RPA maintenance: the work, outsourcing costs, and the team
An RPA robot’s maintenance starts the day it goes live. By morning, a robot has stopped. A customer’s screen changes and it needs fixing. The steps of the work change, and someone asks for the robot to follow. A product upgrade arrives. As this piles up, change requests wait in a queue and payments to an outside maintenance vendor go out every month.
This article splits RPA maintenance and operation into the six things they actually involve, then covers fixing in-house versus outsourcing, the prices outside maintenance services publish, how to set up the team, and how to shrink the load. It applies to WinActor, UiPath, BizRobo!, and Power Automate Desktop alike; the prices are for Japan. For how to total licenses and fixing effort, see what RPA costs to keep running.
Key points
- RPA maintenance and operation means six things: handling stops, fixing after screen changes, handling change requests, upgrades, license management, and keeping the robot ledger
- Product support and robot maintenance are different things. Fixing a stopped robot falls to your own staff or a maintenance vendor
- Few maintenance services publish prices. Those that do sell tickets, a flat monthly fee, or consulting. Check in the contract whether fixes for screen changes are included
- For the team, name who manages the whole, an owner and operators for each robot, and who fixes. Even if you outsource, keep the ledger and the owners in-house
What RPA maintenance and operation are
Operating RPA means running robots every day, checking the results, and responding when they stop. Maintaining it means fixing robots, changing them, and keeping the environment they run in working. In practice, the same people usually do both.
The RPA guidebook for local governments from Japan’s Ministry of Internal Affairs and Communications (July 2025) says that keeping RPA in use means reworking scenarios continually, for stops and malfunctions, changes in regulations, and so on. It also says robots are unstable right after they go live, and even once stable they are affected whenever the systems they operate change, so stops and malfunctions should be treated as likely.
The six things are:
| Task | What it involves | When it comes |
|---|---|---|
| Handling stops | Notice the stop, check how far it got, then cover by hand or fix it and run the rest | Without warning, often on busy days such as closing day |
| Fixing after screen changes | Find the robots whose screens changed, fix them, and check them | Each time the other system is updated; outside systems often give no notice |
| Change requests | Change a robot’s steps when the work, forms, customers, or regulations change | Whenever business teams ask; bunched around regulatory changes |
| Upgrades | Check that every robot still runs after updates to the RPA product, OS, browser, or Excel | On the product’s update cycle and support deadlines |
| License management | Know how many licenses you have, when they renew, which PCs use them, and which sit unused | At the yearly renewal, and when PCs or robots come and go |
| Robot ledger | Keep the list of what each robot does and who is responsible for it current with every change | When a robot is built or changed, or its owner changes |
1. Handling stops
When a robot stops, work through these steps:
- Notice: send failure notifications to a team, not to one person
- Check how far it got: use the error message and the log to see which items finished and where it stopped
- Keep the work going: decide whether to wait for a fix or do the work by hand. If closing day is near, do it by hand first
- Fix it and run the rest: rerun without the items already done, so nothing is entered twice
- Record it: the date, the cause, the fix, and the time it took
The guidebook calls it most important to build safeguards in from the start: when unexpected data arrives, or the system being operated stops or slows down, either halt the automatic processing and add a step where a person decides, or skip the item so a person can check it later. A robot that keeps going and does the wrong thing does more damage than one that stops.
Why robots stop, and how to build ones that stop less, is covered in why RPA robots stop when screens change.
2. Fixing after screen changes
For robots that operate outside web systems, this is most of the maintenance work. A button moves slightly, or fields change order, and the robot can’t find what to act on and stops.
Fixing means finding the affected robots, fixing them, checking them with test data, and putting them back into production. If the ledger’s “connected systems” column is filled in, you can find every robot that operates the changed system at once. The guidebook likewise says to use the ledger to assess the impact on robots whenever a system is updated or an outside system’s specification changes.
For internal systems, a rule that update plans reach the RPA team in advance lets you fix robots before they stop.
3. Change requests
When the work changes, the robot’s steps have to change too: a new customer, an extra column on a form, a different approval flow, a regulatory change. When few people can fix robots, this is where requests pile up.
