AI in ERP, EPM and HCM: Where It Already Works and Where It Is Still on Slides

Artificial intelligence is on the agenda of every board meeting. Yet there is a gap between “we have AI” and “AI makes us money”, and you cannot cross a gap in two jumps. Here is a look without the magic: what AI can already do in an ERP system, in budgeting and in payroll, where it is too early to trust it, and where an owner should start.

The report card: a pass, but only just

In spring 2026 Gartner analysts surveyed 204 finance leaders and gave finance AI something like a school report card. The picture is honest: 66% of the organisations that adopted AI name higher productivity as a top benefit, yet 63% say implementation was slower than expected.

Another figure from the same survey explains why boards get bored by AI updates: 45% of AI investments lean towards productivity and only 20% towards decision quality. The first means doing the same thing faster. Owners need the second.

The money is flowing anyway. Gartner estimates that by 2027, 62% of ERP spend will go to applications with generative AI, up from 14% in 2024. The train has left the station. The question is whether you have a seat or are hanging on to the door.

Where AI can be used today and where there is no need to rush

ERP: AI arrives through the conductor

In Oracle JD Edwards EnterpriseOne AI is not hidden behind one magic button. The system has Orchestrator, an automation tool that can call external services. Since Release 24 it has been able to connect to Oracle Cloud AI services: document understanding, computer vision and generative models. And the JD Edwards roadmap lists the Model Context Protocol, a standard that lets AI assistants talk to the ERP system by the rules instead of guessing.

What this means in practice:

• A supplier invoice arrives as a PDF. The service reads the details and amounts, Orchestrator matches them to the purchase order and creates the document. The accountant reviews exceptions instead of retyping numbers.
• A camera on the line or in the warehouse spots a defect, and an inspection request appears in the system automatically.
• A manager asks a question in plain language and gets a summary of system data instead of an export with forty tabs.

One caveat: Orchestrator does not invent anything. It executes a documented process. If your purchasing procedure lives in one person’s head, AI will inherit the same chaos, only faster. We covered linking processes and money in project business in the article on Primavera P6 and JD Edwards.

EPM: a forecast that explains itself

Budgeting adopted AI before anyone else: it has a history of numbers, and machines learn well from history. IBM Planning Analytics includes an assistant built on watsonx. You can type “show me last year’s revenue for mobile phones” and it builds the right view of the data.

Three capabilities make it worth a look:

• Multivariate forecasting. The system works out which variables affect the result most: exchange rate, season, raw material prices.
• Cell explanation. Click a number and see the formulas, assumptions and sources behind it. The “where did this figure come from” argument ends in a minute.
• Agents. Through watsonx Orchestrate the assistant not only answers but also triggers actions: approvals, scenario recalculation, updates to connected systems.

IBM itself reports that its own financial planning reached model accuracy of about 95%. Your number will be your own, but the direction is clear: people stop assembling the forecast and start discussing it. If your budget still lives in spreadsheets, start with our article “Budgeting in Excel vs IBM Planning Analytics”.

HR and payroll: trust, but verify

A payroll error is the most expensive kind because every employee sees it. The figures are telling. In a Symmetry survey, 78% of payroll teams already use AI, but only 45% trust it to stay tax-accurate. In another study, 98% of HR professionals said they do not trust generative AI to make workforce decisions.

That is a sound attitude. In payroll, AI works best as a second reviewer, not as the calculator:

• flagging anomalies before payday: pay that doubled, a timesheet that does not match the schedule;
• answering routine employee questions about leave or a payslip;
• supporting workforce analytics: turnover, overtime, payroll budget.

The calculation itself has to stay predictable: the same data gives the same result, and every amount can be traced back to its source entries. That is the principle behind RBC HCM Solutions. For how to change systems without missing payday, see our article on HR and payroll migration.

Three conditions AI cannot do without

1. Clean data. Garbage in, garbage out, only now in a confident tone. AI will not fix duplicate vendors or three versions of the item master. It will multiply them.
2. A documented process. You can automate what is written down. “It has always been done this way” does not turn into an algorithm.
3. A human in the loop. A person is accountable for the payment, the journal entry and the payslip. AI proposes, a person approves, at least where a mistake costs money.

Where to start: a 90-day plan

1. Pick one process with a lot of routine and a clear cost of error: incoming invoices, sales forecast, payroll checks.
2. Record the baseline: hours and errors today. Without a “before” you cannot show an “after”.
3. Clean up the data for that process only, without trying to fix everything at once.
4. Run a 6–8 week pilot with an owner from the business, not only from IT.
5. Decide by the numbers: scale, rework or stop. A stopped pilot is a result too. Measure twice, cut once.

Frequently asked questions

Do we need to replace our ERP system to get AI?

Usually not. In JD Edwards, AI services connect through Orchestrator; in IBM Planning Analytics the assistant and forecasting are part of the product. Get the most out of what you already own first.

Will AI replace accountants, planners or payroll specialists?

It replaces manual entry and stitching spreadsheets together. Review, judgement and accountability stay with people, and that is exactly what they gain time for.

Is it safe to give AI financial and HR data?

That is a matter of architecture, not faith. You need to know where the data is processed, who has access and what is logged. For personal and payroll data this is settled before the pilot, not after.

Sources

• Gartner: CFOs struggle to turn AI into business value (survey of 204 finance leaders)
• Gartner: 45% of finance AI investments lean towards productivity
• Gartner: share of ERP spend on applications with generative AI by 2027
• Oracle: Artificial Intelligence with Oracle JD Edwards EnterpriseOne (PDF)
• Oracle: JD Edwards roadmap (PDF)
• IBM: Planning Analytics Assistant and Multivariate AI Forecasting
• IBM: agentic capabilities in IBM Planning Analytics
• IBM: its own planning and reporting case study
• Symmetry: AI in payroll statistics, 2026 survey
• HR Dive: what CHROs think about AI