Financial Services Practice

Extracting value from

AI in banking: Rewiring
the enterprise
To gain material value from AI, banks need to move beyond experimentation to
transform critical business areas, including by reimagining complex workflows with
multiagent systems.
This article is a collaborative effort by Carlo Giovine, Larry Lerner, Renny Thomas, Shwaitang Singh, Sudhakar Kakulavarapu,
and Violet Chung, with Yuvika Motwani, representing views from McKinsey’s Financial Services Practice.

December 2024
Much has been written about the power of AI, helping to pinpoint which loans might go bad,
including generative AI (gen AI), to transform enabling the bank to take steps to intervene and
banking. Beyond ushering in the next wave of support the client.
automation, AI promises to make banks more
intelligent, efficient, and better able to achieve A regional bank, meanwhile, used gen AI to boost
stronger financial performance. the productivity and efficiency of its software
developers. Seeking to optimize resources and
While the buzz is undeniable, many banking C-suite accelerate time to market of new developments, the
leaders are increasingly asking questions about bank launched a proof-of-concept study to assess
the realization of value in light of the headwinds the impact of gen AI tools on coding productivity.
facing the sector. Will AI live up to expectations? Productivity rose about 40 percent for the use
After initial experimentation, how can banks go cases that were part of the study; more than
from proof of concept to proof of value and truly 80 percent of developers said gen AI improved their
reimagine and transform the enterprise using AI? coding experience.
How soon, if ever, can banks see a tangible return
on their investments in AI? In this article, we detail a blueprint to help
financial-services leaders chart the complex path
These questions are gaining relevance as the global of extracting at-scale value from AI across the
banking sector contends with challenges such as enterprise. We begin with what banks that excel
uneven labor productivity results, including falling in AI do differently. We then outline a road map
productivity at US banks, despite high technology that roots the AI transformation in business value,
spending relative to other sectors. Banks also face ascertaining which key business problems need
slowing revenue and loan growth and competition to be solved and harnessing technology, including
from businesses beyond banking—such as private AI, to help with the process. Next, we describe a
credit firms, fintechs, neobanks, payment solutions comprehensive AI capability stack for banking
businesses, and nonbank providers—for the powered by AI agents. Finally, we explore the
largest profit pools. To maintain the current return elements needed to sustain and scale value from AI
on tangible equity margins, banks will need to cut beyond the initial rollouts.
costs much faster as revenue growth slows. AI
has the potential to chip away at these problems
and put banks on more solid footing in the years to Delivering on the promise of AI
come, particularly in boosting labor productivity as in banking
employees continue to delegate a growing number The latest McKinsey Global Survey on AI shows
of routine tasks to increasingly sophisticated and that adoption has increased significantly across
capable AI systems. organizations and industries. However, the
breadth of adoption (measured by the deployment
Some institutions are raising the bar and creating of AI across multiple enterprise functions)
strategic distance from their peers by effectively remains low, and many organizations are still in
scaling AI, including gen AI. For example, a the experimental phase.
large bank is using AI across the enterprise
to improve experiences for its customers and Still, a few leading banks stand out in their ability to
employees, enhance efficiency, and boost revenue deploy AI, including gen AI, across the enterprise,
and profitability. In retail banking, the bank is and have begun to capture material gains from the
harnessing AI to generate personalized nudges use of AI (see sidebar “What does it mean to be an
to help customers with investing and financial AI-first bank?”).
planning. In the small-business segment, AI is

Extracting value from AI in banking: Rewiring the enterprise 2

What does it mean to be an AI-first bank?

AI is enabling broad changes in all sorts The essentials of building an AI-first bank — Modernizing core technology
of industries, including banking, but many include the following: required for the backbone of the AI
banks are still in the experimental phase. capability stack, including automated
— Reimagining the customer
Given how far AI has come and the promise cloud provisioning, an application
experience by providing personalized
it holds, experimenting is not enough. To programming interface, and
offers and streamlined, frictionless
thrive in this new world, banks will need streamlined architecture to enable
use across various devices, for
to become AI-first institutions, adopting continuous, secure data exchange
bank-owned platforms as well as
AI technologies enterprise-wide to boost among various parts of the bank.
partner ecosystems.
value—or risk being left behind.
— Setting up a platform operating model
A successful AI transformation spans — Using AI to help with decision making,
that brings together the right talent,
several layers of the organization. It’s significantly enhancing productivity by
culture, and organizational design.
important to invest in each of the building the architecture required to
interdependent layers, as underinvestment generate real-time analytical insights
in one section can sabotage the entire and translating them into messages
AI transformation. addressing precise customer needs.

