The problem
As IBM’s workforce grew globally, its HR department faced siloed processes, inconsistent policies across regions and heavy reliance on manual, repetitive work, the company says. Employees met multi-channel experiences that led to low satisfaction, and handoffs delayed responses (IBM). HR professionals spent their days on non-strategic tasks: payslip questions, letters, vacation requests and policy lookups across 7,000 policy pages.
IBM’s diagnosis was that earlier digital tools had automated fragments of the work without changing the experience. It set out to give employees one place to ask and act, and to move HR staff toward higher-value work.
What they did
IBM first introduced AskHR, then a benefits assistant chatbot, in 2017. Chief human resources officer Nickle LaMoreaux told Fortune that the company treated it as a technical change, and “nobody used it” (Fortune). In 2018, IBM told more than 20,000 workers they would have to use the chatbot instead of asking the HR department.
After a backlash, HR began asking managers directly how the chatbot could be better, and redesigned it around what employees valued: real-time, personalized answers drawn from their own career data. IBM says it applied an “eliminate, simplify, automate” principle and a “transform, not transfer” method, so no inefficient process was automated as it stood (IBM).
Today AskHR runs a two-tier model: AI handles routine inquiries, while human advisors take complex needs. It integrates with Workday, SAP and Concur, and automates about 80 tasks, including employment verification letters, vacation requests, payroll access and manager workflows such as transfers, compensation changes and organizational updates. In 2025 IBM rebuilt it on its own watsonx Orchestrate product: large language models classify each prompt, route it to an HR domain such as benefits, payroll or careers, and then trigger a task or generate an answer. Sensitive matters such as low performance or misconduct are still handled by people, LaMoreaux told Fortune.
IBM is both the user and the vendor here: AskHR is a showcase for a product it sells, which matters when reading its figures.
Results
All the headline figures are self-reported by IBM in its case study (IBM). The company says AskHR helped contribute to a 40% reduction in the HR team’s operational costs over the past four years; it does not attribute the whole reduction to the agent. It reports a 94% containment rate of common questions, a 75% reduction in support tickets raised since 2016 and more than 11.5 million employee interactions in 2024 alone. Adoption has reached 99% among managers, and IBM says some HR task areas improved productivity by as much as 75% between 2022 and 2024.
IBM’s own volume figures are not consistent. The same case study also describes AskHR as handling over 2.1 million employee conversations annually, and in 2024 LaMoreaux cited around 10.1 million interactions per year (Fortune). IBM does not say what each figure counts.
Employee sentiment recovered. LaMoreaux told Fortune that HR’s satisfaction score now sits consistently in the +80s, against a score of -35 in the year after use became mandatory.
What didn’t go to plan
The first rollout failed. Forcing 20,000 workers onto the chatbot in 2018 drove HR’s net promoter score to -35 within a year, against an employee satisfaction score of +19 before, according to Fortune. “We were trying to force the behavior change, but we thought about it from the perspective of the HR team,” LaMoreaux said (Fortune). It took a redesign around user feedback to win employees over.
The agent has also cost HR jobs. In May 2025, chief executive Arvind Krishna told The Wall Street Journal that AI had taken over the work of a few hundred HR staff, as reported by People Matters, while saying the savings funded hiring in programming and sales (People Matters). IBM has not said when those cuts happened, and People Matters notes open questions about displacement, transparency and reskilling.
The evidence has limits. No independent audit backs the containment or cost figures; IBM’s page itself says actual results will vary; and the interaction counts conflict. The investment is not disclosed.
Lessons
- Design for the employee, not the HR team. IBM’s mandate-first launch drove its HR score to -35; asking managers what they wanted is what turned adoption around.
- Redesign before you automate. IBM’s “transform, not transfer” rule kept the agent from hard-coding inefficient processes.
- Keep humans on sensitive cases. AskHR routes performance and misconduct matters to people, which protects trust in the routine tier.
- Plan the workforce consequences openly. Krishna confirmed a few hundred HR roles went to AI; say early where the savings will be redeployed.
- Define your volume metrics once. IBM’s own interaction counts range from 2.1 million to 11.5 million, which makes the trend hard to judge.
Adapt it in your organization
Is this for you?
- You have thousands of employees asking the same payroll, benefits and policy questions.
- Your HR policies and data sit in systems such as Workday or SAP that an agent can call.
- Policies are documented and consistent enough for an agent to answer from them.
- You can route sensitive cases to named human advisors.
- You can measure ticket volume, resolution without handoff and employee satisfaction today.
A 90-day version
Pilot: put an AI agent in front of the ten most common HR questions and two transactions, such as employment letters and vacation requests, in one country, with handoff to a human advisor. Owner: chief human resources officer. Success measure: at least 70% of those questions resolved without a ticket, with employee satisfaction no lower than the baseline.
Metric to track
Containment rate: the share of common questions resolved without a human, tracked alongside employee satisfaction. IBM reports 94% after eight years of refinement; a first-year pilot should expect far less. For a customer-facing contrast, see the Klarna case.
Failure modes to pre-empt
- Mandating the tool before it is useful, as IBM did in 2018.
- Letting the agent answer from inconsistent regional policies.
- Reporting cost savings that other programs also drove as the agent’s alone.
- Cutting HR roles without a stated plan for the savings and the people affected.
Questions to take to your leadership team
- Which HR questions could an agent answer today from our documented policies?
- What would we do with the HR capacity it frees, and how will we say so to staff?
- How will we know employees prefer it, not just that they are forced to use it?
Executive FAQ
How many HR questions does IBM’s AskHR answer without a human?
IBM says AskHR achieved a 94% containment rate of common questions, a self-reported figure.
Did IBM replace HR workers with AI?
Yes, in part: Krishna said in May 2025 that AI took over the work of a few hundred HR staff, while IBM’s overall hiring grew in other areas.
About the curator & AI disclosure
Curated by Luciano Ayres
Luciano Ayres is Director of Artificial Intelligence at ArcTouch, with more than 20 years of experience in software and technology leadership and specialized studies in deep learning. Connect with him on LinkedIn.
AI-generated briefing. This case was produced through an automated AI research and writing workflow curated by Luciano Ayres. Sources are linked for independent verification; summaries and analysis may contain errors and do not replace professional due diligence.