| Last Week in HR AI: Issue #19 | Week of November 2, 2026 |
Your next agent will hit the workarounds nobody wrote down
IHG, Deutsche Bahn, and two more Oracle customers say process, data, and legal review gate every agent.
|
1 Nature2 koreaherald.com3 SHRM4 hartfordbusiness.com
Ikona's Take on the past week (November 2, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
Every agentic HR story worth reading this week was really a story about groundwork. The through-line: the return on HR AI is decided before the agent is switched on, by whether your processes, data, and task definitions are documented well enough for software to act on them. SK Telecom shows it in a single number. AI-written descriptions of fields across roughly 9,000 databases lifted answer accuracy from about 40% to over 80%. The lift came from describing the data properly.
The Oracle customers described the same thing from the other side. Once AI touched their processes, it exposed regional workarounds nobody had written down. Providence reports 90% less manager admin work, and its first lesson was to fix the process before automating it. Now Workday and Dayforce are breaking jobs into tasks and splitting them between people and software. That generated task list becomes a record that works councils and European Union (EU) AI Act reviews may scrutinize. If you cannot name its owner today, you have learned something about your readiness. And do not expect the tools to close the skill gap for you. In the Nature Human Behaviour trial, consultants performed better with ChatGPT and were no better at technical questions once it was taken away.
Before you fund the next agent, ask whether you can produce the documented process, the defined data, and the named owner that agent will depend on. Reply and tell me where your team stands. I'd welcome the conversation.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
Researchers from BCG, Boston University, and OpenAI ran a randomized controlled trial with 487 management consultants, giving some of them ChatGPT access and training before assigning data science tasks outside their expertise. The treated group scored higher than the control group on three tasks but did not reach the level of expert data scientists. When tested later without the tool, the treated consultants did no better on technical questions. The study appears in Nature Human Behaviour.
Workday's Work Intelligence and Dayforce's Strategic Workforce Planning Agent both break jobs into tasks and divide them between people and software; one is in early access and the other is due in 2027. The author contends that the task lists these tools generate will become a new system-of-record artifact. Redundancy consultations, works councils, and reviews under the European Union (EU) AI Act could all examine those lists. The piece recommends a named owner, manager validation, versioning, and access boundaries.
Four Oracle human capital management (HCM) customers spoke at Oracle's AI World about their agentic HR plans: IHG, UL Solutions, Deutsche Bahn, and Acosta Group. Each described progress as gated by process redesign, legal review, data cleanup, and token costs. Applying AI surfaced regional workarounds in existing processes, and legal review centered on intent and human oversight. The customers said early data cleansing and a single-platform data landscape speed adoption, and that no agent goes live without effort.
Also covered by: hrchiefmagazine.com, futurumgroup.com
|
From Ikona
Your task list and data definitions live in people's heads
We diagnose how work and knowledge actually flow through your HR function and capture the judgment that lives in people's heads. Then we structure it into a knowledge asset your team owns and your AI tools can stand on. Talk to us about yours → |
More than 7,500 SK Telecom employees use AI agents each month, the company says, and over 4,000 have built their own, about 1,500 of which are in regular use. The carrier is moving from individual tools to department-wide deployment, where data access and privacy have become the main obstacles. Having AI write descriptions of fields across roughly 9,000 databases raised answer accuracy from about 40% to over 80%. No cost or productivity figures were disclosed.
Providence, a health system with about 119,000 employees, built an agent called RITA with IBM Consulting on watsonx Orchestrate. RITA runs across its existing Oracle HCM and recruiting systems, automating requisitions, approvals, and internal transfers. Since the August 2025 launch, Providence reports 90% less manager administrative work, a 60% lower cost per hire for internal transfers, and internal moves making up more than 35% of hiring. Its stated lessons are to bring IT in early and fix the process first.
Also covered by: newsroom.ibm.com, ibm.com
XPO says AI tools helped its North American freight unit handle nearly 6% more shipments in the third quarter while hiring more slowly. An in-house forecasting model now matches service center staffing to expected volume, and other tools plan dock assignments, routes, and loading feedback. Damaged freight fell 20%. Payroll costs still rose 11%, mostly from higher volume, wage inflation, and incentive pay, and management is targeting 1.5 to 2 points of productivity gain a year.
Ikona's post argues that six HR leadership roles face six different mandates but depend on the same missing input: an honest picture of how HR actually runs, beyond the org chart and the documentation. That is the groundwork this week's agent stories keep pointing to, because no agent can act on a process, data set, or task list that nobody has captured.
The SK Telecom number is the one I keep coming back to: accuracy jumped from about 40% to over 80% because someone described what the fields across those 9,000 databases meant, not because anyone found a smarter agent. It is the least glamorous win in this whole issue, and it is within reach of most HR teams well before a Workday or Dayforce task list lands in front of a works council. On Monday, pick the one dataset your next agent would query first, ask who on your team can explain every field in it without opening a ticket, and treat any silence as the first line of your roadmap.
If your team is wrestling with workforce intelligence, AI readiness, or a transformation that has to land, we'd love to help. Reach out and tell us what you're working on.
Get in touch →Last Week in HR AI
Practitioner perspectives and diagnostic frameworks, delivered the week they publish. Be the first to know when new research drops.
No spam, ever. Unsubscribe anytime.