| Last Week in HR AI: Issue #11 | Week of September 7, 2026 |
IAG doubled HR deflection by rewriting 730 knowledge articles
Plus UBS ties pay to AI fluency, and Claudeforce moves CRM inside Claude. The model was never the bottleneck.
|
1 infosectoday.io2 Human Resources Director3 Federal Reserve Bank of St. Louis4 VentureBeat
Ikona's Take on the past week (September 7, 2026) in HR + AI
|
IO
|
Ian O'Keefe
Co-founder & CEO, Ikona Analytics
|
IAG's number is the one I keep coming back to: HR self-service deflection went from 30 percent under the old chatbot to a 67 percent average in Now Assist's first month. The unlock was not the model. It was reformatting roughly 730 knowledge articles so a machine could read them. Kansas City's eight agents work for the same reason: they are grounded strictly in city policy, which someone had to write down first. That is the week's through-line. The value is in the machinery around the model: readable knowledge, policy grounding, incentives, and honest job redesign.
The St. Louis Fed shows what happens when you skip that work. Across nearly 14,000 workers, adoption is a mile wide and an inch deep, with most occupations below 50 percent, and exposure scores predicting which occupations adopt but barely predicting which individuals do. Learning beats job content. UBS drew the obvious conclusion and made AI fluency a formal factor in hiring, promotion, and pay, backed by a weekly power hour. That is incentive design, not procurement.
The manager research is the uncomfortable one. Only one in 10 managers would miss having direct reports, and employees sometimes preferred AI-driven promotion feedback to an average manager's. An agent does not rescue a role that was already failing; it just surfaces the failure faster. So the question I would put on your staff meeting agenda: which of your HR processes actually has its rules written down? Reply and tell me where you landed. I read every one.
| The six to read | The stories that matter for the Office of HR — with our take on each. |
Kansas City has deployed a set of named AI agents built on Microsoft 365 Copilot to help managers with coaching conversations, discipline documentation, interview questions, and benefits queries at any hour. The roster has grown from three agents to eight. Each is grounded strictly in city policy and was tested for bias and privacy exposure before release. Manager use is voluntary, and final accountability stays with the human.
Australian insurer IAG swapped ServiceNow's basic HR chatbot for the Now Assist AI suite to handle employee self-service queries. According to the company's digital engagement manager, the project required reformatting roughly 730 HR knowledge articles into machine-readable form, using Copilot with subject-matter expert review. Deflection rose from 30 percent with the old chatbot to a 67 percent average in Now Assist's first month, against a target of 80 percent. Agentic HR workflows are the next phase.
Also covered by: itnews.com.au, traveltrade.today
UBS has made AI proficiency a formal requirement for hiring, promotions, and pay starting September 2026, embedding it as a permanent factor in performance reviews at the wealth manager. The bank tracks AI tool usage at an aggregate, departmental level rather than monitoring individuals, and runs a weekly "AI power hour" training session to build staff capability.
|
From Ikona
Your AI stalls where your rules were never written down
We diagnose how work and knowledge actually move through your HR function, capture the judgment sitting in people's heads, and structure it into something an AI system can stand on. Talk to us about yours → |
Researchers from SAP SuccessFactors' Future of Work Research Lab and Gartner argue that firms flattening management layers while handing managers AI tools are automating a role that needs rebuilding. Their data shows only one in 10 managers would miss having direct reports, and the long-standing engagement premium managers held over individual contributors has vanished. Ninety-three percent of managers say AI has changed their job, largely through informal adoption rather than designed change. Employees in some cases preferred AI-generated promotion feedback to feedback from an average manager, and the researchers propose untethering managers from fixed teams in favor of project-based deployment.
Also covered by: hrreview.co.uk, Human Resources Director
A St. Louis Fed working paper introduces task-level measures of generative AI use, based on nearly 14,000 workers surveyed through the Real-Time Population Survey. Use turns out to be broad across occupations but thin within them: most occupations show adoption below 50 percent, and few individual tasks clear that mark. Occupation-level exposure scores predict aggregate adoption reasonably well but do a poor job of predicting which specific people adopt. The authors point to experience and learning as stronger drivers than job content.
Salesforce and Anthropic launched Claudeforce, which embeds Salesforce CRM inside Anthropic's Claude CoWork app through 37 pre-built sales skills, so sellers can read and update live CRM records without opening Salesforce's own interface. Salesforce executives position the move as extending the platform's value into data and workflows rather than eroding it. The arrangement introduces consumption-based API pricing on top of Anthropic's inference fees. That pricing structure marks a step away from seat-based enterprise software economics.
Dashboards answer "what happened" instantly but go mute on "why," and that gap is exactly what IAG's 730 rewritten knowledge articles and Kansas City's policy-grounded agents just closed. The teams winning this wave are not shopping for smarter models; they are building the context layer underneath them.
The IAG detail I keep turning over is the one nobody would put in a press release: someone sat down and reformatted roughly 730 HR knowledge articles so a machine could read them. That is the least glamorous work described anywhere in this issue, and it produced the biggest number in it, deflection going from 30 percent to a 67 percent average in a single month. Kansas City's agents hold up for the same unremarkable reason, because city policy was written down before anyone tried to ground an agent in it. So on Monday, take the HR process your people ask about most, and go find out whether its rules live anywhere other than in one person's head.
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.