The short version
Stereologic discovers the inputs. Other approaches ask you to define them first.
That is the difference, and almost everything else follows from it. Recording starts, and the applications and screens in use, the business data and its values, the case each piece of work belongs to and the process connecting them are all recognised as the work is observed.
From there the evidence assembles itself: validated recognition outputs are associated to corresponding cases, and combined into end-to-end process paths and variants — including work that was interrupted and resumed, carried out concurrently, or handed between several people.
The other architectures ask for those same inputs up front: capture rules naming which applications and screens to record, the data fields that matter, a configured case identifier, an event log prepared before analysis begins. Each is a sound engineering choice, and each is work your team does before the first finding arrives.
Process mining sees the shape. Task mining sees the detail. Stereo sees both.
So the question worth asking a vendor is simple: how much do we have to know before you can tell us anything? The answer sets your time to first evidence, how much of the operation you end up seeing, and how much of it you had to guess at in advance. The comparisons below answer it for each platform, from that vendor’s own documentation — Celonis, Skan.ai, Mimica, Apromore and Soroco all describe real-work and desktop-observation capability, and these pages take them at their word.
