AI Layoff Math Fails in Production: Two Reversals and the Numbers Behind Them

By Daniel Okafor

Commonwealth Bank of Australia canceled the elimination of 45 customer-service positions weeks after announcing them in July 2025. The bank apologized and admitted it had not adequately considered business requirements when assessing the roles. The reversal came only after employees and their union reported that call volumes were rising — while the bank offered overtime and assigned team leaders to answer calls.

That is the pattern this story examines. AI pilots promise savings. Production delivers complications. The staffing math then breaks.

What happened at Syndio?

Pay intelligence software vendor Syndio shifted toward an AI-centered business model. In May, the company eliminated several positions, including that of labor economist Jonathan Vidales. Around the same time, it redesigned remaining jobs, adding responsibilities and sales quotas.

CEO Maria Colacurcio expected the surviving employees to accept the new terms, particularly with the job market souring for workers. "We assumed they wouldn't blink," she wrote in an August account. Several resigned instead. By August, Syndio had rehired Vidales into a newly opened position, and Colacurcio acknowledged the company had underestimated its existing employees and their ability to adapt.

Note what is not said here: no one claims the AI system failed to replace a worker. Syndio was changing its business model. Its staffing assumptions proved more complicated than anticipated. That is the caveat, and it follows every promise of clean AI substitution.

What does the workforce data show?

Research released in October by talent assessment firm SHL, based on a Propeller Insights survey of 2,036 U.S. and U.K. workers, suggests employers may not understand the people already working for them:

• Only 42% believed their managers fully understood their skills beyond their current positions.
• 84% wanted employers to identify internal opportunities matching their capabilities.
• 19% of U.S. respondents said they had been laid off, replaced by AI, and then asked to return by the same employer.

Treat that 19% with care. SHL did not verify that AI actually caused those layoffs, senior vice president Lucy Beaumont told TechTarget. Self-reported boomerang cases are a signal, not a verified count.

"Many organizations are treating AI as a way to take cost out of existing work rather than as a reason to redesign it, and that is the risk," Beaumont said.

Gartner adds a third data point. In a survey of 321 customer-service and support leaders, 85% reported expanding human agents' responsibilities as AI changed their work, while 31% had implemented or planned AI-driven frontline layoffs through the first quarter of 2027. Those figures are not contradictory. AI can reduce some responsibilities while expanding others, and the hard part is seeing how the changes fit together before making permanent cuts.

Why doesn't a successful pilot settle it?

Assessing the operational and workforce impact of new technology is established IT management practice. AI should not be exempt. But even a successful pilot cannot establish what production will show — how the volume of complex calls shifts, or whether routine-request automation leaves employees with more exceptions, more oversight and harder problems.

"Redesign work around AI, not just around tools. Decide which tasks AI should take on and where human judgment, context and relationships add the most value," Beaumont said.

The open questions for CIOs:

• AI handles certain tasks — what happens to the complete workflow?
• Productivity improves in testing — will gains persist in production?
• Time spent on particular activities drops — what work falls on remaining employees?
• New roles are designed — do employees have the skills, capacity and willingness to perform them?
• Financial projections show savings — do actual savings justify permanent staffing changes?

Who gets a voice in the assessment?

Employees who understand how the work actually gets done should participate, according to the analysis. A representative group can surface exceptions, informal responsibilities and dependencies that never appear in job descriptions or process documentation. Their participation is not only a matter of fairness; it can produce better implementations and a clearer picture of the workforce the business will need afterward.

Beaumont argues technology and HR leaders must make these decisions jointly. "In my experience, the poorest decisions come when AI programs are led by technology alone," she said, noting that questions about skills, changing roles and adoption become afterthoughts.

Candid participation is hard when employees suspect they are helping build their own replacement. A September Gallup survey found 27% of U.S. workers worried technology could make their jobs obsolete, up from 13% in 2017 — roughly a doubling in eight years. Promises that AI will supplement rather than replace people ring hollow when employers themselves do not yet know the staffing impact.

Commonwealth Bank's reversal proves the point. Workers and their union challenged management's account of what was happening in the call centers, and the operational evidence forced the bank to reverse course. Employees had information the initial assessment lacked.

What should happen before cuts become permanent?

Some AI implementations will genuinely reduce headcount. Others will create new responsibilities or free capacity the business previously lacked. Not every position must remain unchanged; the work AI can perform does not tell the whole story about which employees a company will need, either.

Before converting projected efficiencies into permanent reductions, CIOs, HR, finance and business leaders should verify what the organization has actually gained, what work remains, and what additional costs or responsibilities the technology created — including whether employees can be retrained or reassigned.

An organization can revise an implementation, change a workflow or adjust its operating model. Rebuilding lost expertise, replacing experienced employees and restoring trust could prove considerably harder. The challenge is not knowing what AI can do. It is understanding what the business and its people will need once AI is doing it.