Tech, ExplainedEnterprise SaaS and software
Part

Task mining

How the work moves across each desk.

INSIDE: PROCESS MININGHOW IT WORKS · THREE WEEKS OF CLAIMS CHECKING IN BHUBANESWARA small programrecords thescreens of 60claims employeesfor three weeksComputer visionrecognises eachapplication,field and clickAbout 9,000claims arejoined into onemap of the workEach claim takes14 minutesacross fourapplicationsFour of thoseminutes go intocopying thepolicy numberand nameA robot takesover thecopying: about600 hourssaved everythree weeksALSO TALKS TORobotic process automation (RPA)INSIDE: PROCESS MININGHOW IT WORKS · THREE WEEKS OFCLAIMS CHECKING IN BHUBANESWARA small program recordsthe screens of 60 claimsemployees for three weeksComputer visionrecognises eachapplication,field and clickAbout 9,000 claims arejoined into one map ofthe workEach claim takes14 minutes acrossfour applicationsFour of those minutes gointo copying the policynumber and nameA robot takes over thecopying: about 600 hourssaved every three weeksALSO TALKS TORobotic process automation (RPA)INSIDE: PROCESS MINING · THREE WEEKS OF CLAIMSCHECKING IN BHUBANESWARA small program records the screens of 60claims employees for three weeks↓Computer vision recognises eachapplication, field and click↓About 9,000 claims are joined into onemap of the work↓Each claim takes 14 minutesacross four applications↓Four of those minutes go into copying thepolicy number and name↓A robot takes over the copying: about 600hours saved every three weeksALSO TALKS TORobotic process automation (RPA)
Task mining records the screens of a claims team, recognises each application and click, and shows that four of every 14 minutes go into copying data that a robot can then take over.
opens a page inside this one a page in the same scene a page in another scene a concept page a label only, with no page

Task mining is software that records how employees do their work on screen, using computer vision, and builds a map of how work moves between people and applications at each desk. It answers a question that process mining cannot: what happens between two entries in a system.

Process mining reads the event logs that systems keep. But much of an employee's work leaves no log. She reads an email, looks something up in a spreadsheet, or copies a number from one window into another. Task mining installs a small program on each employee's computer. With the employees' knowledge, the program captures what is on the screen at each step. Computer vision, software that recognises what is in an image, works out which application is open, which field is being filled and which button is clicked. Joined together, these steps show how the work is done at each desk.

Consider the claims team of a health insurer in Bhubaneswar, where 60 employees check hospital bills. The company runs task mining for three weeks, and records about 9,000 claims. The result shows that each claim takes 14 minutes on average. In that time, an employee switches between four applications: the claims system, the policy system, a spreadsheet of hospital rates and email. Four of the 14 minutes go into copying the policy number and the patient's name from one application into another, and checking them. Six minutes go into reading the hospital's bill, which needs judgement.

So the company automates the four minutes of copying with a software robot, and leaves the six minutes of judgement with its staff. Across 9,000 claims, four minutes each comes to about 600 hours of work saved every three weeks.

Recording screens raises privacy questions. The screens show patients' names and medical details, as well as the employees' own work. India's Digital Personal Data Protection (DPDP) Act requires a clear purpose and care with such personal data. These duties apply from 13 May 2027. So task-mining tools blank out personal data on the screen before storing it, record only work applications, and report on the process, not on each employee's speed. Companies usually tell their staff in advance what is recorded, and why.

Tech, Explained is a part of ingrid.fyi. Every page explains one piece of technology in plain words.