AI-powered decision making for the bank of the future
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Global Banking & Securities
AI-powered decision
making for the bank of
the future
Banks are already strengthening customer relationships and
lowering costs by using artificial intelligence to guide customer
engagement. Success requires that capability stacks include the
right decisioning elements.
by Akshat Agarwal, Charu Singhal, and Renny Thomas
© Getty Images
March 2021The ongoing transition to digital channels creates engaging with them continuously and intelligently
an opportunity for banks to serve more customers, to strengthen each relationship across diverse
expand market share, and increase revenue at lower products and services.
cost. Crucially, banks that pursue this opportunity
also can access the bigger, richer data sets required — Lower operating costs. Banks can lower costs
to fuel advanced-analytics (AA) and machine- by automating as fully as possible document
learning (ML) decision engines. Deployed at scale, processing, review, and decision making,
these decision-making capabilities powered particularly in acquisition and servicing.
by artificial intelligence (AI) can give the bank a
decisive competitive edge by generating significant — Lower credit risk. To lower credit risks, banks
incremental value for customers, partners, and can adopt more sophisticated screening of
the bank. Banks that aim to compete in global prospective customers and early detection of
and regional markets increasingly influenced behaviors that signal higher risk of default and
by digital ecosystems will need a well-rounded fraud.
AI-and-analytics capability stack comprising four
main layers: reimagined engagement, AI-powered As banks think about how to design and build a highly
decision making, core technology and data flexible and fully automated decisioning layer of
infrastructure, and leading-edge operating model. the AI-bank capability stack, they can benefit from
organizing their efforts around four interdependent
The layers of the AI-bank capability stack are elements: (1) leveraging AA/ML models for
interdependent and must work in unison to deliver automated, personalized decisions across the
value, as discussed in the first article in our series customer life cycle; (2) building and deploying AA/
on the AI bank of the future.¹ In our second article, ML models at scale; (3) augmenting AA/ML models
we examined how AI-first banks are reimagining with what we call “edge” capabilities³ to reduce
customer engagement to provide superior costs, streamline customer journeys, and enhance
experiences across diverse bank platforms and the overall experience; and (4) building an enterprise-
partner ecosystems.² In the current article, we focus wide digital-marketing engine to translate insights
on the AA/ML decisioning capabilities required to generated in the decision-making layer into a set of
understand and respond to customers’ fast-evolving coordinated messages delivered through the bank’s
needs with precision, speed, and efficiency. Banks engagement layer.
that leverage machine-learning models to determine
in (near) real time the best way to engage with each
customer have potential to increase value in four Automated, personalized decisions
ways: across the customer life cycle
If financial institutions begin by prioritizing the use
— Stronger customer acquisition. Banks gain an cases where AA/ML models can add the most value,
edge by creating superior customer experiences they can automate more than 20 decisions in diverse
with end-to-end automation and using advanced customer journeys. Within the lending life cycle, for
analytics to craft highly personalized messages example, leading banks are relying increasingly
at each step of the customer-acquisition journey. on AI and analytics capabilities to add value in five
main areas: customer acquisition, credit decisioning,
— Higher customer lifetime value. Banks can monitoring and collections, deepening relationships,
increase the lifetime value of customers by and smart servicing (Exhibit 1, next page).
1
Suparna Biswas, Brant Carson, Violet Chung, Shwaitang Singh, and Renny Thomas, “AI-bank of the future: Can banks meet the AI challenge?”
September 2020, McKinsey.com.
2
Violet Chung, Malcolm Gomes, Sailee Rane, Shwaitang Singh, and Renny Thomas, “Reimagining customer engagement for the AI bank of the
future,” October 2020, McKinsey.com.
3
Edge capabilities refer to next-generation AI-powered technologies that can provide financial institutions an edge over the competition.
Natural language processing (NLP), voice-script analysis, virtual agents, computer vision, facial recognition, blockchain, robotics, and
behavioral analytics are some of the technologies that we classify as “edge capabilities.” These capabilities can be instrumental in improving
customer experience and loyalty across multiple dimensions (engagement channels, intelligent advisory, faster processing), personalizing
offers with highly accurate underwriting, and improving operational efficiency across the value chain (from customer servicing to monitoring,
record management, and more.).
