What Is an Agentic Target Operating Model?

Every bank has an AI strategy. Most of them answer the wrong question.

They ask where they can use AI. The better question is what the organisation looks like when AI is a participant in it.

McKinsey calls this the "agentic organization." IBM describes an "always-on operating model." BCG talks about "machines that manage themselves." The consulting industry has recognised the shift. But their frameworks are designed for every industry. They do not address the specific governance, regulatory, and structural realities of banking.

What is missing is a target operating model. A TOM, the kind of document a Swiss bank's COO would use to redesign how work actually gets done, who is accountable for which decisions, and how authority flows through a system that now includes non-human actors.

That is what this site describes.

Why "target operating model" matters

An AI strategy says "we will deploy AI in these areas." A target operating model says "this is how the organisation works when AI is a first-class participant." The first can coexist with the current org chart. The second replaces it.

IBM's 2025 CEO Study, based on 800 executive interviews, found that 78 percent believe capturing full value from agentic AI requires a new operating model where humans and systems learn together. McKinsey Global Institute's "Agents, Robots, and Us" report (2025) estimates that current AI technologies could automate roughly 57 percent of work hours in banking. BCG's 2025 banking outlook projects that AI could increase bank profitability by 30 percent and reduce costs by 30 to 40 percent by 2030.

These numbers are meaningless without a design. The design is the TOM.

What makes agents different

We are talking about agents, not automation. Automation executes predefined rules on predefined inputs. A bot that reconciles 50'000 transactions per day is executing instructions.

An agent pursues goals. It operates in dynamic environments. It makes decisions under uncertainty and learns from outcomes. Give it an objective, process this mortgage application and return a risk-adjusted offer, and it determines how to achieve it, what to escalate, and when to stop.

This changes governance entirely. A rule-based RPA bot can be governed like software: deploy, monitor uptime, fix bugs. An agent must be governed like a member of staff, with onboarding, performance expectations, accountability structures, and an offboarding process.

Banks that manage their agents like software are under-governing them. Banks that try to specify their behaviour fully in advance are misunderstanding what they are.

Three workforces

The agentic TOM rests on a taxonomy. Conflating all non-human actors is one of the most common mistakes banks make. McKinsey's Global Institute describes three distinct categories: people, agents, and robots. For banking, the taxonomy is:

Bots are the industrial backbone. Rule-based, auditable, deterministic. Reconciliations, data extraction, sanctions screening, payment routing. No judgment. By design. They belong in the same governance category as infrastructure.

Agents operate as knowledge workers at scale. They pursue goals, use context, and make probabilistic decisions within a policy envelope defined by humans. A mortgage pricing agent evaluates risk, calibrates against the bank's book, and generates an offer. A compliance monitoring agent reviews transactions for patterns suggesting financial crime. When the stakes exceed their authorisation level, they escalate by design.

Humans hold board-level decision proxies within operations. They own irreversible decisions, regulatory accountability, and relationship trust. They handle the estate planning conversation, the complex credit exception, the client who needs a trusted person in the room. They never touch routine.

The shift most organisations have not absorbed: human roles concentrate at the edges. Exceptions, governance, relationships. PwC describes this as the shift from a pyramid to a diamond: fewer junior roles, a strong middle layer of agent orchestrators, and a small senior leadership tier.

From functions to outcomes

A conventional bank organises around functions: operations, compliance, risk, client service. Each function optimises its own process.

Handoffs between functions are where most operational risk lives. They are also where most of the cost sits. A mortgage application passes from sales to credit to legal to operations to client service. Each handoff adds time. Each boundary creates potential for error.

In an agentic TOM, the organising principle is the outcome. An agent system owns an end-to-end workflow, mortgage origination from application to offer, KYC remediation from trigger to resolution, client onboarding from first contact to active account, and coordinates across what used to be functional boundaries.

The department does not disappear. But it stops being the unit of execution. It becomes the unit of governance. The head of operations governs the agents that manage the process. The head of compliance sets the policy envelope within which the compliance agent operates and reviews its performance.

What changes for leadership

For a CEO, the agentic TOM is a strategic question: what is the target state, and how fast are we moving toward it? DBS Bank completed nine operating model transformation initiatives in a single year and identified 11'000 employees for reskilling. That is the pace of a bank that has decided to move. Most Swiss banks have not.

For a COO, it is an operational redesign. Which processes become outcome-owned agent systems? What is the transition path? McKinsey envisions one human managing 20 to 30 AI agents. Accenture describes a "10x bank" where a single individual leads a team of AI co-workers delivering exponentially greater output. These ratios require different management skills.

For a CRO, it is a governance challenge. How do you ensure accountability when decisions are made by systems that adapt? FINMA Guidance 08/2024 provides the regulatory frame: accountability cannot be delegated to AI, autonomous operation requires demonstrated reliability, and the institution must retain the expertise to override. The CRO's job is to design the governance that satisfies these requirements at scale.

For a CHRO, it is a workforce transformation. KPMG's "Agents of Change" report (2025) found that 87 percent of leaders say agents require redefining performance metrics. BCG found that 29 percent of heavy adopters already report fewer traditional entry-level roles. The skills profile changes, and the people who thrive in the new model may not be the ones who thrive in the current one.

The cautionary tale

Klarna offers the most instructive lesson. In 2024, the company deployed an AI assistant that it said replaced the work of 700 human agents. Resolution time dropped from 11 minutes to under 2. The company claimed CHF 40 million in annual savings.

Within a year, they reversed course. Quality slipped on complex cases. The CEO acknowledged the strategy "went too far." Klarna began rehiring humans.

The lesson: full automation without designed escalation, without human governance of the edges, without a clear framework for what stays human, breaks. The agentic TOM is that framework.

Why now, and why Switzerland

Three forces are converging.

The technology is ready. According to Evident Insights, agentic AI deployments at the top 50 global banks grew 13 times in a single year (2024-2025). 70 percent of financial services organisations are deploying or exploring agentic AI, but only 14 percent have reached full-scale implementation. The gap between experimentation and operating model redesign is where most banks are stuck.

The demand side is shifting. Customers will increasingly interact with banks through their own AI agents, querying multiple providers simultaneously, benchmarking fees in real time, filing structured complaints. The volume and speed implications are severe for banks that still operate at human pace.

The regulatory framework permits it. Switzerland's principles-based regulatory philosophy, reinforced by FINMA Guidance 08/2024, creates space for agentic systems that the EU's rules-based AI Act constrains. The AI Act classifies credit assessment as high-risk and imposes detailed prescriptive requirements. FINMA governs outcomes and accountability. Swiss banks have a structural advantage that most have not yet recognised.

What this site explores

This post defines the concept. The rest of the site explores its implications:

Escalation by Design: why human oversight is an architecture decision.

The Swiss Agentic Advantage: three structural advantages Swiss banks have not yet recognised.

The Tsunami: what happens when customers use AI agents to interact with your bank.

Mortgage Lending in an Agentic TOM: a concrete case sketch showing the three-layer model applied to one process, grounded in FINMA 08/2024.

From ABS to Autopilot: the five levels of banking autonomy, and why you cannot skip them.

Who Manages the Agents?: the organisational design question behind agent governance.

The framework page visualises the three-layer model. The assessment reveals where your institution stands today.

The first step is conceptual

The implications for leadership, governance, and technology are substantial. The first step is not technical.

Your bank already has a target operating model. It was designed for a workforce that was entirely human. That assumption no longer holds.

The question is whether you redesign deliberately, or discover, too late, that the model redesigned itself around you.