The screens behind each step
Every activity is tied to the user interface screen where it happened, so the step can be reviewed rather than described.
Activity level
See the finest details of employee activities and data.
Go past which application was open to what was actually done inside it: individual activities, the business data on the screen, interruptions, application switches and task-level causes — at mouse-click and keystroke granularity, and with non-intrusive, no-install desktop-recording software.
Illustrative view of discovered task detail.
Why this matters commercially
Knowing where the time went is not the same as knowing what got done.
Application usage tells you somebody spent half a day in the claims application. It does not tell you whether they processed forty claims or one, approved or rejected them, got stuck, or accomplished anything useful. Deep Task Mining™ goes past which screens were used to the employee activities and business data inside them.
Task-level truth
What you know about a task once it has been discovered.
Never lose a detail of what people do in office tools and business applications, how their time is spent, and which tasks can be standardized, improved or automated.
- When the task started and when it finished.
- How much time was spent actually working on it.
- How many times it was interrupted — and what interrupted it.
- Where a task was interrupted and later resumed, reconnected into one piece of work rather than counted twice.
- Where two tasks were being worked concurrently, separated rather than merged into one.
- Where a task was carried out by several people collaborating, recognised as a single task across all of them.
- Whether it was completed successfully or put on hold.
- Whether the work involves real decision-making, or data handling that could be standardized — and whether it is value-adding at all.
A patented approach, not a new one
Stereologic filed for its task-discovery patent in 2009, covering the use of optical character recognition together with user-system interaction to discover employee tasks. The U.S. patent was granted in 2019. Task mining is not an add-on here — it is where the technology started.
- 18 years of operational discovery
- 3 granted patents
Inside each task
Detail captured for every activity in the task.
Task detail is what makes a change defensible: a reviewer can see the actual step, in the actual system, with the actual time against it.
Time per activity and per task
Duration for each activity and for the task end to end, across the observed executions rather than a single walkthrough.
The inputs people make
User inputs are captured as part of the activity — checking a checkbox, choosing a menu item, and the rest of the small decisions that carry the work.
The system that carried the step
Which application each activity happened in — mainframe, ERP, CRM, browser, spreadsheet or a custom internal client.
How it is captured
Task detail without software on employee devices.
The reason task mining usually stalls is not analytics. It is the endpoint installation, the privacy conversation and the noise in the data.
No desktop installation
Employee tasks are extracted and mapped with non-intrusive, no-install desktop-recording software.
Interruption noise removed
Interruptions and distractions that normally distort task measurement are separated out, so task time reflects the work.
Private data obfuscated
Obfuscation removes private employee and client data from what is captured and shown.
Any mix of systems
Recognition holds regardless of how many systems are involved, including Excel-heavy work.
What the detail decides
Decide what to standardize, what to automate, and what to leave to people.
Frequency and duration are the obvious way to rank tasks for automation, and they are rarely enough on their own. Task detail also shows the effect a task has on service quality and customer experience, whether it carries real judgement, and how much of its elapsed time is waiting rather than working.
That is the difference between a list of automation candidates and a prioritized one. Where a task is worth automating, the same discovery output — steps, screens, inputs and systems — becomes the specification the build team works from, instead of a fresh round of workshops.
Quantify it first
Turn task detail into timing, variation and sized opportunities.
Quantify it firstProcess IntelligenceAutomate what is worth automating
Decide what should be standardized, improved, automated or kept human.
Automate what is worth automatingAI & Automation ReadinessStandardize complex work
Expose variation and workarounds before writing the standard.
Standardize complex workProcess StandardizationEvaluation questions
What evaluators ask about task mining.
Is anything installed on employee desktops?
No. Employee tasks are extracted and mapped with non-intrusive, no-install desktop-recording software, and without asking people to change how they work. Deployment and data handling are set out on the Security & Deployment page.
What happens when people switch between tasks constantly?
Interrupted work is still measured correctly. Even when tasks are frequently interrupted and people move between them, each task is documented separately, with its own activities, durations and outcome.
Does it work for spreadsheet-heavy work?
Yes. Task recognition holds regardless of the number and type of systems involved, including work that is eight hours a day of complex spreadsheet manipulation.
How does Deep Task Mining relate to Process Mining?
Process Mining answers where the work goes across systems and teams. Deep Task Mining answers what work was actually accomplished inside the applications. Both feed the same Process Intelligence.
AI in this capability
This is where AI does the most work.
Recognition is AI-driven. Approval is not.
Recognizing that a screen belongs to the claims application is one thing. Recognizing that the employee rejected a claim, on what basis, against which policy number, is what turns activity into business meaning — and that is AI work.
- Activity and microactivity recognition identifies what was actually done inside a screen, not just that the screen was open.
- Business-data recognition identifies the business data on that screen and assembles it into a data dictionary tied back to the process.
- Decision recognition identifies where the work branched and on what basis, so variation can be explained rather than only counted.
- A process engineer reviews, corrects and approves recognised output at defined checkpoints before it is used to assemble anything. Corrections feed back and improve subsequent recognition.
Time is not throughput
Knowing where the time went is not enough.
Illustrative claims-processing example. The same three hours and fifty-two minutes, read two ways.
- 3h 52mspent in the claims application
This shows where time was spent.
- 41 claims approved — per case, per team
- 12 rejected — decision branch and recorded reason
- 9 re-opened — after a correction
- 6 stalled — where work stopped, waiting on a third party
This shows what work was actually accomplished.
Stereologic connects this task-level understanding with the end-to-end process — cases, flows, handoffs, variants and outcomes.
Illustrative example for explanation. Not an actual client result.
Take the next step
Find out where the task time is actually going.
Choose one task your team believes it understands. We will show what task-level discovery reveals about its steps, its interruptions and its real duration — before anyone builds anything.