Back to all posts
Projects & Retrofits7 min read·November 15, 2026

RPA or a Proper Interface for Legacy Plant Software?

TB
ThynkBlox Automation Team
Automation Engineering

If a legacy plant system has a supported interface, such as a database view, a file exchange or an API, use it. A proper interface is usually more stable, shows errors more clearly and is easier to audit than robotic process automation (RPA) that clicks through screens. Keep RPA for systems that offer no interface and whose screens rarely change, and treat it as a temporary bridge until you have a better option. Never use RPA for safety-related functions.

What is the difference between RPA and a proper interface?

Screen-driven RPA copies what a person does. It reads text from screens, types into fields and clicks buttons, in the same order an operator or clerk would. The software being automated has no idea a robot is using it.

A proper interface exchanges data through a channel built for that purpose. Common examples in plants:

  • Database interface: reading from or writing to agreed tables or views, preferably ones the system's owner supports for integration.
  • File exchange: one system drops a structured file in an agreed location in an agreed format, and the other picks it up.
  • API: a defined set of requests and responses, with documented fields and error codes.

The "RPA vs workflow automation" question usually comes down to this choice: automate the user interface, or build a workflow around a real data path. For how RPA compares with newer agent-based tools, see the main blog's article on the difference between AI agents and RPA.

How do they compare on stability, change risk and maintenance?

CriterionScreen-driven RPAProper interface (database, file, API)
StabilityDepends on screen layout, field positions, timing and pop-upsDepends on a defined data contract
Change riskA software update, new screen resolution or renamed button can break it silentlyChanges to the contract are usually visible and can be versioned
Error visibilityFailures can look like success if the robot clicks the wrong fieldErrors come back as codes, rejected files or failed transactions
AuditLogs show actions on screens, which are harder to tie to data recordsRecords can be logged by message, file or transaction identity
Speed and loadLimited by screen response; runs one session at a timeUsually handles batches and runs without a user session
Who maintains itOften an office automation team, away from plant engineeringUsually shared between system owner and integrator under change control
Effort to startOften quick to build a first versionNeeds interface design, testing and sometimes help from the system owner

The fast start is RPA's real advantage. The cost comes later, when the robot breaks after a change nobody told it about.

Why does error visibility matter so much on the shop floor?

A person at a screen notices when something looks wrong. A robot does not, unless someone has scripted that exact check. If a pop-up shifts focus, the robot may type a batch quantity into the wrong field and carry on.

With a proper interface, the receiving system checks the data and rejects what does not fit. A rejected file or failed API call can raise an alert, so someone can follow it up before it reaches production. That matters when the data controls what the plant makes, such as production orders or recipe parameters. Related reading: exchanging production orders between ERP and the shop floor.

When does each option make sense?

Use the system's interface availability and how often its screens change as the two main questions.

Does the system offer a usable interface?How often do screens change?Suggested approach
Yes, supported and documentedAnyUse the interface
Yes, but undocumented (for example, raw database tables)AnyConfirm with the system owner what is safe to read; prefer read-only access; avoid writing to undocumented tables
NoRarely (frozen legacy version)RPA may be acceptable as a bridge, with monitoring and a replacement plan
NoOften (regular updates)Avoid RPA; look at a gateway, middleware, an upgrade or a manual step with checks
No, and the function is safety-relatedAnyDo not automate through screens; involve a qualified assessor

If the data you need sits in a PLC rather than in office-side software, a gateway is often a better route than either option. See how to choose an industrial protocol gateway and collecting data from an existing PLC.

Why should RPA stay away from safety-related functions?

Safety functions need predictable behaviour, defined responses to faults and a design checked by a qualified assessor. Screen automation cannot reliably provide any of these. It depends on display timing, user sessions and screen layout, all of which can change without warning.

Do not use RPA to acknowledge alarms, bypass interlocks, change safety settings or start equipment. If a task touches a safety function, take it out of scope for screen automation and plan it with a qualified assessor. The guide to planning a safety assessment for a machine retrofit covers how to begin.

How should you govern whichever option you choose?

Both options are plant software, and both need change control. Record:

  • What the automation does, which systems it touches and who owns it.
  • The data fields, files or screens it depends on.
  • How failures are detected, who is alerted and what the manual fallback is.
  • Which software updates must trigger a retest.
  • Where versions and configuration are backed up.

Bring RPA scripts and interface configurations into the same process you use for PLC and HMI changes. See controlling plant software changes after commissioning.

What would a worked decision look like?

Illustrative example: a plant wants to copy daily production counts from an old line-reporting application into a newer planning tool.

The team checks the old application. It has no API, but it can export a daily report file. The vendor no longer issues updates, so the screens are frozen.

  • The team chooses the file export over RPA, even though both would work, because a file can be checked for completeness and logged.
  • The planning tool imports the file and rejects rows with missing line or shift identities.
  • A failed import sends an alert to the production supervisor, and a manual entry step serves as fallback.

If the application had no export at all, RPA reading the summary screen could have served as a bridge, provided it had a completeness check and a plan to replace it.

What mistakes should you avoid?

Choosing RPA because it was quick to demo. Judge it on how it behaves after the next software update.

Writing to undocumented database tables. Reading may be acceptable with the owner's agreement. Writing can bypass the application's own checks.

Running robots under a shared operator login. Give the automation its own account and permissions so its actions can be audited. See setting HMI user roles and permissions.

No owner after go-live. Name who fixes it when it breaks, and include it in the support agreement.

Letting a bridge become permanent. Set a review point for every RPA bridge to decide whether it should be replaced.

When does this comparison not apply?

It does not apply to real-time machine control, which belongs in the PLC or control system, not in either approach described here. It also does not apply to safety functions, which need a qualified assessor whatever approach you consider. And where a task runs only occasionally, a well-documented manual procedure may be simpler than automating it at all.

Frequently asked questions

Is RPA ever the right long-term answer for plant software?

Occasionally, for a stable, frozen system with no interface and a low-consequence task. Even then, treat it as plant software: give it an owner, monitoring, change control and a manual fallback. Review it each time the surrounding systems change, and replace it if a supported interface becomes available.

Is reading directly from a legacy database safe?

It can be, if the system owner agrees and access is read-only, using views or agreed tables. Check the effect on the application's performance and confirm what happens to your queries after an upgrade. Avoid writing to the database directly, because that can bypass validation the application normally performs.

How do we detect when an RPA robot has silently failed?

Add checks the robot cannot fake. Compare record counts between source and destination, confirm expected values on the screen before you submit, and alert when a scheduled run does not complete. Have someone outside the robot's own logs review the results regularly, especially after any software update.

Who should maintain the automation after handover?

Name one owner for each automation, usually whoever owns the system it writes to, with integrator support set out in an agreement. Record its dependencies, credentials and backup location in the handover files. Make sure plant engineering knows it exists, so updates trigger a retest rather than an unexpected failure.

If you want help assessing which option fits your systems, see our services or contact us.

Next step

Want to apply this to your plant?

Every project starts with a fixed-fee assessment, and you can stop after the report.