Our experience suggests that banks excelling in AI tools, can (see sidebar “What are multiagent
do four things well: systems?”). Expanding these systems to
the entire enterprise requires setting up a
— Set a bold, bankwide vision for the value AI comprehensive AI bank stack.
can create. Leading banks have an expansive
outlook on the role that AI can play, viewing — Sustain and scale value by setting up critical
the technology not just as a driver of cost enablers of the AI transformation. These include
efficiencies but also as a way to enhance cross-functional business, technology, and
revenues and significantly improve customer AI teams along with a central AI control tower
and employee experiences. that coordinates enterprise decisions across
functions, drives governance and adoption of
— Root the transformation in business value by standardized risk guardrails, and promotes the
transforming entire domains, processes, and reusability of AI capabilities.
journeys rather than just deploying narrow
use cases. Banks that excel in AI resist the
temptation to launch narrow use cases such Setting a bold, bankwide vision
as a chatbot or a conversational Q&A tool in for the value AI can create
isolation. Although these might be fast to launch McKinsey’s experience with hundreds of companies
and potentially low risk, in isolation, they won’t across various industries shows that capturing
unlock material financial value. value from digital and AI transformations requires a
fundamental rewiring of how a company operates.
— Build a comprehensive stack of AI capabilities This involves six critical enterprise capabilities:
powered by multiagent systems. Running a business-led digital road map, talent with the
complex banking workflows, such as evaluating right skills, a fit-for-purpose operating model,
a commercial customer’s loan application, technology that’s easy for teams to use, data that’s
involves highly variable steps and the processing continually enriched and easily accessible across
of a mix of structured and unstructured data. the enterprise, and adoption and scaling of digital
While traditional automation cannot handle solutions. These elements are interconnected, and
such tasks, gen-AI-enabled multiagent all have to function well for the transformation to be
systems, combined with predictive AI and digital a success.

Extracting value from AI in banking: Rewiring the enterprise 3

What are multiagent systems?

Multiagent systems, also known as multiagent systems are expected to Eventually, gen AI agents could act
agentic systems, have been around for improve over time. as virtual coworkers. For instance, an
years but have been kicked into a higher engineer could use everyday language
These systems could be capable of
gear in the past two years, thanks to the to describe a new software feature to a
planning actions, using tools to complete
natural-language capabilities of generative programmer agent, which would then
those actions, collaborating with other
AI (gen AI). Although they are still in a code, test, iterate, and deploy the tool it
agents and people, and improving their
nascent phase, and much of the value helped create.
performance as they learn by doing.
they could generate remains hypothetical,

AI can do much more than just automate processes Rooting the transformation in
and boost efficiency. Banks that extract value business value
from AI view the technology as a transformational Launching a chatbot, creating a document
tool and use AI for core strategic priorities such as summarizer, using off-the-shelf gen AI tools to
boosting revenue, differentiating the bank from create ads and write emails—although these types
competitors, and driving higher satisfaction for of AI endeavors allow banks to experiment and learn
customers and employees. with minimal risks involved, the results are typically
incremental and, in isolation, rarely lead to material
Leading banks embed AI in the strategic planning changes in financial outcomes.
process, requiring every business unit to revamp
its operations and set bold financial and customer Using AI to significantly boost business value will
goals. They focus on innovation by prioritizing the require banks to do the following:
most high-impact areas that are core to strategy,
versus experimenting in peripheral areas seen as — Choose the right scope of transformation by
safe bets or taking the “peanut butter” approach rewiring entire domains and subdomains.
by spreading investments across many disparate Instead of letting a thousand flowers bloom
initiatives. Next, they invest in enabling the with many disparate, siloed AI projects,
scalability of AI initiatives by setting up the right leading banks are using AI to reimagine entire
data and technology platforms. business domains—such as risk, sales, and
operations—and within them, subdomains
Leading banks also ensure that major AI initiatives such as relationship management, collections,
are business led, not just technology led. This and contact-center servicing and operations.
means business executives take ownership of A typical bank has roughly 25 subdomains
shaping the design of interventions, ensuring what (Exhibit 1). Once bank executives choose the
is built is tightly aligned with what the business subdomains for transformation, they reimagine
needs, and holding joint accountability with each one end to end, using the full range of AI
technology leaders to deliver outcomes. and digital technologies to achieve the desired
financial outcomes.