2 AI-powered decision making for the bank of the futureExhibit 1
Banksshould
Banks shouldprioritize
prioritizeusing
usingadvanced
advanced analytics
analytics (AA)
(AA) andand machine
machine learning
learning (ML)
(ML) in decisions across the customer life
in decisions across the customer life cycle. cycle.
Customer life cycle
Customer Credit decisioning Monitoring and Deepening Smart servicing
acquisition collections relationships
Hyperpersonalized Early-warning Intelligent offers (eg,
Credit qualification Servicing personas
offers signals next product to buy)
Probability of Dynamic customer
Customer retargeting Limit assessment Churn reduction
default/self-cure routing (channel, agent)
Propensity-to-buy VAR-based customer Real-time recom-
Pricing optimization Channel propensity
scoring segmentation1 mendation engine
Agent–customer AI-enabled agent
Channel mapping Fraud prevention Fatigue rule engine
mapping review and training
Monthly customer- Credit-approval Average days past Deposit/AUM Net promoter score,
acquisition run rate turnaround time, % due, nonperforming attrition rate,2 cost of servicing
of applications assets products 6 per
approved customer
Key metrics
1
VAR is value at risk.
2AUM is assets under management.
Customer acquisition profile of each customer, including financial position
The use of advanced analytics is crucial to the design and provisional credit scoring. Based on real-time
of journeys for new customers, who may follow a analysis of a customer’s digital footprint, banks can
variety of paths to open a new card account, apply for display a landing page tailored to their profile and
a mortgage, or research new investment opportunities. preferences.
Some may head directly to the bank’s website, mobile
app, branch kiosk, or ATM. Others may arrive indirectly These tools can also help banks tailor follow-up
through a partner’s website or by clicking on an ad. messages and offers for each customer. Replacing
Many banks already use analytical tools to understand much of the mass messaging that used to flow to
each new customer’s path to the bank, so they get an thousands or tens of thousands of customers in a
accurate view of the customer’s context and direction subsegment, advanced analytics can help prioritize
of movement, which enables them to deliver highly customers for continued engagement. The bank can
personalized offers directly on the landing page. select customers according to their responsiveness
Following local regulations governing the use and to prior messaging—also known as their “propensity
protection of customer data, banks can understand to buy”—and can identify the best channel for
individuals’ needs more precisely by analyzing how each type of message, according to the time of
customers enter the website (search, keywords, day. And for the “last mile” of the customer journey,
advertisements), their browsing history (cookies, AI-first institutions are using advanced analytics
site history), and social-media data to form an initial to generate intelligent, highly relevant messages
AI-powered decision making for the bank of the future 3and provide smart servicing via assisted channels to usage data, and more. This decisioning process
create a superior experience, which has been shown to is automated from end to end, so it can be
contribute to higher rates of conversion.⁴ completed nearly instantaneously, enabling
the bank to predict the likelihood of default for
Credit decisioning individuals in a vast and potentially profitable
Setting themselves apart from traditional banks, segment of unbanked and underbanked
whose customers may wait anywhere from a day to a consumers and SMEs. As banks build and refine
week for credit approval, AI-first banks have designed their qualification model, they can proceed
streamlined lending journeys, using extensive gradually, testing and improving the model—for
automation and near-real-time analysis of customer example, by using auto-approvals for customers
data to generate prompt credit decisions for retailers, up to a certain threshold with significantly lower
small and medium-size enterprises (SMEs), and default risk and using manual verification to
corporate clients. They do this by sifting through a review those estimated to have a higher default
variety of structured and unstructured data collected risk and then gradually shifting more cases to
from conventional sources (such as bank transaction automated decisioning.
