Hyperautomation: how to combine RPA, AI and workflow into a single process

The analyst firm Gartner described hyperautomation as combining multiple automation and machine learning tools to carry out a whole piece of work rather than a single step. It also pointed out that RPA on its own is not yet hyperautomation.

RPA, or robotic process automation, is software that clicks through applications and enters data just as a person would. It is useful with older systems that cannot exchange data with other programs.

In practice, a process consists of the four types of work listed below. Each needs a different tool. Problems start when one tool is expected to do everything, e.g. when an RPA robot is supposed to read unstructured emails.

Take a hypothetical trading company. Today a complaint arrives by email. Someone reads it, looks up the order in the ERP, logs the case in a spreadsheet, asks the manager to approve a refund and passes the case on to accounts. Every handover means an email and a wait.

Once the tools are connected, the same process looks like the steps below. People still make decisions, but only where a decision requires judgement.

The change is not that the case “handles itself”. It is that nobody retypes data, nobody hunts for the status in emails and nobody forgets the case, because the workflow sends reminders about deadlines.

The most common design mistake is using a powerful tool in the wrong place. The table below helps you assign process steps to tools.

The general rule: the closer you are to systems and data, the more integration pays off. The closer you are to unstructured content, the more you need AI. An RPA robot is the solution for places where there is no other way.

For years, the textbook example of hyperautomation was AI reading invoices from PDFs and a robot entering them into the accounting software. In Poland, this example is now largely out of date.

KSeF, Poland’s National e-Invoicing System, is a government platform through which companies issue and receive invoices. Since 1 February 2026, every company has had to receive invoices through it, and since 1 April 2026 most sellers have had to issue them there. Most invoices from Polish suppliers therefore arrive as structured data rather than as an image.

So the purchase invoice process looks different today. The workflow downloads the invoice from KSeF and matches it to the purchase order. It sends discrepancies for approval and passes approved data to accounts through an integration or, with older software, through an RPA robot.

AI for reading documents is still needed, but for a narrower range: foreign invoices, receipts, delivery notes and other paperwork from outside KSeF.

A common misunderstanding goes: “we have Microsoft 365, so hyperautomation comes with the package”. Some of the tools are indeed included in Microsoft 365 plans, but key elements require additional licences. The table below shows the situation according to Microsoft’s price lists as of October 2026.

These are list prices in US dollars on annual billing. The price on a partner’s invoice may differ, and Microsoft changes its licensing rules several times a year. For example, from 1 November 2026, new Premium licences no longer include AI Builder credits, so AI features in Power Automate are billed in Copilot credits bought separately. Check them before signing a contract.

You rarely have to start from scratch. Most companies already have one of the components and can add the missing ones. Below are the three most common starting points.

Whatever your starting point, the rule is the same: one process at a time. Measure handling time and the number of errors before the change, and again after a few months. Only the result will show whether it is worth taking on the next process.

The mistakes below recur regardless of industry and company size. Each can be caught before you start, at the process design stage.

If you recognise any of them in your current plans, change the plan before buying licences. A fix on paper costs less than rebuilding a process that is already running.

Combining several tools makes sense for processes repeated many times a week. If a case comes up a few times a month, the cost of licences and maintenance will exceed the savings. A simple template and a checklist are enough in that case.

It also makes no sense when the process has not been agreed and each department does the same thing differently. You first need to settle on one way of working, and only then automate it.

Finally, if the company’s main system is due to be replaced soon, do not build robots around it. Wait for the new system, which will most likely have an interface for exchanging data.