Extracting value from AI in banking: Rewiring the enterprise 4

Exhibit 1

Banks can identify business areas for AI transformation and then rewire them
to boost value.
Examples of subdomains that AI could transform in retail banking¹

Domains Sales and Risk Servicing Digital Human Other

marketing and operations technology resources functions

Digital-led Customer Self-service via Developer Recruitment Legal

customer underwriting digital channels productivity and staffing processes
acquisition such as mobile
banking

Frontline sales Risk-based Assisted service IT operations Performance Regulatory

enablement pricing via contact management, compliance
center, branch, training, and skill and controls
and digital development

Relationship Transaction Middle- and Technology Employee Business

Subdomains management fraud back-office modernization satisfaction intelligence and
and advisory prevention operations and well-being analytics

Partner Portfolio Complaints Product Employee

collaboration for optimization and management and service development
product and monitoring development and for key role
service sales management fulfillment

Engagement, Collections
cross-selling,
and customer
retention

Enterprise knowledge management

Examples of subdomains that AI could transform in private banking¹

Domains Sales and Risk Servicing Digital Human Other

marketing and operations technology resources functions

Digital-led Client risk Self-service via Developer Recruitment Legal

customer profiling and digital channels productivity and staffing processes
acquisition due diligence such as mobile
banking

Relationship Wealth and Relationship IT operations Performance Regulatory

management, portfolio risk management management, compliance
affluent management and concierge training, and skill and controls
clients² services development

Relationship Credit risk Assisted service Technology Employee Business

Subdomains management, management via contact modernization satisfaction intelligence and
HNW³ and center, branch, and well-being analytics
UHNW⁴ clients and digital

Partner-led Risk-based Complaints Product Employee

client pricing management and service development
acquisition and development and for key role
cross-referrals management fulfillment

Engagement, Fraud and Middle- and

cross-selling, financial crime back-office
and customer prevention operations
retention

Enterprise knowledge management

1
A typical bank has ~25 subdomains that could be rewired with AI. This list is not comprehensive.
²Clients with personal financial assets of $100,000–$1 million.
³High-net-worth clients are those with personal financial assets of $1 million–$50 million.
⁴Ultra-high-net-worth clients are those with personal financial assets of >$50 million.

McKinsey & Company

Extracting value from AI in banking: Rewiring the enterprise 5

Exhibit 1 (continued)

Banks can identify business areas for AI transformation and then rewire them
to boost value.
Examples of subdomains that AI could transform in corporate and commercial banking¹

Domains Sales and Risk Servicing Digital Human Other

marketing and operations technology resources functions

Digital-led Customer Self-service via Developer Recruitment Legal

customer underwriting digital channels productivity and staffing processes
acquisition such as mobile
banking

Partner-led Risk-based Relationship IT operations Performance Regulatory

sales pricing management management, compliance
and concierge training, and skill and controls
services development

Relationship Transaction Middle- and Technology Employee Business

Subdomains management fraud back-office modernization satisfaction intelligence and
and advisory prevention operations and well-being analytics

Frontline sales, Portfolio Complaints Product Employee

generalist, and optimization and management and service development
product led monitoring development and for key role
management fulfillment

Engagement, Loan renewals Assisted service

cross-selling, management via contact
and customer center, branch,
retention and digital

Enterprise knowledge management

Examples of subdomains that AI could transform in investment banking¹

Domains Sales and Risk Servicing Digital Human Other

marketing and operations technology resources functions

Relationship Customer Relationship Developer Recruitment Legal

management underwriting management productivity and staffing processes
and advisory and concierge
services

Relationship Liquidity risk Middle- and IT operations Performance Regulatory

manager–led management back-office management, compliance
deal sourcing operations training, and skill and controls
development

Engagement, Transaction Complaints Technology Employee Business

Subdomains cross-selling, fraud management modernization satisfaction intelligence and
and customer prevention and well-being analytics
retention

Market risk Product Employee

management and service development
development and for key role
management fulfillment

Enterprise knowledge management

1
A typical bank has ~25 subdomains that could be rewired with AI. This list is not comprehensive

McKinsey & Company

Extracting value from AI in banking: Rewiring the enterprise 6

 — Decide which subdomains to transform transformation. Together, these subdomains
with AI and in which order. To select these can drive 70 to 80 percent of total incremental
subdomains, banks can consider the overall value from an AI transformation.
business impact and technical feasibility
of driving an AI transformation of a specific In terms of business impact, banks will need to
subdomain (and the likelihood that the assess whether the value of an AI transformation
chosen subdomain includes components of a particular subdomain can be accurately
that can be reused in subsequent subdomain quantified, how well the proposed solution aligns
transformations) (Exhibit 2). In our experience, with the bank’s strategic objectives, how well end
a typical bank has fewer than ten subdomains users (whether clients or employees) are equipped
that could most benefit from an AI overhaul to adopt the solution, and whether the solution will
and should be the first candidates for be a priority for the business.

