Automation & RPA

The umbrella term: technology takes over steps of work previously done by people — following fixed, predefined rules. Automation pays off where work is frequent, uniform and rule-based: transferring data, checking documents, producing reports. The key word is “rule-based”: classic automation executes, it does not decide. Every exception the rules did not anticipate goes back to a human.

RPA software operates existing applications the way a person would: it clicks buttons, fills in forms, reads screen content and switches between applications — without those systems needing APIs. From around 2015 this made RPA the fastest way to connect legacy systems; vendors like UiPath grew big on it.

The strength is also the weakness: an RPA bot replays a recorded script. If the screen changes, a field moves or an unexpected case appears, the process breaks — the bot cannot improvise.

Attended bots work next to a person: a clerk triggers them on demand, for instance to pull data from three systems together during a customer call. Unattended bots run on servers with nobody watching, usually at night or on a schedule, controlled by an orchestrator. The distinction is more than technical: it defines who intervenes when something goes wrong — and how quickly anyone notices.

The control room of a bot fleet: the orchestrator assigns jobs to bots, schedules runs, manages credentials, collects logs and reports failures. In mature RPA organisations it is the central operations tool — and the predecessor of what an operations hub does for agents today: one place where you can see what the digital workforce is doing right now.

Before automating, you need to know how a process really runs — not how the handbook says it does. Process mining reconstructs the actual flow from the digital traces in your systems (timestamps, status changes, handlers) and reveals loops, waiting times and exceptions nobody knew about. The result is the most honest basis for any automation or agent roadmap.

The first place machine learning entered automation: IDP reads unstructured documents — invoices, delivery notes, forms — and turns them into structured data. Classic OCR only recognises characters; IDP additionally understands which field is the invoice number, even though every supplier uses its own layout. This raised the share of cases passing through without a human — but the exceptions remained.

The metric automation is measured by: the share of cases that pass from start to finish without a human touch. An STP rate of 70% means 7 out of 10 invoices post themselves — and 3 land on a desk. Every stage of automation's evolution, from RPA through IDP to agents, is at heart an attempt to raise this rate without losing control.

Automation not as a project but as an ongoing service: a partner builds the bots (today: the agents), runs them, monitors them, keeps them current and charges by usage or outcome — instead of the company building platform, team and operations itself. The model moves the risk to where the experience sits, and makes automation accessible to mid-sized companies without an automation team of their own.

BOTFORCE was founded on exactly this idea — the story begins in 2019, long before “agent” became a buzzword.