history, credit reports, and tax returns) and new ones
(including location data, telecom usage data, utility — Limit assessment. Leading banks are also using
bills, and more). Access to these nontraditional data AA/ML models to automate the process for
sources depends on open banking and other data determining the maximum amount a customer
sharing guidelines as well the availability of officially may borrow. These loan-approval systems, by
approved APIs and data aggregators in the local leveraging optical character recognition (OCR)
market. Further, while accessing and leveraging to extract data from conventional data sources
personal data of customers, banks must secure data such as bank statements, tax returns, and
and protect customer privacy in accordance with utilities invoices, can quickly assess a customer’s
local regulations (e.g., the General Data Protection disposable income and capacity to make regular
Regulation in the EU and the California Consumer loan payments. The proliferation of digital
Privacy Act in the US). interactions also provides vast and diverse data
sets to fuel complex machine-learning models.
By using powerful AA/ML models to analyze these By building data sets that draw upon both
broad and diverse data sets in near real time, banks conventional and new sources of data, banks
can qualify new customers for credit services, can generate a highly accurate prediction of
determine loan limits and pricing, and reduce the risk a customer’s capacity to pay. Just a few data
of fraud. sources that may be available for analysis (with
the customer’s permission) are emails, SMS, and
— Credit qualification. Lenders seeking to determine e-commerce expenditures.
if a customer qualifies for a particular type of
loan have for many years used rule-based or — Pricing. Banks generally have offered highly
logistic-regression models to analyze credit standardized rates on loans, with sales
bureau reports. This approach, which relies on representatives and relationship managers
a narrow set of criteria, fails to serve a large having some discretion to adjust rates within
segment of consumers and SMEs lacking a formal certain thresholds. However, fierce competition
credit history, so these potential customers turn on loan pricing, particularly for borrowers with a
to nonbank sources of credit. In recent years, strong risk score, places banks using traditional
however, leading banks and fintech lenders approaches at a considerable disadvantage
have developed complex models for analyzing against AI-and-analytics leaders. Fortified with
structured and unstructured data, examining highly accurate machine-learning models for risk
hundreds of data points collected from social scoring and loan pricing, AI-first banks have been
media, browsing history, telecommunications able to offer competitive rates while keeping their
4
Erik Lindecrantz, Madeleine Tjon Pian Gi, and Stefano Zerbi, “Personalizing the customer experience: Driving differentiation in retail,” April
2020, McKinsey.com.
4 AI-powered decision making for the bank of the futurerisk costs low. Some are also using their decisioning (e.g., fraud by sales agents), customer fraud, and
capabilities to quantify the customer’s propensity payment fraud (including money laundering and
to buy according to the customer’s use of different sanctions violations). Banks should continuously
types of financial products. Some even leverage update their fraud detection and prevention
natural-language processing (NLP) to analyze models, as we discuss later with regard to edge
unstructured transcripts of interactions with sales capabilities. Ping An, for example, uses an image-
and service representatives and, in some cases, analytics model to recognize 54 involuntary micro-
collections personnel. By basing the offered rate expressions that occur before the brain has a
on both creditworthiness and propensity to buy, the chance to control facial movements.⁵
bank can optimize the balance of total asset volume,
risk, and interest income within a lending portfolio. AI-driven credit decisioning can build the business
while lowering costs. Sharper identification of risky
— Fraud management. As competition for credit customers enables banks to increase approval
relationships becomes concentrated in digital rates without increasing credit risk. What is more, by
channels, the automated processing of loan automating as much of the lending journey as possible,
applications and use of AA/ML models to expedite banks can reduce the costs of support functions and
credit approval and disbursement of funds not strengthen each customer’s experience with faster
only positions the bank to acquire new customers loan approval and disbursement of funds, fewer
and increase market share, but also opens new requests for documentation, and credit offers precisely
opportunities for fraud. The costliest instances of tailored to meet customer needs. Exhibit 2 illustrates
fraud typically fall into one of five categories: identity how AI-enabled decisioning capabilities underpin a
theft, employee fraud, third-party or partner fraud customer’s onboarding journey.
5
“Chinese banks start scanning borrowers’ facial movements,” Financial Times, October 28, 2018, ft.com.