Web <2024>
<Ai in Banking>
Exhibit 2
Exhibit <2> of <8>

Bank subdomains with high business impact and high technical feasibility
should be first in line for an AI transformation.
Illustrative example of how business impact and technical feasibility can inform the transformation
HIGH
o Portfolio optimization and o Assisted service through contact
monitoring center, branch, digital channels
o Product and service o Collections
development and management o Customer underwriting Examples of top
o Regulatory compliance and o Developer productivity candidates for an
controls AI transformation
o Digital-led customer acquisition at a typical bank.
o Risk-based pricing
o Engagement, cross-selling, and These subdomains
o Technology modernization customer retention will vary from bank
o Frontline sales enablement to bank.
o Relationship management
o Self-service through digital
channels such as mobile banking
Business
impact
o Business intelligence and o Complaints management
analytics o Enterprise knowledge
o Collaboration with partners management
to sell products and services o IT operations
o Development of employees o Middle- and back-office operations
to fill key roles
o Legal processes
o Employee satisfaction and
well-being o Performance management,
training, and skill development
o Recruitment and staffing
o Transaction fraud prevention

LOW

LOW Technical feasibility HIGH

McKinsey & Company

Extracting value from AI in banking: Rewiring the enterprise 7

 Regarding technical feasibility, it is important Once selected for an AI transformation, each
to ascertain the availability and quality of data, subdomain can be deconstructed into a series of
including special considerations for handling executable modules that need to be built, delivered,
sensitive data, techniques for scaling the solution and adopted to drive business value. For example,
across other domains and business units, the transforming the customer underwriting subdomain
reusability of the solution’s components for other end to end involves gen AI, traditional analytics, and
use cases, and the presence of legacy technology digital tools and platforms all working together to
infrastructure that may not be compatible with more reimagine end-to-end workflows and processes
modern AI solutions. (Exhibit 3).

Web <2024>
<Ai in Banking>
Exhibit 3
Exhibit <3> of <8>

Banks can rewire the customer underwriting subdomain by using a

combination of gen AI, traditional analytics, and digital tools and platforms.
Elements and use cases Generative AI Traditional analytics Digital tools and platforms
in customer underwriting (illustrative)

Document collection Preassessment Credit assessment Contract generation

s Loan application and s Question generator: s Voice to memo: s Automation of the final step:
document checker: Check for Come up with Summarize insights and Generate contracts, such as
errors, incomplete data, and questions for a actions after personal confirmation of an applicant’s
potential fraud and follow up personal discussion discussion with the eligibility for a loan and loan
with applicants directly with the applicant applicant covenants

s Third-party data validation: s Automated decisions: Make instant decisions to

Verify accuracy of application approve or decline applications based on predefined
details using sources such criteria and risk thresholds
as credit bureaus and
government databases

s Document collection: Allow for multichannel uploading or s Data assessment: Give

scanning of collateral, financial, and know-your-customer estimates for probability
documents; convert them to the required format; and work of default, expected loss,
with customers to get missing or additional documents climate risk (using internal
and external data sources)

s Document analyzer: Check documents for correctness, eg, accuracy of the loan period,
weed out potentially fraudulent documents, and assess income and other data to make
a credit decision

s Unstructured risk elements assessment: Assess risk

elements from unstructured sources, eg, applicant’s
social media footprint and potential reputational damage

s Automated credit memo generation: Generate a credit

memo, a summary of why a customer needs a loan, and
other details for a bank employee to review

s Customer chatbot: Answer customers’ queries and guide them to submit documents, then provide updates on credit decision
and contract finalization

s Employee chatbot: Answer employees’ questions, allow for sending action alerts to teams such as relationship managers, and
drive employees’ communications with customers

s Workflows workbench: Run workflows for end-to-end application management, collateral valuation, legal review, reassignment
of tasks, etc

McKinsey & Company

Extracting value from AI in banking: Rewiring the enterprise 8

 Enabling value through an AI stack investment and attention to unlock the full power of
powered by multiagent systems AI for the enterprise.
To embed AI seamlessly across the enterprise,
banks can implement a comprehensive capability Given the advent of new technologies such as
stack that goes beyond just AI models. This AI gen AI, we have updated the AI capability stack
bank stack contains four key capability layers: (Exhibit 4) from a previous iteration published in
engagement, decision making, data and core tech, 2020. Each layer’s foundational elements are
and operating model. Each layer will need to receive supplemented by several new elements.

Exhibit 4

To drive sustainable value, banks need to put AI first and revamp the entire
technology stack.

