As more Indonesian organizations digitalize human resources, the role of HCM technology is beginning to change. The first stage of that transformation focused largely on replacing paper, spreadsheets and disconnected processes with systems that could manage employee records, attendance, payroll, recruitment and performance more consistently. But as these digital foundations mature, HR leaders face a new question: what comes next?
Artificial intelligence appears to offer the next step. Yet adding AI to an HCM platform does not automatically take HR management to a higher level. A chatbot may make information easier to retrieve, and automation may make an existing process faster, but neither necessarily improves the decisions being made. The larger opportunity is to combine AI with the workforce data and business context already captured by HCM systems, allowing organizations to move from recording what happened to understanding why it happened, anticipating what may happen next and deciding how to respond.
That shift changes the ambition of HR technology. A conventional HCM system can report how many employees resigned, how payroll changed or how long recruitment took. An AI-powered decision engine could help explain the factors behind rising turnover, identify an unusual payroll transaction before salaries are processed or determine which applicants warrant a recruiter’s attention. It does not merely produce more information. It creates a path from data to analysis, from analysis to recommendation and from recommendation to a controlled human decision.
At the Indonesia Human Capital and Beyond Summit 2026, DataOn founder and CEO Gordon Enns argued that this is the next stage of HCM development. The goal is not to place AI on top of existing HR processes as another interface. It is to connect reliable data, appropriate analytical models, clear permissions and human judgment so that AI can support decisions management can understand, examine and act upon responsibly.
From Systems of Record to Systems of Decision
Digital HCM gives organizations something AI needs: a structured history of the workforce. Employee movements have dates, payroll and attendance exist at transaction level, roles follow a defined organization structure, and performance, recruitment and learning records can be connected to the same employee identity. When these foundations are consistent, the system can begin to analyze relationships and changes over time rather than simply retrieve isolated records.
Many organizations have not reached that point yet. HR.com’s State of People Analytics 2025-26 found that only 26 percent of organizations often or always combine HR data with non-HR data, while 41 percent rarely or never do so. Even within HR, the information behind one decision may sit across payroll, attendance, performance, recruitment and engagement processes.
This fragmentation limits what AI can reliably conclude. A model assessing resignation risk, for example, may need to consider tenure, overtime patterns, manager changes, internal applications and time since the employee’s last pay or role change. If those histories cannot be connected, the organization receives an incomplete answer regardless of how advanced the model appears.
“You need data and technology to achieve results, in that order,” Gordon Enns said. “Prediction on unreliable reporting produces confident nonsense.”
Five Levels of Decision Support
A useful way to judge AI capability is to separate five levels of decision support. The first is reporting: accurate and agreed information about what happened. The second level is diagnostic, where the system helps explain why a result changed by identifying the relevant drivers, differences or anomalies. At the third level, the predictive one, the system estimates what is likely to happen. It might assign a resignation risk band, forecast a future quantity or flag a transaction that does not fit the expected pattern. The fourth level, prescriptive capability, goes one step further; it recommends a response and explains the constraints behind it. At the highest, the agentic level, AI can retrieve information or carry out specifically permitted actions across systems, subject to approval and an audit trail.
These levels build on one another. An organization cannot make dependable predictions when it still disputes basic workforce numbers. It should not allow an AI agent to act when access rights and approval responsibilities remain unclear. Each new level depends on the data definitions and controls established below it.
The maturity ladder also clarifies the role of generative AI. A language model can understand a question and explain an answer in accessible language, but another analytical method may produce the underlying result. Classification can place an employee into a risk band. Forecasting can project a future cost or workload. Anomaly detection can identify a payroll entry that requires attention. Generative AI helps the user investigate and understand those findings.
What This Evolution Enables
The value becomes clearer when analysis enters the workflow while a decision can still change. In payroll, anomaly detection can identify negative net pay, an unusually large deduction or salary continuing for someone recorded as resigned before the pay cycle closes. The system can then explain what changed and direct the payroll team to the affected records.
Predictive resignation analysis changes a different decision. Instead of reviewing turnover only after people leave, HR can examine where risk appears to be increasing and which factors contribute to it. A pattern concentrated in one team may point toward workload, management or compensation. The purpose is to direct attention and support an informed conversation, not to label an employee or predetermine an outcome.
Recruitment offers another example. AI can convert submitted resumes into structured candidate information and compare that information with the requirements of a role. Recruiters receive a prioritized order for reviewing candidates and can see why a candidate appears relevant. The recruiter still decides who advances, but spends less time retyping information and searching through applications that do not meet the stated criteria.
The practical change in each case is timing. The system brings relevant evidence forward while payroll remains open, retention action is still possible or a recruitment shortlist is still being formed. That gives managers an opportunity to influence the outcome instead of receiving a report after the event.
Across these examples, one benefit remains consistent: the result can stay grounded in company facts. Recommendations can link back to current employee records, historical patterns or approved policy documents. Managers receive the conclusion together with the evidence needed to examine it, question it and decide whether to act.
DataOn is applying this model within SunFish HR through payroll diagnostics, predictive resignation, recruitment support, conversational access and controlled access for external AI assistants. These capabilities serve different decisions, but follow the same principle: the system must show where the answer came from, respect the user’s authority and keep consequential action under human control.
Governance Determines the Value
The use of AI in HCM introduces responsibilities that ordinary process automation does not. Indonesia’s Ministerial Circular on Artificial Intelligence Ethics identifies transparency, humanity and security among the principles for responsible AI. In workforce decisions, those principles require practical controls: permission before access, explanation before intervention, named approval for consequential actions and an audit trail after the event. After all, explaining how to find an HR form carries little risk, but ranking applicants, identifying resignation risk or changing pay-related information requires stronger oversight.
Human review is therefore part of the decision engine rather than a correction applied after it. A resignation-risk flag should prompt a manager to look more closely at the working environment and speak with the employee where appropriate. It should not become a performance rating or an unchallengeable fact in the employee’s record.
Organizations also need to monitor whether the capability works after launch. Predictions should be compared with actual outcomes. Payroll warnings should be checked against confirmed errors. Managers’ acceptance or rejection of recommendations can reveal whether a model, workflow or use case needs adjustment. Business conditions and employee behavior change, so accuracy can drift even when the software remains unchanged.
What HR Leaders Should Ask Next
HR leaders evaluating AI should look beyond the number of features shown in a demonstration. They should ask whether a prediction uses their own history or a generic model, whether the factors behind a score are visible, where employee data is processed and whether it trains a shared model. They should also ask what the AI can write back, which actions require approval and how every access or action is logged. These questions separate convenient AI from accountable decision support.
And that is precisely what the next stage of HR digitalization will be about: not how many AI features appear inside an HCM platform, but whether those capabilities help organizations understand what is changing, anticipate what may happen next and make better decisions while keeping results explainable, actions authorized and people accountable. The goal is not to automate judgment. It is to give judgment better evidence, at the moment a decision can still be changed.
HCM technology began by creating more reliable processes and records. Its next contribution can be to make those records useful at the moment management has to choose what to do.
