than a rejection of remote hosting: keep the gating functions local with buffering and reconciliation, and let planning, costing and reporting live wherever is convenient.

Distance from the machines is not just latency

Systems that talk to control equipment need to sit inside the plant network for reasons beyond response time. The segmentation between control and business networks is deliberate, and a remote application reaching directly into machine networks weakens it. The common pattern places an on-site component adjacent to the equipment, with only aggregated or event-level data crossing to the hosted layer. Deciding which component sits where is a design task at the outset, because retrofitting a local tier to a remotely hosted product after go-live is expensive and sometimes not possible.

Upgrade cadence: who decides when you change

On-site software changes when you choose, which means many plants never change it and accumulate an unsupported version. Vendor-operated services change when the vendor chooses, which removes that problem and creates another: a validated process, a customer-approved configuration or a working interface can be altered on somebody else's schedule. Ask what notice is given, whether releases can be deferred, whether a test environment receives changes first, and what happens to your integrations. In regulated production the revalidation burden of frequent change is a genuine cost, not a theoretical concern.

Residency, exit terms and getting your records back

Manufacturing records carry retention obligations that outlast most software contracts, and customers in some sectors ask where their data physically sits. Establish the storage location, whether support staff outside that jurisdiction can access it, and what happens at the end of the agreement. Extraction terms matter more than most buyers expect: a full export in a documented, readable structure, within a stated period, at a price agreed in advance. Discovering after a decision to move that your history is available only as screen prints is a genuinely bad position.

Who is accountable when it fails at night

A hosted arrangement typically involves a hosting provider, the application vendor, an implementation partner and your own team, and outages rarely announce which of them owns the fault. Map the escalation path before you need it: who is called, what response commitment exists at the hours you actually run, and who has authority to declare the problem theirs. Plants running continuously need support cover matching their operating pattern, and a standard business-hours agreement is a common and expensive mismatch discovered during the first serious incident.

Frequently asked questions

Can a vendor-hosted system run a production line?

It can run the planning, costing and order management around a line comfortably. Whether it can run functions that physically gate production depends on whether local components exist to keep those working during an outage. Ask specifically what happens when the site loses connectivity: which screens still function, what is queued, and how reconciliation works when the link returns. A vendor without a clear answer has not deployed into an environment where production waits for their response.

What should stay on site regardless of the model?

Anything that must work while disconnected, anything talking directly to control equipment, and anything whose response time affects an operator's cycle. That usually means machine data collection, label and mark generation, scanning at gating operations, and any local buffer of identifiers or work instructions. The rest can sit anywhere. Drawing that boundary early is what makes a mixed deployment coherent rather than an accumulation of exceptions added after each incident.

Does vendor hosting reduce our internal IT workload?

It removes infrastructure administration and it does not remove the application work: user administration, configuration, master data quality, interface monitoring and testing after each release. Sites that plan for a large reduction in internal effort are usually disappointed, because the work that consumed their time was rarely server maintenance. What genuinely changes is the skill profile needed, shifting from infrastructure towards integration and data governance.