New AI bank of the future

elements
Delighting customers through personalized experiences Empowering employees to serve customers better
Engagement Mobile as the gateway to the rest of the bank, including branches, contact center, relationship managers
Multimodal conversational Intelligent products Omnichannel experiences for
experiences (text, visual, voice) and services customers, employees and partners
Use of digital twins to simulate behavior of customers and employees
AI-powered AI orchestration (including copilots and autopilots that organize workflows)Signature Sales
decision Intent Document fraud skills
making AI agents (AI recognition summarizer detector Enterprise coach
that specializes Risk policy Property Fraud Test case
in narrow collateral pattern knowledge generator
expert analyzer detector search
domains) Ad-banner Legal AI Spend Tax expert
analyzer
Predictive Customer Credit decision Monitoring Retention, Servicing and
analytics acquisition making and collections selling, upselling engagement
models

AI Reusable components Information security Streamlined

enablers and services standards and controls risk protocols

Core Industrial AI
technology Observability Machine learning FinOps1 LLM2 LLM Security
and machine tool stack operations orchestration gateway
and data learning

Enterprise Search and retrieval engine

data
Data Data Vector Data Structured
ingestion preprocessing databases postprocessing data storage

Technology and Tech-forward strategy (in-house capabilities vs buying offerings; in-house talent plan)
infrastructure
Modern API Intelligent infrastructure (AI operations Cybersecurity Core
architecture command, hybrid cloud setup, etc) and control tiers modernization

Operating Platform Autonomous business, technology, and data teams enabled by AI models and agents
model operating
model Agile ways AI control Modern talent Culture and
of working tower strategy capabilities
Value capture office to monitor transformations

1
Financial operations, a framework for managing the operational costs of cloud computing.
2
Large language models.

McKinsey & Company

Extracting value from AI in banking: Rewiring the enterprise 9

The AI bank of the future All together now
To create sustainable value, banks need to put AI Elements across the four layers of the AI bank stack
first and revamp the entire technology stack. The work together to enable transformative change and
rise of innovative technologies such as gen AI has deliver value for the enterprise.
prompted an update to the technology stack from
a previous version published in 2020, with new The key to next-generation innovation and
elements highlighted in shades of blue. productivity: Orchestrated multiagent systems
The decision-making layer is the brain of the
Engagement layer AI-first bank, orchestrating and enabling
Banks will need to reimagine how they engage thousands of AI-powered decisions affecting
with customers, making their experiences as customers (such as which product to recommend
intelligent, personalized, and frictionless as possible to them next) and employees (for instance, should
through the use of AI. Leading banks’ customers they approve credit for a specific customer or flag
are experiencing human-like conversational a transaction as fraudulent) across the full life
interactions with AI via text and voice chats and are cycle of products and services.
moving seamlessly across channels such as mobile
apps, websites, branches, and contact centers, Predictive AI models, a core part of the decision-
thanks to powerful AI capabilities. making layer at most banks, are great at driving
decisions when presented with structured data
AI-powered decision-making layer under controlled conditions. These models,
The brain of the bank, this layer makes and however, struggle to adapt when data is
orchestrates decisions. Historically, banks unstructured and the nature of the tasks is nonlinear
have focused on deploying traditional analytics and requires multistep planning, reasoning, and
modules such as models, but as AI technologies orchestration. Such tasks include, for example,
mature, this layer has expanded to include preparing a credit memo—a summary of why a
agent and AI orchestration sublayers working customer needs a loan and other details—based
in unison with the traditional analytics layer to on multiple interactions with that customer and an
drive superior outcomes. evaluation of various types of documents. Another
example is coaching a low-performing seller on how
Core technology and data layer to improve sales performance.
This layer includes the technology and data needed
for an AI transformation, including reusable tools Orchestrated multiagent systems represent a
and pipelines equipped with machine learning major advancement in the decision-making layer.
operations capabilities needed to run large These systems comprise various AI “agents” that
language models (LLMs) at scale. Other portions can be thought of as virtual coworkers. Enabled
of this layer include the data needed to train by advances in gen AI technology, these agents,
multiagent systems, as well as modern application like humans, have the capacity to eventually be
programming interface (API) architecture and able to plan (for instance, organize a workflow
robust cybersecurity. encompassing a series of tasks), think (come up
with chain-of-thought reasoning), and act (use
Operating model digital tools).
By integrating business and technology in platforms
run by cross-functional teams, banks can break Multiagent systems remain nascent and will need
up organizational silos, boost agility and speed, more technical development before they will be
and better align goals and priorities across the ready to deploy at scale across enterprises, but they
enterprise. An AI control tower tracks the value are nonetheless attracting attention because of the
realized from AI initiatives, among other tasks. promise they hold.

