If you are weighing up how to automate a piece of work, you likely have three options in front of you: a managed AI employee, robotic process automation (RPA), and workflow automation tools such as Zapier and Microsoft Power Automate. The old assumption that only one of these can use AI, or that only one can read a document, is out of date. What actually differs is who builds and runs the automation day to day, and what happens when it needs fixing.
The three approaches, defined fairly
Managed AI employee: a defined role, delivered as a service. You agree a job description in writing that names the jobs, the tools and sources it may use, and who approves what. The provider, Struan in this case, builds the role, runs it every day, corrects mistakes and signs a weekly report. See the full process.
Robotic process automation (RPA): software robots that carry out repetitive tasks, typically by interacting with application interfaces or APIs. Common platforms include UiPath, Automation Anywhere and Blue Prism. Modern RPA platforms are commonly paired with document understanding features that combine RPA with AI models to extract data from structured and unstructured documents, with a human validation step in the workflow, as UiPath's own documentation sets out. UiPath describes Document Understanding as combining RPA and AI for document processing, with availability depending on the deployment.
Workflow automation tools: tools such as Zapier and Microsoft Power Automate connect applications using trigger and action logic: when X happens, do Y. Zapier and Microsoft Power Automate both offer AI capability within their platforms. Zapier's help documentation describes AI steps that summarise, classify and extract data, with a choice of underlying model, defined tool access and stated limitations. Microsoft's AI Builder documentation describes prebuilt and custom AI models available inside Power Apps and Power Automate, including document processing flows. As with RPA, this AI capability typically needs to be selected and configured as part of building the flow.
Where the approaches actually differ
Because most platforms now offer some form of AI capability, the sharper distinction is who owns building, testing and maintaining the automation, and what happens when something breaks.
- Managed AI employee: Struan is responsible for building and managing the agreed role. The written scope identifies approvals, monitoring and how corrections are handled.
- RPA: your team or an implementation partner builds and maintains the robot, including document understanding where the chosen deployment makes it available, and is responsible for fixing it when an application interface changes.
- Workflow automation tools: whoever builds the flow, including any AI step, is responsible for testing it and maintaining it as your tools and data change.
Illustrative example, not a benchmark: a business extracting data from supplier invoices could configure Power Automate's AI Builder to pull fields from each PDF, or use an RPA platform's document understanding feature with a human validation step. Either route still needs someone in the business to build, test and maintain that flow over time. A managed AI employee's job description puts that build and maintenance inside the service instead.
Cost basis, without invented figures
None of these approaches has a single market price, and this article will not invent one. What differs is the shape of the cost, not a specific figure.
- Managed AI employee: a build cost and an ongoing managed-service fee, set out in the written quote. Confirm what each fee covers and agree how a change in scope will be priced.
- RPA: Licensing varies by vendor and deployment; implementation and maintenance still need to be costed.
- Workflow automation tools: Licensing varies by vendor and deployment; implementation and maintenance still need to be costed.
When each approach tends to fit
- RPA tends to fit when you have an internal RPA team, are automating a legacy system without a usable API, and want the automation to stay in house.
- Workflow automation tools tend to fit when you have someone technically comfortable connecting modern cloud apps and want to own the build yourself, including any AI step the platform offers.
- A managed AI employee tends to fit when you want a defined role that is run, monitored and answered for as a service, with named approvals and a weekly report, rather than building and maintaining the automation in house.
Where the approaches overlap
These are not always separate choices. Some businesses keep an existing RPA robot running a legacy system with no API, use a workflow tool for simple app to app connections, and bring in a managed AI employee for a broader role that needs judgement calls and named approval, such as a finance or support role. If you are already running RPA or automation tools and finding the maintenance heavy, that is worth reviewing against your own numbers before switching anything. Struan's full comparison against hiring, software and outsourcing, including the rows where Struan is not the better option, is on the Compare page.
Book a sense-check call to talk through your specific workflows and where each approach would actually fit. The first call is free.
You can also see real deployments written up in full before deciding anything.