Before You Implement AI for HR, Do These Three Things First

A practical sequence for HR transformation — from data foundations to workforce intelligence.

Artificial intelligence has moved to the top of almost every HR leader’s agenda. Boards are asking about it. Vendors are demonstrating it. Employees are already using it, with or without permission. And so, predictably, a wave of organizations are racing to “adopt AI” in their people function, often without pausing to ask whether HR operations are actually ready to absorb it. Too many still run on fragmented systems, inconsistent employee records, disconnected reporting processes, and unprepared teams. In many cases, the challenge is less about AI capability than organizational AI readiness. 

The early results of that race are sobering. Gartner’s research finds that only about one in five AI investments deliver measurable ROI, and only thirty nine percent of technology leaders are confident that their business’s current AI investment will yield a positive impact on financial performance. 

But there are organizations that DO get value and they tell a remarkably consistent story: they invest up to four times more in their data and analytics foundations than peers who chase tools first. Roughly 60 percent of their AI spend goes to data quality, governance, and people — not platforms.

That pattern is even more pronounced in HR than in most functions. People data is messier, more political, and more fragmented than customer or financial data. In Southeast Asia, that complexity is often amplified by multi-entity structures, localized compliance requirements, and disconnected operational workflows across countries or business units. Workforce data frequently lives across payroll systems, attendance applications, recruiting platforms, spreadsheets, and manually maintained processes that no one fully owns. Layering AI on top of that landscape without first cleaning it up is not transformation. It is automation of confusion, at higher speed.

DataOn works with several thousand employers across Southeast Asia and we have come to believe the question is not whether HR should adopt AI. That debate is over. The harder, more useful question is: in what order should companies adopt it?

There is a sequence that works. And there is a sequence that wastes budgets. This article lays out the three steps we ask every client to walk through — in order — before they activate AI as part of their HR operating model.

Why “Where Do We Start?” Is Still the Most Honest Question in HR Tech

When we sit down with HR and business leaders, the conversation almost never opens with “which AI model should we use?” It opens with some version of: we know we need to do something, but we are not sure where to start.

That is the right instinct. The companies that struggle most are the ones who skip this very question.

Two failure patterns appear again and again.

The first is the leap-to-AI pattern.
The organization buys an analytics product, an AI-driven recruiting tool, or a generative-AI assistant, and then discovers, weeks into the project, that the underlying employee data is incomplete, duplicated across systems, or locked inside spreadsheets that only one person knows how to read. The tool works. The data doesn’t. The insights are unreliable, trust erodes quickly, and the project gets quietly shelved, while the underlying business problems — slow hiring decisions, rising overtime costs, retention blind spots, and workforce planning uncertainty — remain unresolved.

The second is the system-without-readiness pattern.
The organization has invested in a modern HRIS. The data is reasonably clean. But the HR team, and the line managers they support, have not been equipped to interpret data-driven insights, let alone act on them. Dashboards get built and ignored. Predictive signals go unread. The technology is fine; the operating model around it never caught up.

Both failures point to the same underlying truth. AI in HR is not a software purchase. It is an operating-model change. Treating it as a buying decision is exactly how organizations end up in the four-out-of-five group that sees no ROI.

The good news is that the path forward is not complicated. It is sequential.

Step 1: Build a Centralized HR Data Foundation

The first step has nothing to do with AI. It has to do with knowing (accurately, in one place) who works for you, what they do, what they earn, how they perform, and how they move through the organization.

Most companies dramatically underestimate how scattered their employee data actually is. It is not unusual that a typical mid-to-large employer in Indonesia runs payroll in one system, attendance in another, recruiting in a third, learning in a fourth, and stores everything else, including performance reviews, organizational charts, compensation history, leave balances, exit interviews, in some combination of spreadsheets, shared drives, and individual managers’ inboxes. Each silo has its own definition of “employee,” its own data updating schedule, and its own quirks. Reconciling them is a quarterly project that no one wants to own.

You cannot run useful AI on top of that landscape. AI is a pattern-recognition technology; it amplifies whatever signal, or noise for that matter, is in your data. If your headcount number disagrees with itself across three systems, an AI model will not resolve this disagreement. It will encode it.

The foundation step is unglamorous but non-negotiable: consolidate the core.

A modern HRIS should hold a single, authoritative record for every employee — organizational structure, compensation, attendance, payroll, leave, performance, and the audit trail that connects them. Every downstream process and every future analytics use case will inherit the quality of that record.

In practical terms, the work at this stage is mostly about discipline rather than software. Define the master data. Decide who owns each field. Eliminate parallel spreadsheets. Resolve the duplicate-employee problem. Standardize position titles and cost centers across business units. None of this requires AI. All of it is a prerequisite for AI.

Companies that do this step properly often discover something surprising: the act of consolidating the data already produces ROI, before any machine learning is involved. Compliance gets easier. Payroll errors drop. Reporting cycles shorten from weeks to hours. Managers stop arguing about which version of the headcount is correct.

That is the basis. Get it straight first.

Step 2: Prepare Your Talent and Your Organization

Once the data foundation is in place, the next bottleneck is almost always human. AI in HR fails far more often because of organizational unreadiness than because of technical limitations.There are two readiness gaps that matter most.