Extracting value from AI in banking: Rewiring the enterprise 10

 These agents, when combined with predictive AI train and define operating procedures for the
models and digital tools, could fundamentally rewire orchestrators to follow; however, the hope is
several domains of the bank, not just unlocking that the technology will evolve to make them
productivity but forming the basis of more engaging more autonomous.
experiences for customers and bank employees.
— The AI agent layer comprises AI focused on
Multiagent systems can automate complex completing specialized tasks as instructed by
decisions and workflows through a twofold use of AI the orchestration layer or by other agents. Each
(Exhibit 5): of these AI agents, powered by LLMs, is fine-
tuned through a combination of domain-specific
— The AI orchestration layer handles complex data and human feedback. For instance, a policy
workflows and task planning. These AI agent, after being provided with the bank’s loan
orchestrators, programmed to achieve certain policies and related exceptions, can suggest
goals, are expected to eventually be able to do the appropriate loan terms for a customer,
things like autonomously plan actions, reach much like a seasoned bank executive would.
decisions, and make use of existing tools, Meanwhile, a collateral inspection agent can
in-house data, and other AI agents to complete be trained on documents and images related to
stated goals. These AI orchestrators could collateral that customers use to apply for a loan,
manifest themselves in the form of increasingly such as photos of small-business storefronts.
popular copilots for employees and customers. A computer vision tool would then work with
For instance, a copilot for a bank’s credit the collateral inspection agent to screen new
managers is capable of not just answering collateral documents and images to spot
questions but orchestrating the entire credit instances of fraud, such as doctored photos that
workflow when a customer applies for a loan. make the storefront look more impressive than
For now, human intervention is still needed to it really is.
1
Financial operations, a framework for managing the operational costs of cloud computing.
2
Large language models.

Web <2024>
<Ai in Banking>
Exhibit 5
Exhibit <5> of <8>

Orchestrated multiagent
systems represent a big AI-powered
advance in banks’ decision
making
decision-making capabilities.
Interacting to solve problems
Multiagent systems are still in a nascent phase, but
eventually AI agents could act as virtual coworkers
capable of planning and executing tasks.

The AI AI orchestration (including copilots and autopilots that organize workflows)Signature Sales
Intent Document fraud skills
orchestration recognition summarizer
layer handles AI agents (AI detector Enterprise coach
that specializes Risk policy Property Fraud Test case
complex collateral pattern knowledge
workflows, in narrow expert generator
domains) analyzer detector search
calling on Ad-banner Legal AI Spend Tax expert
specialized AI analyzer
agents to Predictive Customer Credit decision Monitoring Retention, Servicing and
complete analytics acquisition making and collections selling, upselling engagement
discrete tasks. models
Analytics and
AI enablers AI Reusable components Information security Streamlined
enablers and services standards and controls risk protocols
support these
efforts.

McKinsey & Company

Extracting value from AI in banking: Rewiring the enterprise 11

 The work of these two types of AI is enhanced by agents can be continuously trained to become
AI enablers, including reusable components and better over time, and they can be embedded
services encompassing tasks such as retrieval, across workflows. Humans will continue to
intent detection, and summarization, as well as oversee the agents, frequently auditing the results
safety elements such as data confidentiality and generated by multiagent systems and adjusting
enterprise-level risk guardrails. All of these put as needed.
together ensure that models are deployed with
speed and safety. As an illustration of how this works, consider how
the traditionally complex task of underwriting
Over time, banks could have hundreds of credit for a small-business customer can be
AI agents at their disposal, each trained to revamped through a mix of AI orchestrators and
complete a particular task and ready to be agents (Exhibit 6). The traditional way to do this
called on by other agents or humans. These is for humans to handle every step, moving from

Web <2024>
<Ai in Banking>
Exhibit 6
Exhibit <6> of <8>

The credit manager’s role has the potential to evolve significantly thanks to AI.
How AI can help human credit managers achieve higher efficiency and effectiveness
Manual orchestration and execution AI-enabled orchestration and execution

Credit manager AI frees up credit manager AI handles many

handles tasks to focus on key tasks best AI tasks previously
manually handled by humans done manually

