Flipping the Banking Pyramid at Sibos 2025: AI Moves Value to the Front Office

How AI is "flipping the pyramid" of the banking business model: Sibos 2025

AI is set to touch every layer of banking. The message from Sibos was clear: move from experiments to execution and aim AI directly at growth, cost, and compliance.

For Operations leaders, this is a reallocation problem. Shift effort from brittle middle/back-office stacks to customer-facing impact-without weakening controls.

The three priorities Ops teams must backstop

Growth and performance. Many banks still have return on equity below the cost of capital. Expect more M&A in the next three years, especially in mature markets. Ops needs a repeatable integration motion, not a one-off scramble.

• Standardize due diligence data intake and mapping.
• Rationalize cores and channels with a clear cutover plan.
• Simplify product catalogs and pricing early.
• Stage client migrations with clean communications and fallback playbooks.

Cost and efficiency. Cost-income ratios are stuck: ~62% in the US, ~54% in Europe, lower in LATAM and APAC. Point fixes won't move those numbers. Structural change will.

Compliance and risk. A global bank can face 12,000+ discrete regulations. Gen AI will produce more software and, with it, a larger attack surface. Controls must scale with volume.

Flip the pyramid: simplify the back, fund the front

For decades, two-thirds to three-fourths of investment went into middle and back office. The result: complex, inflexible systems that resist change. That model has hit a wall.

Gen AI, hybrid cloud, and eventually quantum computing can streamline the middle and back office. Free the resources and attention to refocus on customers, channels, platforms, and partnerships.

• Case management: automate intake, triage, and summarization across Ops, Finance, and Risk.
• KYC/AML: investigator co-pilots with full audit trails and policy grounding.
• Policy and procedure assistants: retrieval-augmented agents over approved content.
• Exceptions and disputes: straight-through processing with AI classification and next-best actions.
• Reconciliations: real-time data contracts, anomaly detection, and explainable break resolution.

Banking is behind-by design-but the clock is ticking

Other industries adopted enterprise AI faster due to lower regulatory exposure. Still, over 90% of banks plan to incorporate AI by 2028. The first movers will set the new benchmarks for cycle time, cost, and control quality.

Translate strategy into sequences: common components, reusable patterns, and a pipeline of use cases that compound.

SDLC: the biggest quick win

Gen AI can improve software development productivity by roughly 20%. One bank's contact center now routes and answers ~283,000 monthly queries with gen AI, cutting wait times from ~10 minutes to seconds. That is throughput you can measure.

Inside IBM, "client zero" results point to scale: ~86% of IT tickets and ~94% of HR queries handled by internal gen AI assistants, with about USD 4.5B in annual efficiency gains. The lesson for Ops: codify patterns, instrument everything, and keep humans in the loop where risk is highest.

• Stand up pair-programming and code-assist in a controlled environment (VDI, policy-guarded prompts).
• Enforce provenance: prompt/version logs, SBOMs, code scanning, and policy-as-code gates.
• Automate test generation and regression packs; ship smaller, safer changes more often.
• Target a 10% reduction in L2/L3 tickets via AI triage and self-heal before scaling to customer workflows.

Risk expands with more software-and quantum is coming

More code means more threats. Extend threat modeling, harden endpoints, and monitor gen AI usage with the same discipline as customer data access.

Quantum risk is a timing issue: attackers can steal encrypted data now and decrypt later. Build crypto agility, not a one-time swap.

• Maintain a cryptographic inventory and a migration plan aligned to NIST post-quantum guidance.
• Shorten retention for long-lived secrets and rotate keys on a schedule you can prove.
• Model risk for gen AI: usage tiers, red teaming, explainability logs, and human oversight by process and risk level.
• Automate regulatory obligation mapping across jurisdictions and track implementation evidence.

The data advantage

Banks hold proprietary transaction data that competitors cannot copy. Most of it has yet to be used in a generative context. That's the edge-if governance is tight.

• Stand up a governed feature store and vector store with field-level controls and PII redaction.
• Define golden sources and data contracts; monitor freshness, quality, and lineage.
• Use synthetic data for testing; apply purpose-based access and consent enforcement.

A 90-day execution plan for Operations

• Days 0-30: Form an AI governance council; approve intake, risk tiers, and guardrails. Pick 3-5 use cases tied to cost or control outcomes. Baseline metrics and spin up an SDLC copilot pilot.
• Days 31-60: Launch KYC/AML investigator co-pilot and reconciliation bots. Complete a cryptographic inventory and draft a PQC roadmap. Start a regulatory change engine proof of concept.
• Days 61-90: Move two workflows to straight-through processing. Deploy an employee "Ask Ops" assistant targeting 70% self-service resolution. Publish results and fund scale-out.

KPIs to track weekly

• Cost-income ratio delta and unit cost per transaction.
• Cycle time: claim, dispute, onboarding, and exception handle time.
• STP rate, rework rate, and first-contact resolution.
• Control health: breaches, audit findings cleared, model drift alerts.
• Ticket volume by tier, MTTR, and percent auto-resolved.
• Customer wait time and NPS for assisted channels.

Skills and tooling

• Train Ops teams on prompt patterns, data governance, model risk, and process mining.
• Pilot tools that matter for Finance and Ops efficiency. Curated options here: AI tools for Finance and role-based learning paths: Courses by job.

Compliance resources

• Basel framework overview (capital, liquidity, leverage): BIS

Bottom line

The pyramid is flipping. Simplify the middle and back office with AI, shift investment to customer value, and harden controls as software volume grows. Delay adds integration debt and keeps costs high.

Pick a few high-yield use cases, measure relentlessly, and scale what works.