Exhibit 2
The combination
The combination of
ofAI
AI and
andanalytics
analyticsenhances
enhancesthethe
onboarding journey
onboarding for for
journey eacheach
new customer.
customer.
Name: Joy Family: Married, no children
Age: 32 years Profile attribute: Avid traveler
Occupation: Working professional
Propensity-to-buy
model to identify AA-enabled real-
whom to retarget time credit Joy
Joy’s Channel propensity underwriting, limit completes Hyperpersonalized
Joy online know-your-
landing page to identify right receives a call to assessment, and cross-sell and
shows her per- outreach channel assist in journey; pricing customer form, upsell offers
sonalized offers: provides details for
caller also employment verifi-
personal loan for informs
travel, 5% off on cation, and sets up
Joy about custom an online pay-
travel insurance travel services
Joy receives a ment mandate Loan disbursed to
Joy completes a
WhatsApp
streamlined Joy; curated
reminder for
3-click journey catalog of offers
personal loan
Analytics-backed to see the ahead of Joy’s
for travel at zero Analytics-enabled AI capabilities to
hyperpersonalized customer–caller offer terms and conduct relevant travel sent by
processing
offers based on mapping with conditions fraud checks (eg, email
charges
customer specific cues facial recognition
microsegment provided to caller with know-your-
customer docs)
AI-powered decision making for the bank of the future 5Monitoring and collections variety of external data partnerships for location
Once a bank has employed AA/ML models to data and transaction history can help the bank
automate loan underwriting and pricing, it can also understand both the customer’s position and the
deploy AI and advanced analytics to reduce the most effective approach, or contact strategy, for
burden of nonperforming loans. Increasingly, banks averting default (Exhibit 3).
are engaging with clients proactively to help them
keep up with payments and work more closely Contact strategy. To determine an appropriate
with clients who encounter difficulties. By drawing contact strategy for customers at risk of default,
upon internal and external data sources to build a banks can segment accounts according to value
360-degree view of a customer’s financial position, at risk (VAR), which is the loan balance times
banks can recognize early-warning signals that a the probability of default. This allows banks to
borrower’s risk profile may have changed and that focus high-touch interactions on borrowers that
the risk of default should be reassessed. account for the highest VAR; banks can then use
low-cost channels like telephoning and texting for
Beyond conventional data sources like repayment borrowers posing less risk. Banks have used this
data and credit bureau reports, banks can digitize approach to reduce both the cost of collections
and leverage other interaction data from campaigns, and the volume of loans to be resolved through
field visits, and collection agents’ comments to restructuring, sale, or write-off.⁶
draw insights for collections strategy. Further, a
6
Ignacio Crespo and Arvind Govindarajan, “The analytics-enabled collections model,” April 2018, McKinsey.com.
Exhibit 3
Advancedanalytics
Advanced analyticsandand machine
machine learning
learning cancan classify
classify customers
customers into into
microsegments for targeted interventions.
microsegments for targeted interventions.
