Goldman Sachs signals AI-first operations with targeted layoffs and a new operating model
Goldman Sachs is tightening headcount and running a "limited reduction in roles" as it moves into OneGS 3.0 - a multi-year push to re-engineer how the bank operates with AI at the core. The announcement arrived the same day the firm posted record third-quarter results, underscoring a clear message for operations leaders: efficiency is now a non-negotiable, regardless of performance.
Headcount stood at 48,300 as of Sept. 30, nearly 2,000 higher than a year ago, and the firm still expects a net increase by year-end. Even so, leadership is pressing for productivity gains and front-to-back process changes that reduce cost-to-serve and cycle times across the bank.
What OneGS 3.0 means for Operations
Management outlined six priorities that translate directly into execution for ops teams.
- Enhance the client experience
- Improve profitability
- Drive productivity and efficiency
- Strengthen resilience and capacity to scale
- Enrich the employee experience
- Bolster risk management
The immediate focus is front-to-back workstreams where AI can remove delays, errors, and manual swivel-chair work. Expect end-to-end redesign, not tool swaps.
- Sales enablement and pipeline hygiene
- Client onboarding and KYC
- Lending and documentation
- Regulatory and financial reporting
- Vendor management and third-party risk
Goldman is clear: technology alone won't deliver the outcome. The firm wants faster decision cycles, leaner handoffs, and new operating rhythms that treat AI as a team member embedded in daily work.
AI already in production at Goldman
The bank's GS AI Assistant is in use by thousands, helping summarize documents, draft reports, and analyze data. The next wave targets measurable gains in throughput, exception reduction, and cycle-time compression across core processes.
The industry signal: cut costs early, even in a good quarter
Competitors are moving as well. Morgan Stanley is cutting about 2,000 roles. JPMorgan has announced multiple rounds of reductions this year. Citigroup is removing roughly 20,000 roles over two years alongside a major restructuring and tech investment program.
Research cited in industry coverage projects large-scale automation of routine finance functions over the next five years. For operations leaders, the takeaway is simple: budget for change now, or fund it later under pressure.
Ops playbook: where to act in the next two quarters
- Map front-to-back flows: Document the actual path of work (not the SOP), including systems, handoffs, and rework. Time each step.
- Triage AI use cases: Start with high-volume, rules-heavy tasks and long-tail exceptions that block straight-through processing.
- Stand up "human-in-the-loop" patterns: Define when AI drafts, when humans approve, and what auto-publishes.
- Data readiness: Clean reference data, standardize taxonomies, and route PII securely. Poor inputs kill ROI.
- Controls by design: Logged prompts, versioned models, documented decision rationale, and audit trails from day one.
- Target metrics, not tools: Commit to unit cost per case, cycle time, STP rate, and error rate. Review weekly.
- Workforce planning: Freeze low-value backfills, build internal mobility tracks, and upskill for prompt and workflow engineering.
- Vendor strategy: Consolidate overlapping tools, enforce APIs, and align contracts to measurable outcomes.
- Change management: Script the new "day in the life," train managers to coach with dashboards, and restructure incentives to reflect flow efficiency.
Metrics that matter
- Client onboarding TAT, NIGO rate, and first-pass approval rate
- STP rate by product and exception rate per 1,000 transactions
- Unit cost per case/ticket and rework percentage
- Cycle time and its variability (P50/P90)
- Error rate by step and regulatory finding frequency
- Time-to-detect and time-to-recover for incidents
Headcount and skills: precision over blanket cuts
Expect selective reductions in low-leverage roles and redeployment into controls, data quality, and AI-enabled operations. Hire for process ownership, data stewardship, and prompt/workflow engineering inside ops teams, not just in tech.
Define a redeployment bar (e.g., 6-12 weeks to productivity) and fund an internal academy to get there. Freeze backfills where AI can offset within one quarter.
Risk and governance without slowing delivery
- Model risk practices for LLMs: purpose, limitations, bias checks, and monitoring
- PII minimization, redaction at the edge, and role-based access by default
- Prompt, response, and decision logging for audit
- Vendor due diligence: data residency, IP rights, security attestations
- Guardrails for outbound communications and code generation
90-day rollout plan
- Days 0-15: Stand up an AI Ops PMO, confirm two priority journeys, baseline KPIs, and finalize guardrails.
- Days 15-45: Pilot with production data on narrow slices; measure STP, cycle time, and error deltas weekly.
- Days 45-90: Scale to adjacent steps, retire manual steps, update SOPs, and shift headcount from rework to value-add tasks.
Why this matters for Operations
Goldman's message is direct: profitability and resilience will come from process redesign powered by AI, not headcount growth. Operations leaders who treat AI as a workflow redesign tool - and who commit to hard metrics - will set the standard for the next cycle.
Upskill your team for AI-enabled operations
If you are building an internal academy or need role-based learning paths for ops, explore practical course maps by job function and curated tools for finance.
Your membership also unlocks:
