In 2022, AT&T announced that they had reduced regrettable attrition (employees leaving voluntarily) by 35% using an AI-powered workforce planning system. The system, built in partnership with Eightfold AI, analyzed historical HR data to identify patterns that preceded employee departures—things like changes in performance ratings, reductions in training activity, and shifts in internal job application behavior. By flagging employees at high risk of leaving, AT&T's HR team could intervene with targeted retention offers before the employees had even updated their LinkedIn profiles.
The AT&T case is notable not just for the results, but for the scale. AT&T has 200,000+ employees worldwide, and the AI system analyzed data from all of them. This wasn't a pilot on a single department; it was enterprise-wide deployment of predictive HR analytics. And it worked well enough that AT&T expanded the system to include not just retention prediction, but also skills gap analysis, internal mobility recommendations, and workforce scenario planning.
AT&T's success highlights a broader trend: AI is finally delivering on the promise of "people analytics." For over a decade, HR departments have been collecting massive amounts of employee data—performance reviews, engagement surveys, learning management system logs, email metadata, badge swipe data—but they haven't known what to do with it. AI gives them a way to extract signal from the noise, to identify which employees are at risk of leaving, which skills will be in demand in two years, and how to optimize workforce composition for business outcomes.
Employee turnover is expensive. The Society for Human Resource Management (SHRM) estimates that replacing an employee costs 6-9 months of their salary when you account for recruiting, onboarding, lost productivity, and institutional knowledge loss. For a company with 10,000 employees and a 20% annual turnover rate, that's $100-150 million in turnover costs per year.
Multiply that across the U.S. economy (160 million workers, 47% annual turnover rate in some industries), and you're looking at over $1 trillion in turnover-related costs annually. Even a 10% reduction in turnover would save the economy $100 billion per year. That's the prize that AI workforce planning is chasing.
But turnover prediction is just the beginning. AI workforce planning systems are increasingly being used for:
1. Skills gap analysis: Identifying which skills the organization currently has, which skills it will need in the future, and where the gaps are. This is particularly important in industries undergoing digital transformation, where the half-life of technical skills is estimated at 2.5 years.
2. Internal mobility optimization: Matching current employees to open positions within the company based on skills, interests, and career trajectory. This reduces both time-to-fill and turnover, because internal hires are cheaper and stick around longer than external hires.
3. Workforce scenario planning: Modeling the impact of different business scenarios (market expansion, automation, M&A) on workforce needs. This helps HR and finance teams plan headcount, budget, and hiring strategies proactively rather than reactively.
4. Diversity and inclusion analytics: Identifying where the organization is losing diverse talent and predicting which interventions (mentorship programs, sponsorship, flexible work policies) will be most effective at improving retention and advancement for underrepresented groups.
The AI workforce planning market is dominated by a mix of HR tech incumbents (Workday, Oracle, SAP SuccessFactors) and AI-native startups (Eightfold AI, Visier, Peakon, Culture Amp). The incumbents have the advantage of data—they already have HR data from thousands of customers—but the startups have the advantage of focus and speed.
Workday launched their "Workday Prism Analytics" platform in 2018 and have been steadily adding AI capabilities since. In 2023, they announced "Workday Skills Cloud," which uses AI to parse job descriptions, resumes, and learning records to create a dynamic skills taxonomy for the organization. The system can then recommend employees for open roles based on skills rather than just job titles. Workday claims that customers using Skills Cloud see a 20% increase in internal fill rates.
Eightfold AI is the breakout startup in this space. Founded in 2016 by Ashutosh Garg (formerly of Google) and Varun Kacholia (formerly of Facebook), Eightfold uses deep learning to build "talent intelligence" platforms. Their system creates a "talent genome" for each employee—a vector representation of their skills, experiences, and potential—and uses it to match employees to opportunities (jobs, learning, mentors) within the organization. Eightfold raised $220 million in Series E funding in 2024 at a $2.1 billion valuation.