Check applications Review AI agents’ Plan and monitor tasks during

output AI copilot the credit underwriting process
Collect and authenticate AI orchestrator
documents, then extract insights Have discussion with
customer Check loan application
AI copilot
Have personal discussion with AI agent
customer online or in person Summarize insights, actions,
and next steps Collect documents
AI copilot
Synthesize insights, actions, AI agent
and next steps Visit and inspect site
Authenticate documents and
AI copilot
extract insights from them
Evaluate collateral
documents Present credit offer to AI agent
customer AI copilot Evaluate collateral
Visit and inspect site, such as documents
business location Ask questions and send AI agent
reminders to customers
Assess credit using unstruc-
Evaluate credit using older AI AI copilot
tured data/generate loan offer
tech and generate loan offer
Predictive credit model
Predictive credit model
Check internal rules, policies,
Check internal rules, policies, and compliance terms
and compliance terms
AI agent

Prepare credit memo with Prepare credit memo with offer

offer and credit terms, and and credit terms
present it to customer
AI agent

Prepare contracts Prepare contracts

AI agent
McKinsey & Company

Extracting value from AI in banking: Rewiring the enterprise 12

 document collection to a discussion with the experiences for customers and bank employees.
customer to assessment of collateral and so on. For instance, a multiagent system can help
With orchestrated multiagent systems, agents customers during a loan application process even
can handle most of these tasks. A credit manager if they don’t have all the required documents,
steps in to review the agents’ output and handle enabling them to move on to the next step and
tasks that require the human touch: chatting ensuring that the documents are requested later.
with the customer, visiting the small business in For employees, a multiagent system could help a
question, and the final step, presenting the credit sales associate who is underperforming by creating
offer to the customer. a conversational experience that could offer the
employee specific actions to secure the next sale.
As a result, credit risk teams could evolve into a
mix of humans working with many AI orchestrators Investing in the foundations to enable AI
and agents. value creation
Banks that unlock value from AI are making
When implemented well, multiagent systems can balanced investments across the entire AI capability
fundamentally rewire various domains at a bank. stack. Envisioning this target-state AI stack is
For example, we analyzed the effects of using critical to ensuring that the right capabilities and
multiagent systems to prepare credit memos and innovations are built with an end goal in mind.
found credit analyst productivity gains of 20 to
60 percent, depending on various factors, and To build AI capabilities at scale, it’s necessary
roughly 30 percent faster decision making. to invest throughout the stack, including in the
following crucial sublayers (Exhibit 7):
Beyond boosting productivity, the use of multiagent
systems can form the basis of more engaging

Web <2024>
<Ai in Banking>
Exhibit 7
Exhibit <7> of <8>

To extract value from AI, it’s

important to invest across the
stack, including in enterprise
data and machine learning.
Core
Unlocking AI’s potential technology
and data
Investment in this portion of the stack is integral
to ensuring that large language models (LLMs)
can be run at scale.

These Industrial AI
sublayers Observability Machine learning FinOps1 LLM LLM Security
and machine tool stack operations orchestration gateway
include learning
machine
learning Enterprise Search and retrieval engine
operations data
and data for Data Data Vector Data Structured
training ingestion preprocessing databases postprocessing data storage
multiagent
systems.
1
Financial operations, a framework for managing the operational costs of cloud computing.

McKinsey & Company

Extracting value from AI in banking: Rewiring the enterprise 13

— Industrial AI/machine learning sublayer. This stages. As the transformation proceeds, reusable
sublayer within the data and core tech layer components from use cases in the first subdomain
provides the reusable tools and pipelines, can be used in other subdomains (Exhibit 8). This
equipped with a full suite of machine learning process necessitates building and improving the AI
operations capabilities, needed to cost- stack in phases, as opposed to trying to create it all
effectively deploy and run LLMs at scale. at once.

— Enterprise data sublayer. Another portion of the Setting up the technical elements outlined in the AI
data and core tech layer, this sublayer is used to stack isn’t enough to deliver value. Bank executives
store, transform, and access large unstructured also have the following imperatives as they scale AI
data sets, which are critical for training through the enterprise:
multiagent systems.
— Setting up teams to create value, not just
models. To build capabilities for the future,
Sustaining and scaling value from AI banks should put in place cross-functional
A successful AI transformation of a bank balances teams that seamlessly blend deep expertise in
delivering a positive financial impact in the near banking with technical expertise in technology
term with building lasting AI capabilities for platforms, data, and AI. Leaders should be
the enterprise. thoughtful about creating transformation
teams with adequate representation from all
After setting a business strategy with AI at its stakeholders and ensuring that everyone’s
core and choosing the domains and subdomains incentives are aligned. One key consideration
to be transformed with AI, banks should focus on is embedding risk management expertise
executing that transformation at scale, delivering inside teams from the beginning and not
value from the reusable components that can be treating it as an afterthought. Leaders should
created for one domain and then plugged into other put emphasis not just on what is being built
domains as needed. but also on how the new capabilities will be
deployed and adopted by the bank’s employees
For example, a large bank is going through a and customers.
multiyear transformation focused on improving
performance and delivering analytics at scale — Sustaining value. Extracting value from AI is an
with use cases including hyperpersonalization to enterprise-wide effort that requires centralized
target new customers and cross-sell to existing planning and coordination to deal with the
customers. The bank built reusable assets and an inevitable bumps in the road that come with a
end-to-end analytics pipeline powering more than transformation. To do this, many banks set up a
50 machine learning models, developed a tool to central AI control tower and governance council
consolidate customer leads from different sources that typically plays three roles:
and optimize them based on various factors,
and built a fit-for-purpose, data-driven business • Custodian of the bankwide AI strategy. In this
operating model. Early results are promising, with role, the control tower tracks and monitors
projected revenue increases of 10 percent and the value realized from various AI initiatives;
usage of the resulting assets and framework in more frequently updates which areas to fund and
than 150 use cases. which areas to stop funding, for instance, if
experiments fail to yield results; and refines
A transformation begins with one subdomain the bank’s comprehensive AI road map.
and the development of various use cases in that
subdomain, moving through several phases, from • Coordinator of key enterprise-wide
minimal viable product to more sophisticated decisions. These include design choices
about AI tools and the architecture stack,

