IBM RPA: what it does, what it costs and when it is worth using

The short version

• IBM RPA automates rule-based tasks by operating application screens the way a person would, not by calling APIs.
• RPA is most useful when a system has no usable API and the task follows fixed rules at high volume.
• Screen-driven bots need repair whenever the applications they operate change their layout or flow.
• IBM publishes indicative list prices for IBM RPA that exclude taxes and vary by country.
• An RPA bot follows a fixed script, while an AI agent chooses its next step, which makes the agent more flexible and less predictable.

What is IBM RPA?

IBM RPA is IBM's robotic process automation product. It runs software bots that operate business applications through their screens, the way a person would: open the application, read a field, type into another, click submit. The bot follows a script. It does not decide anything the script did not anticipate.

Like other RPA platforms, it has three parts. There is an authoring tool where developers build bots, a runtime where the bots execute, and a control layer for scheduling, stored credentials and logs.

Bots come in two kinds. Attended bots run on a person's desktop and are started by that person, usually to take the repetitive part of a task off their hands. Unattended bots run on their own, on a schedule or when work arrives in a queue, with nobody watching.

How does IBM RPA work?

It finds elements on a screen and acts on them in a fixed order. A developer builds the sequence in the authoring tool, telling the bot which window to find, which field to read and what to do with the value. At run time the bot repeats those steps exactly.

That exactness is the strength and the weakness. A bot does the same thing thousands of times without getting tired or skipping a step. It also does the same thing when the screen has changed underneath it. If a vendor moves a button, renames a field or adds a confirmation dialog, the bot either stops or, worse, acts on the wrong element.

This is why RPA has a reputation for brittleness. The bot is coupled to the presentation layer of software it does not control. Every upgrade to that software is a possible break, and nobody warns the bot in advance.

What does IBM RPA cost?

IBM publishes list prices for IBM RPA on its pricing page. It describes them as indicative, excluding taxes and varying by country, so the figure you see depends on where you buy and will not be the final invoice. The same page sets out IBM's product and trial terms.

The licence is rarely the whole cost. Budget also for building each bot, testing it against real data, the machines or cloud capacity it runs on, and the ongoing work of fixing it when the applications it drives change. That last line item does not end.

A useful way to price it: estimate the hours a bot will save each month, then subtract the hours someone will spend keeping it working. If the remainder is small, the licence price is irrelevant, because the project is not worth doing.

How does RPA compare with API integration and AI agents?

When is IBM RPA worth using?

When a high-volume, rule-based task runs through a system that has no usable API and whose screens rarely change. That combination is narrower than RPA marketing suggests. Where it exists, a bot is often the only practical way to automate without replacing the system.

Typical fits are older desktop or terminal applications, internal tools nobody will extend, and third-party portals that offer no integration. Another is an organisation that already runs IBM software and wants automation governed by the same vendor, contracts and support.

The second group is where IBM RPA specifically makes most sense. An existing IBM relationship simplifies procurement, identity and support. Without one, pick the platform only after confirming that RPA is the right technique at all.

When is RPA not worth it?

When the systems involved already have APIs, when the process changes often, or when the work depends on judgement. In each of these a bot adds a fragile layer that an API call, a redesigned process or a person would handle better.

Small businesses are a common mismatch. Zapier's 2021 survey of 2,000 US knowledge workers at small and medium businesses was published under the headline '66% of small and mid-sized businesses say automation is essential for running their business'. Zapier sells automation software, so read that as a sign of demand rather than a neutral measurement. The demand is real. But most small firms run cloud tools with APIs, where a direct integration costs less and breaks less often than a bot driving screens.

RPA is also a poor fix for a clumsy process. A bot preserves every clumsy step, now at machine speed and harder to see. Simplify the process first, then decide whether what remains needs a bot.

How is RPA different from an AI agent?

An RPA bot follows a script written in advance. An AI agent uses a language model to decide its next step from the input in front of it. The bot is predictable and rigid. The agent is flexible and harder to predict.

That difference shows up in how each one fails. A bot fails loudly: the screen does not match, the run stops, someone gets an alert. An agent can fail quietly, producing a plausible answer that is wrong. So an agent needs tests on real cases and checks on its output before it touches anything that matters.

The useful pattern is often a combination. Let a model do the reading: classify the email, extract fields from the document, decide which queue a request belongs in. Then hand the action to something deterministic, ideally an API call, or an RPA bot where no API exists, with validation in between. Neither tool should be doing the other's job.

How should you read the published evidence for IBM RPA?

With care, because most of it is produced by vendors or drawn from large organisations. That does not make it wrong. It means it answers a different question from the one a smaller buyer is asking.

IBM publishes a customer story about Primanti Brothers, a restaurant chain, produced in May 2022. It describes one customer, with results reported by that customer and by IBM. It shows the product in use. It does not tell you what to expect in your own processes.

Survey evidence has similar limits. Deloitte's seventh Global Intelligent Automation Survey drew on 479 executives from 35 countries, mostly at larger organisations. It is a reasonable picture of enterprise automation programmes. It says little about a firm with one operations person and a handful of cloud tools.

What should you check before choosing IBM RPA?

• •Whether the target systems really lack usable APIs. Ask the vendor or the IT team, not just the people who use the screens.
• •How often each application's screens change, and who controls those changes.
• •The volume: how many times a month the task runs, and how long each run takes a person.
• •Who will own the bot after launch, watch its failures and fix it.
• •The current IBM list price and trial terms for your country, taken from IBM's own pricing page.
• •Whether any part of the task needs judgement that a fixed script cannot supply.