The guidebook describes local governments that stopped using RPA because, in work with frequent regulatory changes, scenario fixes could not keep up.
To keep requests from piling up, take them in through one channel, set priorities by impact such as closing day, and let business staff make small changes themselves.
4. Upgrades
RPA products set a support deadline for each version. Once a version is past it, you get no product support even when something breaks. The rules for the main products, as of October 2026:
| Product | Versions and support |
|---|---|
| WinActor | When a new minor version comes out, sales and support of the previous latest version end one year after the new release. Versions not on the list are outside the maker’s support. Support for Ver.7.6.1 runs until August 31, 2027 |
| UiPath (on-premises) | One long-term support release a year, typically in October. Mainstream support runs 24 months from release, then extended support for 12 months (some contracts add a further year of limited support). Versions past support receive no support of any kind |
| Power Automate Desktop | A new version almost every month (2610 on October 15, 2026). A flow created or edited in a newer version may not work correctly in an older one. Rolling back to a previous version is not supported |
With every upgrade, check that every robot runs on the new version. Update a test PC first and run all the robots on it; if nothing breaks, update the production PCs. For products like Power Automate Desktop, where flows can stop working when the PC that builds them and the PC that runs them are on different versions, update them together. The guidebook also says to use the ledger to define the scope of work when applying patches or upgrading.
End of support for an OS, browser, or scripting language can bring rework across every robot. Internet Explorer’s end of support and the VBScript deprecation are summarized in what RPA costs to keep running.
5. License management
How you manage licenses depends on whether they are per PC or by concurrent use. A WinActor node-locked license, for example, allows one installation on one PC, while floating licenses use a management product and are sized to the number of PCs running at once.
Track how many licenses you have and when they renew, which PCs and accounts use them, and whether any sit unused. The guidebook lists recording license types, quantities, and the PCs they are installed on in the ledger, finding idle time from usage, and rebalancing licenses and PCs.
Retiring a robot does not shrink the license contract by itself. A few months before renewal, compare the ledger with actual use. Recent price revisions are summarized in what RPA costs to keep running.
6. The robot ledger
Each of the other five tasks goes faster with a ledger, because it tells you at once whom to contact and how to cover by hand when a robot stops, which robots operate a system whose screens changed, which robots an upgrade touches, and which PCs use the licenses.
The columns and update rules are covered in how to build a robot ledger for RPA, and there is an Excel template ready to use.
In-house or outsourced maintenance
The guidebook divides who builds and fixes robots three ways: outside contractors, the IT department, or business departments, and says combinations work too. Companies can think about it the same way.
| An outside maintenance vendor | IT or the DX team | Business departments | |
|---|---|---|---|
| How fast a fix comes | Days, through request, quote, and work | Waits when requests bunch up | Fast |
| How you pay | Contract fees; even small fixes cost money | The team’s time | Business staff’s time |
| What stays in-house | Working robots; the vendor knows what is inside | Company-wide rules, and people who can fix | People who know the work and can fix |
| Weak spot | You keep depending on the vendor; work outside the contract costs extra | The fine details of the work are hard to see | Robots only their builder can fix, and rogue robots, appear easily |
| Suits companies that | Cannot put anyone in-house on fixing | Have many robots and want to manage them company-wide | Can name two people in the business department |
The guidebook says that if you outsource every fix, you need to budget enough for maintenance, and if you can’t, you should arrange to make at least small changes in-house. Conversely, even when business departments fix their own robots, it recommends being able to get outside help when needed, such as answers to questions or on-site support capped at a set number of hours.
One more distinction: product support and robot maintenance are different things. WinActor’s support service, for example, answers questions about what the manuals say and about the product’s specifications (functions, specs, and limits), and covers the latest version. Fixing a robot that has stopped falls to your own staff or a maintenance vendor you contract separately.