Customer type
True low-risk Absentminded Dialer-based True high-touch Unable to cure
Targeted Use least Ignore or use Match agents to Focus on customers Offer debt-
intervention experienced agents interactive voice customers; send live able to pay and at restructuring
provided with set message (segment prompts to agents high risk of not settlements early for
scripts will probably to modify scripts paying those truly
self-cure) underwater
Impact Onscreen prompts 10% of time saved, Matching and Added focus Significant increase
guide agent–client allowing for prompts can addresses higher in restructuring and
conversation based reassignment of increase sense of probability of default settlements
on probability of agents to more connection and rates in this segment increases chance of
breaking promises difficult customers likelihood of paying collecting at least
and specific part of debt
campaigns
Source: Ignacio Crespo and Arvind Govindarajan, “The analytics-enabled collections model,” McKinsey on Payments, August 2018, McKinsey.com
6 AI-powered decision making for the bank of the futureTreatment strategy. If contact strategies through a prediction algorithm to estimate the ideal
various channels are inadequate to help the product-per-customer (PPC) ratio for each
customer resume timely payment, banks must user, based on individual needs. If analysis of
pursue stronger measures, according to the a customer’s needs produces an anticipated
customer’s ability and willingness to pay. Customers product usage ratio of eight but the customer
with high willingness but limited ability to pay in the uses only two products, the relationship
short term may require restructuring of the loan manager receives a prompt to reach out to the
through partial-payment plans or loan extensions. customer and cross-sell or up-sell relevant
In cases where the customer exhibits both low ecosystem products.⁷
willingness and limited ability to pay, banks should
focus on early settlement and asset recovery. Servicing and engagement
Advanced analytics, enabled by unstructured AI-powered decisioning can enable banks to
internal data sources such as call transcripts create a smart, highly personalized servicing
from collections contact centers and external experience based on customer microsegments,
data sources such as spending behavior on other thereby enabling different channels to deliver
digital channels, can improve the accuracy of superior service and a compelling experience
determinations of ability and willingness to pay. with interactions that are fast, simple, and
intuitive.⁸ Banks can support their relationship
Deepening relationships managers with timely customer insights and
Strong customer engagement is the foundation tailor-made offers for each customer. They can
for maximizing customer value, and leaders also significantly improve agents’ productivity
are using advanced analytics to identify less with streamlined preapproved products crafted
engaged customers at risk of attrition and to craft to meet each customer’s distinct needs. Models
messages for timely nudges. As with any customer that analyze voice and speech characteristics
communication in a smart omnichannel service can match agents with customers based on
environment, each personalized offer is delivered behavioral and psychological mapping. Similarly,
through the right channel according to the time of transcript analysis can enable prediction of
day. Rich internal data for existing customers can customer distress and suggest resolution to
enable financial institutions to create a finely tuned the agent.
outreach strategy for each individual customer,
guided by risk considerations.
Deployment of AA/ML models at
Deeper relationships are predicated on a bank’s scale
precise understanding of a customer’s unique Leveraging AI to automate decision making in
needs and expectations. A bank can craft offers to near real time is a complex and costly endeavor.
meet emerging needs and deliver them at the right If banks are to earn the required return on their
time and through the right channel. By doing so, the technology investments, they must begin with a
bank demonstrates that it understands customers’ strategy and road map to capture maximal scale
current position and aspirations and can help them benefits in the design, building, and deployment
get from the former to the latter. For example, by of AA/ML models.
analyzing browsing history and spending patterns, a
bank might recognize a consumer’s need for credit As banks embark on this journey, leaders must
to finance an upcoming purchase of a household encourage all stakeholders to break out of siloed
appliance. Analysis of internal data on product mindsets and think broadly about how models
usage can also reveal areas where the bank can can be designed for uses in diverse contexts
make its offering more relevant to a customer’s across the enterprise. AI-first organizations have
current needs. Ping An, for example, has developed succeeded by organizing the effort around four
7
“Ping An Bank: Change everything,” Asiamoney, September 26, 2019, asiamoney.com.
8
Violet Chung, Malcolm Gomes, Sailee Rane, Shwaitang Singh, Renny Thomas, “Reimagining customer engagement for the AI bank of the
future,” McKinsey.com, October 2020.
AI-powered decision making for the bank of the future 7main elements: First, they prioritize the analytics environment where cross-functional teams can
use cases with the biggest impact on customer experiment with different approaches to achieving
experience and the most value for the bank. the value-generating goals of a particular use case,
Second, they ensure that the data architecture, moving from minimal viable product to scalable
data pipelines, application programming interfaces solution in a matter of weeks. Building AA/ML models
(APIs), and other essential components are available at scale and deploying them across the enterprise
for building and deploying models at scale through depend on matching the right talent and skills with
standardized, repeatable processes.⁹ Third, they each of the roles required for a successful analytics
establish a semiautonomous lab for experimentation lab and factory (Exhibit 4).10
and prototype development and set up a factory for
industrial-scale production of the solution. Fourth, The lab combines talent from business, analytics,
they assemble the right mix of talent for agile, cross- technology, operations, and more. There are two
functional teams and empower them to maximize main technical roles. One is the data scientist,
value in close alignment with enterprise strategy. who is responsible for identifying the analytics
techniques required to meet the business goal and
Several leading banks have established semi- for programming advanced analytics algorithms.
autonomous labs offering a test-and-learn The other is the data engineer, who scopes the data
9
Tara Balakrishnan, Michael Chui, Bryce Hall, and Nicolaus Henke, “The state of AI in 2020,” November 2020, McKinsey.com.