Visier, a Vancouver-based people analytics company, focuses on workforce planning and scenario modeling. Their system ingests HR data from multiple sources (HRIS, ATS, LMS, etc.) and provides dashboards and predictive models for workforce planning. Visier's differentiator is their "benchmarking" data—because they have data from over 17,000 organizations, they can tell a customer how their turnover rate, time-to-fill, or skills gap compares to industry peers.
| Vendor | Founded | Funding/Valuation | Key Differentiator |
|---|---|---|---|
| Workday | 2005 (public) | $60B market cap | Incumbent with full HR suite |
| Eightfold AI | 2016 | $540M raised, $2.1B valuation | Deep learning for talent matching |
| Visier | 2010 | $300M+ raised | Workforce benchmarking data |
| Culture Amp | 2009 | $250M+ raised | Employee engagement + performance |
| Peakon (acquired by Workday) | 2015 | Acquired for $700M (2021) | Real-time engagement analytics |
For all the hype, AI workforce planning systems have a dirty secret: the accuracy of turnover prediction models is mediocre. Most published studies report AUC (Area Under the Curve) scores of 0.70-0.75 for turnover prediction. That sounds impressive until you realize that AUC measures the model's ability to rank employees by risk, not its ability to accurately predict who will leave. In practice, turnover prediction models have false positive rates of 40-60%, meaning that for every employee correctly predicted to leave, there's another employee incorrectly predicted to leave.
The high false positive rate creates a problem: if you act on every "at-risk" flag, you'll waste retention budget on employees who weren't actually going to leave. But if you only act on the highest-confidence flags, you'll miss most of the actual departures. It's a classic precision-recall tradeoff, and most HR teams haven't figured out how to navigate it.
There's also the issue of "algorithmic aversion"—managers who don't trust the AI's recommendations and override them. A 2023 study from Wharton found that when HR managers were shown AI-generated retention risk scores, they overridden the AI's recommendations 60% of the time. The managers said they trusted their "gut instinct" more than the algorithm. But when the researchers compared outcomes, the AI's predictions were 15% more accurate than the managers' instincts.
The solution to both problems—false positives and algorithmic aversion—is better explainability. The AI systems that are succeeding in production are the ones that don't just give a risk score, but explain why the score is what it is. "This employee is at high risk of leaving because their performance rating declined 15% in the past year, they haven't applied for an internal transfer in 18 months, and similar employees in their role and tenure have a 40% attrition rate." That's an explanation a manager can act on (or override with good reason).
AI workforce planning raises significant privacy concerns. These systems analyze not just HR data, but also email metadata, calendar data, Slack messages, and badge swipe data to predict behavior. Employees are often unaware that their digital exhaust is being mined for signals about their engagement and retention risk.
In the EU, GDPR and the AI Act impose strict limits on automated decision-making in employment. Article 22 of GDPR gives individuals the right not to be subject to a decision based solely on automated processing. That means if an AI system flags an employee as "high flight risk" and the manager uses that flag to deny the employee a promotion or raise, the employee can request human review of that decision.
Several high-profile cases have highlighted the risks. In 2023, an Amazon warehouse worker in the UK filed a complaint with the Information Commissioner's Office (ICO) alleging that Amazon's workforce analytics system—which used AI to predict which workers were likely to unionize—violated GDPR. The case is still pending, but it's a warning shot for companies using AI to monitor and predict employee behavior.
Despite the challenges, there are examples of AI workforce planning being implemented effectively. Microsoft, which has been a pioneer in people analytics, uses AI to predict which employees are at risk of burnout and which teams are at risk of losing critical knowledge due to attrition. Their system, built on Microsoft Viva Insights, analyzes email patterns, meeting metadata, and survey responses to generate "wellbeing signals." Managers can see these signals (aggregated and anonymized) and take action—like redistributing work or encouraging time off—before burnout leads to turnover.
Unilever, which has been using AI for recruitment and workforce planning since 2017, reported that their AI system reduced time-to-hire by 75% and increased diversity of new hires by 16%. Their system uses natural language processing to parse resumes and match candidates to roles based on skills rather than just pedigree. It also uses video interview analysis (with consent) to assess soft skills and cultural fit.
The companies that are succeeding with AI workforce planning share a few common practices:
1. They start with a business problem, not a technology solution. "We want to reduce turnover in our engineering organization by 20%" is a good starting point. "We want to implement AI workforce planning" is not.
2. They invest in data quality before AI. AI models are only as good as the data they're trained on. If your HR data is incomplete, inconsistent, or biased, your AI will be too.
3. They combine AI predictions with human judgment. The best systems don't automate HR decisions; they augment them. They give HR teams better information to make decisions, but they don't make the decisions for them.
4. They communicate transparently with employees. Employees are more likely to accept AI workforce planning if they understand what data is being used, how predictions are made, and how the system benefits them (e.g., better career development, reduced burnout).
The $1 trillion retention problem won't be solved by AI alone. But AI is a powerful tool for understanding workforce dynamics, predicting outcomes, and optimizing decisions. The organizations that figure out how to use it responsibly and effectively will have a massive advantage in the war for talent. Those that don't will continue to bleed talent and waste money on preventable turnover.