The prices outside maintenance services publish
Most RPA maintenance services set prices by individual quote, and few publish them. These are the examples we could confirm in October 2026:
| Provider and service | Type | Published price | Terms |
|---|---|---|---|
| Information Development, “ID-Cross RPA Remote Maintenance Service” (WinActor, UiPath) | Tickets | From ¥500,000 for 25 tickets, from ¥950,000 for 50, from ¥210,000 for 10 more (reference prices) | Handles spec changes, error fixes and investigation, license renewal, moving to a new runtime PC, and similar work. One ticket is up to half a person-day (about 4 hours). Contracts run in six-month units, with at most 20 tickets a month as a rule. Announced May 2022 |
| CEC, “RPA Flat-Rate Support Service” (ROBOWARE) | Flat monthly fee | Maintenance plan ¥300,000 a month (build plan ¥600,000 a month) | Includes the ROBOWARE license. Maintains scenarios made under the build plan. Contracts in three-month units |
| ITSO, “RPA Subscription” | Flat monthly fee (building, monitoring, and fixing together) | ¥100,000 a month (¥50,000 a month for each unit from the second) | Unlimited development. Includes updates to follow system changes, and troubleshooting. One-year minimum |
| Digital Vorn, “BizRobo! Online Support” | Consulting | Trial plan ¥40,000 a month before tax for up to four sessions; extra sessions ¥14,000 each | 45-minute web meetings answering development and operations questions. Three-month contract. Reported September 2020 |
Only Digital Vorn states how tax is treated. ID-Cross’s reference prices come to about ¥20,000 per ticket, one piece of work of up to four hours.
What each type suits:
- Tickets or hourly: robots that stop rarely, with fixes that come irregularly. How many tickets a fix takes is decided after you ask
- Flat monthly fee: many robots, with fixes every month. Building and maintenance often come together, and only robots built under the service may be covered
- Consulting: you have someone in-house who fixes robots and want help only when they get stuck. The fixing itself stays in-house
How to compare outsourcing with build costs included is covered in what outsourcing automation costs.
What to check before signing a maintenance contract
- Are fixes for screen changes included in the fee? What about fixes for changes in the work?
- How long from request to fix? Does it cover stops on closing day or at night?
- Is there a cap per month or per fix, and what does going over cost?
- Do you receive a record of what was fixed and why?
- When the contract ends, can the robots and their manuals be handed to your own staff or another vendor?
- Is checking robots after a product upgrade included?
Setting up the team
The team decides who manages the whole, who owns each robot, and who fixes. The guidebook splits this into “governance,” which sets company-wide rules and checks that they are followed, and “management,” which names a manager and an operator for each robot and runs its building, fixing, and operation.
| Role | What they do | Who, for example |
|---|---|---|
| Managing the whole | Sets the rules. Manages the ledger and licenses. Tallies stops and results, and decides what to retire | IT, the DX team |
| Robot owner | Sets the robot’s purpose and priority, and approves changes to its steps | A manager in the business department that uses it |
| Robot operators | Check results daily. Cover by hand when it stops. Request changes | Two people in the business department |
| Who fixes | Investigates and fixes stops. Handles change requests. Checks robots after upgrades | IT, business staff, or a maintenance vendor |
Two operators means someone is still there to act when the robot stops, even after one of them moves on. Taking over robots whose builder has left is covered in taking over an RPA bot whose builder has left.
Rules to agree on
- The path when a robot stops: where notifications go, who looks first, and how to cover by hand
- How change requests come in: the channel, how priorities are set, and what business departments may change themselves
- Notice before systems change: plans to update or change internal systems reach the RPA team in advance
- Checks before production: test with test data, and run in production only after the owner approves
- Ledger updates: update the ledger as part of the same procedure as any change to a robot
- A yearly inventory: compare the ledger with the robots actually running, and with the licenses
What stays in-house even when you outsource
Even with a maintenance vendor, keep the ledger, an owner for each robot, the steps for covering by hand when a robot stops, and the records of fixes the vendor sends you. Without them, when you change vendors or review the contract, no one knows what the vendor was looking after.