10Nayur Khan, Brian McCarthy, and Adi Pradhan, “Executive’s guide to developing AI at scale,” October 2020, McKinsey.com.
Exhibit 4
Diverse roles are necessary for building and deploying AA/ML models at scale.
Diverse roles are necessary for building and deploying AA/ML models at scale.
Lab environment Factory environment
Product owner Data scientist Product owner ML engineer
Leads the squad; Frames the First point of Optimizes ML
typically a business problem contact for models for
business owner and develops external stake- performance and
who provides voice advanced analytics holders; defines scalability; deploys
of the customer algorithms the solution criteria the models into
production
Data engineer Translator DevOps engineer Infrastructure
Scopes the data Interfaces Develops CI/CD1 architect
available; builds between business pipelines to Designs
data architecture and technical automate parts of infrastructure
and data pipelines stakeholders the software- components for the
deployment pipeline analytics use case
Designer Delivery manager Full-stack
Focuses on Responsible for all developer
interaction aspects of delivery Develops software
between end of the analytics components for
users and the solution to meet the back and front
analytics solution squad goal ends of AI solutions
output
1
Continuous integration and continuous deployment.
8 AI-powered decision making for the bank of the futureavailable, identifies major sources of data to be Some edge technologies already afford banks
consolidated for analytics, develops data pipelines to the opportunity to strengthen existing models
simplify and automate data movement, and sets up with expanded data sets. For example, many
data architecture for storage and layering. In addition, interactions with customers—via telephone,
the role of translator is crucial to ensure consistent mobile app, website, or increasingly, in a
communication and smooth collaboration between branch—begin with a conversational interface
business leaders and analytics specialists. to establish the purpose of the interaction and
collect the information required to resolve the
On factory teams, one of the primary technical query or transfer it to an agent. A routing engine
roles is the DevOps engineer, who is responsible can use voice and image analysis to understand
for developing continuous integration (CI) and a customer’s current sentiment and match the
continuous deployment (CD) pipelines for deploying customer with a suitable agent. The models
software. In addition, the full-stack developer is underpinning virtual assistants and chatbots
responsible for developing software components employ NLP and voice-script analysis to increase
for other layers of the stack. The machine-learning their predictive accuracy as they churn through
engineer prepares models for deployment at scale, vast unstructured data generated during
and the infrastructure architect ensures that the customer-service and sales interactions.
analytics solution is compatible with the architecture
of the core tech and data layer of the capability stack. While each customer-service journey presents an
opportunity to deepen the relationship with the
The lab-and-factory setup requires flexible and help of next-product-to-buy recommendations,
scalable technologies to handle the changing banks should constantly seek to improve their
requirements of analytics engines. It is also important recommendation engines and messaging
to give analytics teams access to the centralized campaigns. Feedback loops, for example, can
data lake, and these teams must be able to draw help marketing teams and frontline officers
upon raw data from diverse sources to generate data gauge the effectiveness of an offer by analyzing
sets to be used in building models. The technology customers’ ongoing browsing and transaction
supporting the solution must be modular to allow the activity within the bank’s digital ecosystem and
transfer of developed solutions to factory production beyond (Exhibit 5, next page).
using DevOps tools. Finally, it is crucial to embed
performance management and risk controls within As edge capabilities become more powerful,
models to avoid adverse impacts on operations. leaders are developing new, increasingly
complex analytics solutions to create a superior
Once the lab has developed a model, the factory experience and introduce distinctive innovations.