Readiness Gap 1:
HR Team Capability

AI does not eliminate the need for HR judgment — it raises the bar on it. An HR professional who used to spend two days assembling a turnover report now sees that report in real time, alongside three contextual signals they have never had before. The skill required is no longer “produce the report.” It is “interpret the signal, ask the right follow-up question, and decide what to do.” That is a different competency, and most HR teams need deliberate development to build it. Data literacy, scenario thinking, the ability to challenge a recommendation rather than accept it on faith — these become core skills.

Readiness Gap 2:
Line Manager and Employee Readiness

AI-supported HR processes only work if the rest of the organization is prepared to participate in them. If managers do not trust the system’s flight-risk signal, they will not act on it. If employees do not understand how their development data is being used, they will disengage from the platform that captures it. Change management is not a sidebar to AI adoption. It is half the program.

This is also the stage where talent strategy itself needs sharpening. The richer your data, the more clearly you can see where the organization is strong, where it is fragile, and where the next two years of capability gaps are forming. That visibility is only useful if you also have the talent management practices (internal mobility, succession planning, structured development, performance conversations that actually happen) to act on it. AI cannot substitute for a weak talent operating model. It can only accelerate a working one.

In our experience, the companies that move through this stage most successfully treat it as a quiet, multi-quarter investment.

They:

• build internal champions. 
• retrain HR partners. 
• write down (sometimes for the first time) how decisions about pay, promotion, and movement actually get made.

By the time they are ready for step three, they have something more valuable than a tool: they have an organization that knows what to do with one.

Step 3: Activate AI and Workforce Intelligence

Only at this point (after clean data and prepared people) does AI start producing the returns the original business case promised.

The use cases at this stage are no longer abstract. They become concrete because the foundation supports them. A workforce intelligence and people analytics layer running on a unified HR data set can show you the trends that previously took a consultant six weeks to find: where overtime costs are accelerating, which teams have unusual attrition patterns, where performance and compensation have drifted out of alignment, which roles will face supply shortages over the next planning horizon.

Generative and agentic AI capabilities come into play here as well, not as replacements for HR judgment, but as a way to compress the time between question and answer. A line manager can ask a natural-language question and get a defensible response from the same data the HR team would use. A recruiter can have a first-draft job description, screening rubric, and interview guide ready in minutes instead of days. An HR leader can run scenarios on what happens to our payroll cost if we shift this team to a different structure, or open a new site, or absorb an acquisition without commissioning a project.

The shift, when it lands, is qualitative as much as quantitative. HR moves from being a function that reports on what happened to one that anticipates what is coming. It moves from describing the workforce to shaping it. That is the real promise of AI in HR, not faster paperwork, but better strategic decisions about people.

More importantly, this is also the stage at which AI becomes a competitive differentiator. The companies that arrive here first will set the benchmarks that the rest of the industry has to meet on hiring speed, retention, internal mobility, and workforce cost efficiency. 

The companies that skipped steps one and two will be trying to catch up with foundations they should have built two years ago.

Why the Sequence Cannot Be Reversed

The temptation, particularly at the board level, is to start with step three. AI is the visible part of the program. It is what gets the headline and the budget approval. Step one (data consolidation) and step two (capability building) feel like infrastructure. They are less prominent and therefore receive less publicity.

But the sequence is not arbitrary. Each step earns the next.

Skip step one, and AI produces unreliable insights from unreliable data. Skip step two, and reliable insights go unused. Both failure modes are common, and both are expensive. Doing the steps in order is slower at the start and dramatically faster at the end. It is also the only path we have seen consistently produce the kind of returns Gartner’s top twenty percent is reporting.

That doesn’t mean the three steps have to run sequentially in calendar time. The high-ROI organizations in Gartner’s research tend to invest in foundation, capability, and AI activation in parallel. What distinguishes them is proportion, not order. The majority of their AI spend goes to foundations and people; the remainder goes to tools. The sequence is a way of thinking, not necessarily a project plan. What we have seen consistently fail is when the proportion is inverted: tools first, foundation later, capability never.

The most useful perspective for a leader weighing where to begin is this: AI in HR is not a product you buy. It is a transformation you sequence. 

The buying decisions get easier once that is understood. The pace gets more honest. And the eventual investment in AI tools produces returns rather than press releases.

A Note on How We Approach This at DataOn

The three-step sequence in this article is the same one we use to guide HR transformations through SunFish HR, our integrated HR platform, and the SunFish AI Hub that sits on top of it.

SunFish HR is designed to support that progression, from centralized HR Core and payroll foundations, through talent management and workforce readiness, and into workforce intelligence and AI-assisted decision support. The architecture mirrors the sequence described in this article: building the foundation first before activating the intelligence layer on top of it. 

We mention this not because the framework requires a specific vendor, but because most HR leaders we speak to want to know how the principles translate into something they can actually deploy. If you would like to explore what the three-step path looks like for your organization specifically, the team at DataOn is happy to walk through it.

When Should You Start?

The honest answer is: now — but not at step three.

Begin by being clear-eyed about where your HR data actually lives, how clean it is, and who owns it. Be equally clear about whether your HR team and your line managers are equipped to use data-driven insight if you put it in front of them. Then build accordingly.

The companies that approach AI in HR as a sequenced HR digital transformation rather than a one-off purchase are the ones that will look back, three years from now, and recognize that the conversation about “AI for HR” was really a conversation about how their entire people function evolved. Technology will be the part everyone talks about. The foundation will be the part that actually made it work.