How to shrink the maintenance load
- Retire robots no one needs: when the work is gone, or the volume is small enough to do by hand, retire the robot and cut the license too
- Replace screen work with official integrations: if the other system offers an API or CSV downloads, stop operating its screens. Screen changes can no longer stop the robot
- Make the ledger answer impact questions at once: when screens change, when upgrades come, and when owners change, you find the affected robots right away
- Record stops and start with the robots that stop most: keep, for each robot, how often it stopped and how long fixes took, and you will often find the effort concentrated in a few robots. How to turn that time into cost is covered in what RPA costs to keep running
- Use robots that relearn when screens change: if a robot repeats the operations it recorded and relearns only those a changed screen no longer matches, small screen changes no longer need a fix each time
Move the robots that keep stopping to Kitewell
The robots that stop most can also be moved to Descarty’s Kitewell. After the move, maintenance work changes like this:
- When a screen changes: a workflow repeats what it learned in the same order every time, and AI relearns only the operations a changed screen no longer matches, so the work goes on. Which operations it relearned stays on record with screenshots. Steps that read PDFs or Excel files, or check a screen, use AI every run
- Change requests: the person in charge asks Kitewell in a chat to change the steps, and checks the change before it is applied
- When it stops: failures are sent to email, Slack, Microsoft Teams, and more. After a fix, the rest runs from the step that stopped
- The ledger: every workflow can be exported to Excel as a robot ledger
- Upgrades: automatic updates, so there is no upgrade work
Fixes can come from your own staff asking in a chat, or from maintenance support by Descarty or a partner; you choose.
Kitewell does not read robot files: you describe your current robot’s steps in words, and AI builds the workflow again on the real screens. Kitewell Team is ¥14,800 per computer a month, or ¥148,000 a year (excluding tax), the same for building and running computers. You can try Kitewell free for 14 days, or ask for a 30-minute demo. See RPA migration for how it works.
Summary
- RPA maintenance and operation means six things: handling stops, fixing after screen changes, change requests, upgrades, license management, and the robot ledger
- Product support and robot maintenance are different things; fixing a stopped robot falls to your own staff or a maintenance vendor
- Few maintenance services publish prices; those that do sell tickets, a flat monthly fee, or consulting. Check in the contract whether screen-change fixes are included, how long fixes take, and that you receive records and can hand the robots over
- For the team, name who manages the whole, an owner and two operators for each robot, and who fixes. Even if you outsource, keep the ledger and the owners in-house
- To shrink the load, cut the number of robots, replace screen work with official integrations, and start with the robots that stop most
Sources
- Ministry of Internal Affairs and Communications: RPA guidebook for local governments (July 2025; in Japanese)
- WinActor: Lifecycle policy (September 1, 2026; in Japanese), License prices (in Japanese)
- NTT DATA: WinActor support service terms (version of September 25, 2025; in Japanese)
- UiPath: Product lifecycle support terms, Release strategy (checked October 28, 2026)
- Microsoft: Released versions for Power Automate for desktop (October 16, 2026), Update or roll back Power Automate for desktop
- Information Development: Launch of the RPA remote maintenance service (May 31, 2022; in Japanese)
- CEC: RPA flat-rate support service (checked October 28, 2026; in Japanese)
- ITSO: RPA subscription (checked October 28, 2026; in Japanese)
- Cloud Watch: Digital Vorn’s online support for BizRobo! (September 18, 2020; in Japanese)
Prove Kitewell on your own systems
Try it free for 14 days on your own PC. To see it working first, ask for a 30-minute demo. There is also a one-month onboarding (¥500,000 before tax), where we build your first job together.
What the onboarding checks
- Which method connects each target system: API, file, database, or screen
- Whether Kitewell can work a system with no API through its screens, as a person would
- Whether AI relearns a step and processing continues when a screen changes
- Whether errors are handled as agreed: stop, notify, or retry
- Whether it runs in your environment, with the AI model you choose
- Whether processing time and AI usage cost fit daily operation
How the onboarding runs
- Week 1Requirements and success criteria
- Weeks 2 to 3Build
- Week 4Verify on production screens
- Final dayReport
One-month onboarding
- Length
- One month
- Cost
- ¥500,000 (tax excluded). Your staff build, with the developer beside them
- If you go on to a license
- The onboarding fee is credited to the first year's contract
- Without the onboarding
- At Team scale, start on Team monthly (¥14,800 per computer a month, tax excluded), and cancel in any month
You can try it first on your own PC, in the free 14-day trial.