takes over, running 24/7 to put the model into Use of computer vision and voice-to-script
production and deploying it at scale in diverse use conversion can speed the completion of forms—
cases across the enterprise. for instance, enabling a customer to respond
orally to questions and upload documents
from which relevant data can be extracted
Augmented AA/ML models with edge automatically using optical character recognition
capabilities (OCR). Facial and sentiment analysis during an
The rapid improvement of AI-powered technologies in-person consultation or videoconference can
spurs competition on speed, cost, experience, and support frontline representatives with messages
intelligent propositions. To maintain its market and offers finely tuned to the customer’s needs
leadership, an AI-first institution must develop and aspirations.
models capable of meeting the processing
requirements of edge capabilities, including natural- Several banks use voice recognition to verify
language processing (NLP), computer vision, facial customer identity for certain low-value, high-
recognition, and more. volume transactions. Some are using facial
AI-powered decision making for the bank of the future 9Exhibit 5
Edgecapabilities
Edge capabilitiesenhance
enhance customer-service
customer-service journeys.
journeys.
Digital self-cure Voice-recognition- AI-enabled Voice-analytics- AI-enabled Feedback loop via
channels for enabled IVR,1 customer profiling, and NLP-enabled2 service-to-sales engagement
customers (eg, frontline bots customer–agent customer- engine channels
WhatsApp, mobile handling 30–50% matching sentiment analysis
app, website) of queries
Customer may If query is not Customer Contact-center Customers are Postcall feedback
seek immediate resolved through connected to the agents supported prompted for any and automated
resolution of automated appropriate agent by live feedback specific offers follow-up occur via
queries through channels, (chat or call) based and prompts to preapproved for digital channels
digital self-service customer may on type of query sustain superior them
channels contact bank via and customer customer
chat or request profiling experience
call with live agent
1
Interactive voice response.
2Natural-language-processing-enabled.
recognition to authenticate customers’ identity as soon of customer behaviors, banks must go the last mile to
as they enter a branch, approach an ATM, or open the ensure that these analytical insights have an impact
banking app on a mobile device. As noted earlier, facial on customer behavior, such as purchasing a product,
analysis is also useful in identifying making a loan payment, or exploring new service
potential fraud. offers. In other words, an organization must establish
a mechanism to “translate” analytical outcomes
Leading banks are using blockchain to create smart into compelling messages to communicate to the
contracts, secure trade documents and automate the customer at the right time, through the preferred
release of funds upon delivery of goods, and establish channel—be it email, SMS, mobile app, website,
shared utilities to reduce the burden of know-your- branch staff, or a relationship manager—according
customer (KYC) and anti-money-laundering (AML) to the time of day.
compliance for banks and customers. Edge capabilities
deployed as part of an enterprise strategy to enhance This last mile from decisioning to messaging is the
the AI bank’s value proposition have the potential not domain of the digital marketing engine. Seamlessly
only to improve credit underwriting and fraud integrated with applications across the full AI-and-
prevention but also to reduce the costs of document analytics capability stack with the help of APIs,
handling and regulatory compliance. from data infrastructure to engagement channels,
this engine supports the nearly instantaneous
processing of raw data to produce tailored messages
Enterprise-wide digital marketing engine communicated via engagement channels. Exhibit 6,
While the automated decisions generated by AA/ML on the next page, illustrates the position of the digital
models provide highly accurate, real-time predictions
10 AI-powered decision making for the bank of the futureExhibit 6
The digital
The digitalmarketing
marketingengine
engine requires
requires fullfull stack
stack capabilities.
capabilities.
Engagement layer
Branch Contact ...
Email SMS Mobile app Website
center
Channel
Decisioning layer analytics/
Martech stack Design and activation feedback loop
Customer
relationship Data mgmt. Content Ad tech Campaign Testing
management platform mgmt. mgmt.
(DMP) platform server platform platform
(CRM)
AA/ML1 models: Decisioning and personalization engine
Edge Voice Virtual Computer Facial Blockchain Robotics Behavioral
NLP2
capabilities analysis agents vision recognition analytics
Data lake layer
Raw data lake Curated data lake
Other use cases Customer 360 Personalization store
Additional derived
elements for personalization
Comprehensive set of data sources: internal structured data (eg, applications, product holding, payment behavior),
unstructured data (eg, call logs), CRM, external data (eg, telco, clickstream data), campaign performance data
1
Advanced-analytics and machine-learning.
