Most organisations know about obvious administrative work. Fewer can see the small fragments of manual effort distributed across dozens of roles: downloading a report, renaming a file, copying a value into another system, checking whether somebody replied, rebuilding context from email, updating a status field because another platform cannot, or reconciling two lists before a meeting.
Each activity can look trivial. The commercial problem appears when the same friction repeats across people, customers, orders, weeks and systems. Because the work is embedded inside larger jobs, it is rarely captured as a project, queue or cost centre. It becomes invisible operating overhead.
Hidden manual work is usually not one big task. It is a pattern of small compensating behaviours around weak systems and unclear workflows.
Do not start with a survey asking what should be automated
People are poor at estimating fragmented effort, especially when work has become routine. Asking teams what they want automated tends to surface the most annoying task, not necessarily the most expensive or structurally important one.
Start instead with evidence. Follow a transaction, request, customer, invoice or case from start to finish. Record every system touched, every handoff, every re-entry of information, every waiting point and every manual check required to keep the work moving.
Look for five signatures of hidden manual work
- Repeated transfer: information is copied between email, spreadsheets, CRM, finance, ticketing or shared drives.
- Status reconstruction: somebody has to ask several people or inspect several systems to answer a basic progress question.
- Human reminders: work advances because one person remembers to chase, follow up or escalate.
- Manual reconciliation: two reports or lists must be compared before anyone trusts the result.
- Exception absorption: a person repeatedly fixes predictable cases that the normal workflow does not handle well.
These signals matter because they show where labour is being used to compensate for missing integration, unclear ownership, poor data design or incomplete workflow logic.
Measure frequency before debating technology
For each recurring manual step, estimate frequency, average handling time, number of people involved, waiting time created, error or rework risk and downstream dependency. A three-minute task performed a few times a month may not deserve attention. The same task performed hundreds of times across a team is different.
Do not convert every minute directly into cash. Time saved only becomes economic value if it removes overtime, avoids future headcount, reduces delay, increases throughput, improves control or lets people spend more time on measurable higher-value work.
Use process evidence where it exists
System logs can reveal work that workshops miss. Microsoft continues to expand Power Automate Process Mining and task-mining capabilities, including rework detection, root-cause analysis, process comparison, custom metrics and object-centric process mining for processes that do not fit a simple one-case-one-sequence model. That does not mean every company needs a process-mining platform. It does reinforce the value of event evidence when the real process is too distributed or variable to reconstruct reliably from memory.
A practical approach can range from simple work sampling and transaction tracing to process-mining tools, depending on scale and available data.
Separate removal, redesign, integration and automation
Once the manual work is visible, do not automatically automate it. Some steps should be deleted because the underlying requirement no longer matters. Some should be moved earlier or later in the workflow. Some disappear when systems are integrated. Some remain human because judgment is valuable. Only after those decisions should automation be considered.
This distinction prevents an organisation from spending money to automate work that should not exist.
What better looks like
A better operating environment makes routine work observable. Teams can explain where effort goes, which handoffs create delay, which manual controls are intentional and which are compensating for weak systems. Repeated low-value work is reviewed periodically rather than allowed to accumulate indefinitely.
The objective is not to eliminate humans from operations. It is to stop using human attention as invisible middleware between systems and processes that should carry more of their own operational load.