Shadow AI in HR begins where payroll preparation requires too much manual work

Published on 1 June 2026

AI is now present in the workplace. Also in HR, planning and payroll preparation.

Employees use AI to write texts, summarise documents, analyse data, find errors or detect patterns. Often, this happens with the best intentions. People want to work faster. They want to reduce manual corrections. They want to understand more quickly where something is going wrong.

But that is exactly where a new risk emerges: shadow AI.

Shadow AI refers to the use of AI tools by employees or teams without formal approval, control or governance from IT, security or management. Think of a planner uploading an Excel file with employees, customers, sites and availabilities into a public AI tool. Or an HR employee asking a chatbot to summarise payroll data, absences or internal rules. Or a payroll team using a standalone AI tool to interpret complex allowances or mobility reimbursements.

It may seem efficient. Until you realise what kind of data may be involved.

Employee names.
Hours and performances.
Absences.
Mobility.
Allowances.
Internal wage agreements.
Project information.
Customer data.
Site locations.
Certificates.
Costs and margins.

So the problem is not that HR or payroll wants to use AI. The problem is that sensitive data ends up outside the controlled business flow.

According to Gartner, 69% of surveyed organisations suspect or know that employees are using prohibited public GenAI tools. Gartner warns of risks such as IP loss, data exposure and increased security risks. Gartner also predicts that by 2030, more than 40% of enterprises will experience security or compliance incidents linked to unauthorised shadow AI.

In 2025, Gartner also identified shadow AI as a top concern for risk leaders.

But for HR, pre-payroll and operations, the key question is not only:
Which AI tools are our employees using?

The better question is:
Why do they need those standalone AI tools in the first place?

That is where the real business case lies.

Because shadow AI rarely comes from bad intentions. It emerges where processes are too slow, too fragmented or too manual. Where people have to copy data. Where Excel remains the emergency bridge. Where exceptions only become visible at the end of the month. Where planning, registration, approval, rules and payroll preparation are not connected on the same data line.

In other words: shadow AI is often not an IT problem. It is a symptom of operational friction.

HR and pre-payroll work with sensitive and complex data. It is not just about "hours". It is about the interpretation of those hours.
  

• Was someone scheduled or actually present?
• Was there mobility?
• Was there a deviation?
• Was there a surcharge applicable?
• Is the performance approved?
• What regime, which collective labor agreement, or which internal rule applies?
• Which wage code should ultimately go to the payroll service provider?

Today, that shortcut is increasingly AI.

Payroll preparation is often the endpoint of all operational deviations. Anything that was not correctly captured in planning, registration or approval eventually ends up with HR or pre-payroll.

Just before payroll closing, everything needs to become clear quickly.

Which hours are correct?
Which mobility should be taken into account?
Which allowance applies?
Which absence affects the calculation?
Which exception still needs to be approved?
Which wage code belongs to this situation?

When that information is scattered, AI becomes attractive. An employee copies a situation into an AI tool and asks:

“Which wage code should I use?”
“How should I interpret this mobility allowance?”
“Which allowance applies to this performance?”
“Create a formula for this exception.”
“Check this Excel file with hours and premiums.”

That may seem harmless. But payroll cannot rely on guesswork.

An AI tool can give a convincing answer and still be wrong. Especially when it involves collective labour agreement rules, internal agreements, mobility, premiums, overtime, shifts, night work, weekend work or exceptions. A wrong answer does not only lead to corrections. It leads to discussions, frustration and a loss of trust among employees.

With VIRO, that complexity is not interpreted ad hoc by AI. VIRO applies rules centrally, transparently and reproducibly. Raw registrations are automatically processed into correct wage codes and payroll-ready output. Deviations become visible before payroll closing, not after.

On top of that, Flo can support the process as an AI co-pilot. For example, by flagging deviations, identifying patterns or answering questions based on validated data. But the foundation remains clear: rules, calculations and validations must be traceable.

AI can support payroll. It should not replace it with uncontrolled standalone interpretation.

Many payroll problems do not start in payroll. They start earlier.

In planning.
On site.
During registration.
With last-minute changes.
With sickness, absences, extra assignments or changed shifts.

For example, a planner may hear on Thursday evening that an employee is ill, a site has shifted and a customer needs extra capacity. The schedule for Friday needs to be right again.

When planning is managed in Excel or in a limited visual planning board, the pressure quickly builds. The planner has to take into account availabilities, skills, certificates, shifts, equipment, locations, travel times and internal rules.

That makes it tempting to take an export and ask AI:

“Create a better schedule from this.”
“Find conflicts in this planning.”
“Assign these employees to these sites.”
“Reschedule tomorrow based on availability and skills.”

But that export often contains sensitive data: employee names, customer data, site addresses, certificates, availabilities, work schedules and sometimes project information.

On top of that, AI may suggest a plan that does not work operationally: someone without a valid certificate, a double booking, incorrect shift coverage, an unavailable employee or an unrealistic combination of tasks.

That is why planning should not be optimised outside the process, but within the process.

With SOLUTIO, planning works based on operational reality. Not just based on blocks in a calendar, but on constraints such as availabilities, absences, skills, certificates, equipment, capacity and rules. When changes occur, planning can be adjusted faster and more realistically, without sensitive data having to be copied into standalone tools.

The mobile registration then brings the reality of the field back to the back office: hours, mobility, materials, photos, checklists, digital work orders and deviations.

This turns planning from a standalone schedule into the starting point of a reliable data flow towards approval, pre-payroll and post-calculation.

Management feels the pressure too.

Overtime is increasing.
Mobility costs are rising.
Projects are becoming more expensive than expected.
Post-calculation comes too late.
Teams are deployed differently than planned.
Deviations only become visible when the month is almost closed.

The question is logical:

“Where are we losing money?”

When the right reporting is missing, managers quickly turn to exports. Data from planning, ERP, time registration, HR, payroll and finance is brought together in Excel and then analysed with AI.

“Where are the biggest deviations?”
“Which teams have the most overtime?”
“Which projects deviate from the planning?”
“Why is this site running over budget?”
“Summarise these cost differences for management.”

This may produce insights in the short term. But once again, it creates risk. Project costs, employee data, customer information, rates, margins and internal financial data may end up outside the controlled environment.

On top of that, the output is only as reliable as the input. AI on top of incomplete or incorrect data does not speed up decision-making. It speeds up wrong conclusions.

GO-VIRTUAL takes a different approach. By better connecting planning, mobile registration, approval, VIRO processing and ERP post-calculation, one more reliable data line is created. As a result, deviations become visible earlier and AI can help analyse based on validated information.

Not trying to explain afterwards why the margin has disappeared. But seeing faster where the deviation starts.

Companies cannot solve shadow AI by simply banning AI. Usage will shift to private accounts, browser tools or standalone applications. Employees will continue to look for speed, especially when pressure is high.

The better approach is to offer a controlled alternative.

That starts with the operational flow.

At GO-VIRTUAL, we see that flow as one connected chain:
master data → planning → mobile registration → approval → pre-payroll → payroll-ready output → ERP post-calculation

When that chain breaks, manual work appears.
When manual work increases, dependence on Excel grows.
When dependence on Excel grows, employees look for shortcuts.
And today, those shortcuts are often AI tools.

That is why AI should not sit next to the process. AI must be embedded in a controlled data flow.