2Natural-language processing.
marketing engine (or martech stack) within the and modified; (2) the ad tech server, which
decisioning layer of the AI-bank capability stack. automates advertisements based on data
analysis; and (3) the campaign management
The digital marketing engine comprises platforms platform, which supports the creation and
and applications fulfilling four main functions: management of marketing campaigns, which
data management, design and activation, are conducted automatically according to the
measurement and testing, and channel analytics. microsegmentation generated by the data
The data management platform, which forms part management platform.
of the core tech and data infrastructure layer of
the AI-bank capability stack, supplies the data Just as the AI-and-analytics capability
used to create and manage target customer stack entails fundamental changes in the
segments. The design and activation function organization’s talent, culture, and ways of
has three elements: (1) the content management working, the success of digital marketing
platform, where messages, offers, advertisements, capabilities depends on an agile operating
and other interventions are created, managed, model. This model consists of autonomous
AI-powered decision making for the bank of the future 11cross-functional teams (or pods) drawing upon (CBA) leverages its mobile app to test messages
the talent of different parts of the enterprise, such and learn within hours what works and what must
as business units, marketing, analytics, channels, be changed. This cadence enables rapid scaling of
operations, and technology. Each pod should also campaigns to similar customer segments.11
include representatives from partner organizations
crucial to the digital marketing effort—for example, In measurement of campaigns’ impact, scientific
user-interface and user-experience designers, rigor is crucial. To allow for precise measurement
who lay out the campaign’s look and feel and its of the incremental value of the campaign, each
flow, and copywriters, who finalize the language target segment should include a control group of
of any intervention. The members of each pod customers excluded from the campaign. The tools
collaborate on developing, managing, and improving and capabilities for evaluating the effectiveness of
engagement campaigns, and each member is customer-engagement campaigns help employees
accountable for campaigns’ impact according to across the organization understand how they can
clearly defined key performance indicators (KPIs). enhance their impact on individual customers and
add value to an AI-oriented culture.
To achieve the desired outcome, an AI-first bank
launching daily personalized communications to
millions of customers must build tools for continual
testing and learning. The measurement and testing The rapid improvement of AI-powered technologies
platform flags potential aspects of content or spurs competition on speed, cost, experience, and
distribution to improve, thereby enabling teams to intelligent propositions. To remain competitive,
evaluate in real time the effectiveness of campaigns. banks must engage customers with highly
personalized and timely content to build loyalty.
Another source of continual feedback is channel Personalized offers with tailored communication
analytics, which includes tools and dashboards delivered at the right time through the customer’s
for real-time tracking of engagement across each preferred channel can help banks maximize the
target segment. Every day, each pod leverages the lifetime value of each customer relationship and
channel analytics and measurement and testing reinforce the organization’s market leadership.
platforms to closely track various indicators, To achieve these benefits, banks must build
including delivery rates, email open rates, click- AI-powered decisioning capabilities fueled by a rich
through rates by channel for customers seeking mixture of internal and external data and augmented
more information (the first call to action), conversion by edge technologies. The core technology and
rates, and more. These diagnostics help members data infrastructure required to collect and curate
of the pod experiment with potential enhancements increasingly diverse and voluminous data sets is the
to messages, advertisements, and campaign design. topic of the next article in our series on the AI-bank
As an example, Commonwealth Bank of Australia capability stack.
Akshat Agarwal is an associate partner in McKinsey’s Bangalore office. Charu Singhal a consultant and Renny Thomas is a
senior partner, both in the Mumbai office.
Copyright © 2021 McKinsey & Company. All rights reserved.
Paul McIntyre, “CommBank’s analytics chief on how its AI-powered ‘Customer Engagement Engine’ is changing everything,” Mi3, September
11
21, 2020, mi-3.com.au.
12 AI-powered decision making for the bank of the futureYou can also read