Extracting value from AI in banking: Rewiring the enterprise 14

Web <2024>
<Ai in Banking>
Exhibit 8
Exhibit <8> of <8>

In an AI transformation, reusable components from the first subdomain

to be rewired can be implemented in other subdomains.
An example of an AI transformation of several subdomains at a bank

Strategy Prioritization At-scale execution with a focus on delivering value with reusable components

Bank leaders Leaders choose Minimal viable Phase 2 Phase 3

set a business which domains product
strategy with and subdomains
AI at its core to transform with
and clear AI and in which
financial goals order, linking the Prioritized use cases Continuing business impact
decisions to
strategy 1
1.1 1.2 1.3 1.4 1.5

Subdomain
2
2.1 2.2 2.3 2.4

3
3.1 3.2 3.3

Subdomain front- Other subdomains

Evolving AI technology stack runners and use cases and use cases
Use cases and capabilities built
for one subdomain are reused in
other subdomains, improving
efficiency and delivering value.
Reusability increases as various
use cases, and the AI stack, are
built out over time.

McKinsey & Company

such as what should be uniform and what can other parts of the bank, saving incremental
be federated. In this role, the control tower effort and accelerating time to market for
establishes a set of central standards and product and service launches.
risk protocols to ensure cohesion across the
bank’s units.

• Driver of enterprise AI assets’ reusability. AI holds the potential to revolutionize the way
Here, the control tower works with various business is done, but getting there will require
business segments and functions, connecting more than mere experimenting. Organizations that
dots across teams and accelerating the employ the strategies outlined in this article can
dissemination of best practices, and ensuring harness the power of AI to achieve scale and drive
that AI capabilities built in one area of the lasting, material value. Although only a few leading
bank, to the extent possible, are reused in banks are currently generating material value from

Extracting value from AI in banking: Rewiring the enterprise 15

 AI transformations, it is possible that more could join — reimagining entire business domains, including
Find more content like this on the
them within the next few years. optimizing internal processes and operations
McKinsey Insights App
(versus deploying narrow use cases in isolation)
Banks can use the following checklist to assess
whether they are truly rewiring the enterprise to put — using multiagent systems to automate complex
AI first: workflows (versus training one model and
expecting it to serve as a jack of all trades)
— setting up a comprehensive, bankwide vision
for AI and measuring the ROI of AI investments — ensuring the reusability of components (versus
(versus dabbling in AI within silos) building each AI project from scratch with
Scan • Download • Personalize
limited coordination)
— taking a full-stack approach by blending gen
AI with analytical AI and digital tools (versus
seeking a transformation from gen AI alone)

Carlo Giovine is a partner in McKinsey’s London office; Larry Lerner is a partner in the Washington, DC, office; Renny Thomas
is a senior partner in the Mumbai office, where Shwaitang Singh is a partner; Sudhakar Kakulavarapu is a partner in the
Bengaluru office; Violet Chung is a senior partner in the Hong Kong office; and Yuvika Motwani is an associate partner in the
Gurugram office.

The authors wish to thank Antonio Castro, Archit Taluka, Rickard Ström, Romil Shah, and Vik Sohoni for their contributions to
this article.

This article was edited by Jana Zabkova, a senior editor in the New York office.

Designed by McKinsey Global Publishing

Copyright © 2024 McKinsey & Company. All rights reserved.

Extracting value from AI in banking: Rewiring the enterprise 16