Which workflows will move from copilot to genuine labor and budget substitution first?

Thesis

As of 15 September 2026, bounded transaction queues look likeliest to substitute for labor first: AP and invoice matching, then simple customer-support and IT-service requests, followed by bounded software work. This is an analytical ranking, not demonstrated deployment timing.

The decisive test is less purchased labor at comparable volume and quality, not minutes saved. Avoided hiring substitutes for future capacity but is not an observed payroll decline. Net budget savings must also deduct software, integration, oversight and exception-handling costs.

Why workflow control matters

Agents need write permissions, reliable records, audit trails and affordable escalation. ServiceNow's vendor-sponsored 2026 survey reports only 16% with integrated foundations, 18% scaling agentic processes across key functions and 26% with governance/compliance systems. These findings suggest implementation constraints, not causal proof of substitution. [1]

Finance: strongest mechanism, incomplete payroll proof

Coupa reports GameStop reduced AP headcount 20% while processing 20% more invoices, with 70% shorter average processing time and an 82% increase in first-time matching—not 82 percentage points. Comparable volume and staffing definitions imply 50% more invoices per AP employee: 1.2 ÷ 0.8 − 1. However, the case explicitly describes resources redirected to strategic initiatives. It supports functional labor displacement, not necessarily enterprise payroll savings, and does not isolate generative agents from broader AP automation. [2]

Expense review, supplier onboarding and close tasks are plausible extensions, but their readiness is not established by this case.

Service queues: partial substitution, potentially reversible

Gartner's October 2025 survey of 321 service leaders found 20% reported AI-driven staffing reductions, while 55% maintained staffing despite higher volumes; 42% were hiring specialized AI roles. These are respondent shares, not percentages of jobs eliminated. [3]

Gartner separately forecasts that by 2027 half of companies attributing service cuts to AI will rehire similar functions under different titles. Standardized requests are better candidates than emotional, complex or regulated interactions; escalation and AI-operations hiring can offset frontline savings. [4]

ServiceNow's TP case reports 10% improved employee case deflection, without clarifying relative versus percentage-point change; 5–6 minutes saved per expert daily; and 15–25 minutes/day in major-incident management, with an unclear per-person denominator. No headcount reduction is reported. Autonomous resolution removes work, but only lower purchased labor establishes budget substitution. [5]

Coding: capacity before broad developer cuts

NBER's matched event study covers more than 500,000 GitHub developers. Estimated cumulative gains shrink sharply between commits and releases, consistent with review, testing and release bottlenecks—not proof of equivalent staffing savings. [6]

Maintenance, test generation, documentation and migrations with acceptance tests are plausible early targets. Expect pressure on incremental hiring and contractor hours before broad developer replacement; cheaper coding could instead expand demand.

Generic copilot: coordination limits conversion

Microsoft's six-month randomized experiment covered 6,000 workers. Email time fell by 1.4 hours weekly on an intent-to-treat basis; documents appeared moderately faster to complete, but meeting time did not change significantly. No labor-budget reduction was reported. Individually reclaimed time does not automatically remove coordinated organizational roles. [7]

Investment exposures

These are mechanism-based exposures, not valuation-backed recommendations. Financial amounts below are U.S. dollars; ARR is not quarterly recognized revenue.

• ServiceNow (NYSE: NOW): workflow-control beneficiary. Its Accenture program offers over 300 pre-built agent skills and workflows. Integration and governance create demand, but implementation revenue can precede customer labor savings. [8][9]

• Salesforce (NYSE: CRM): monetization, not payroll proof. Q2 FY27 ended July 31, 2026: Agentforce/Data 360 ARR approached $3.9 billion, up over 210% year over year; included Agentforce ARR exceeded $1.5 billion, up over 240%. Beginning that quarter, Agentforce ARR includes AI offerings, Slackbot and Headless 360, complicating product comparisons. Reported 3.2 billion quarterly Agentic Work Units indicate usage, not customer job removal. [8][10]

• UiPath (NYSE: PATH): automation/orchestration exposure. Q1 FY27 ended April 30, 2026: revenue was $418 million and ARR $1.901 billion. Management describes agentic products moving into production; durable usage and net customer savings remain the tests. [8][11]

• Accenture (NYSE: ACN): mixed exposure. Q3 FY26 ended May 31, 2026: bookings were $19.32 billion and revenue $18.72 billion; 104 quarterly client-booking instances of at least $100 million were reported year-to-date. These are not AI-only bookings. Implementation benefits can coexist with managed-services automation pressure. [8][12]

• Concentrix (NASDAQ: CNXC): labor-intensive risk proxy. S&P identifies sector-level outsourcing and pricing risks, not proof of Concentrix-specific AI losses. Deflection can reduce billable contacts; agent assistance may instead improve margins under fixed-price or outcome contracts. Contract structure determines who captures savings. [8][13]

Milestones and invalidation

Track volume-adjusted processing costs, exception/reopen rates, sustained autonomous resolution, paid staffing and contractor hours, and releases per developer. Separate recurring AI usage from bundling and implementation.

The strongest counterargument is demand expansion: higher service volumes with stable staffing already illustrate augmentation. The ranking weakens if AP remains redeployment-only, service cuts reverse, quality deteriorates, or coding gains stay upstream of releases. Operational automation can arrive well before net budget substitution.