ROT364
PUTTING
INTELLIGENCE
AI has already shown itself to be highly e ective in creating
value across industries and functions, leaving little doubt
that it will fundamentally transform business.
by Philipp Gerbert, Martin Hecker, Sebastian Steinhauser and Patrick Ruwolt
AI gives companies the opportunity to
room and entered the world of mainstream business. Today’s AI -
algorithms already support remarkably-accurate machine sight, actions. The stakes are huge: Brands that integrate advanced
hearing and speech, and can access global repositories of infor- digital technologies and proprietary data to create personalized
mation. Thanks to deep learning and other advanced techniques, experiences can increase revenue by six to 10 per cent — two to
a staggering level of growth in data, and continuing advances in three times the rate among brands that don’t. In retail, health-
raw processing power, AI performance continues to improve,
leading to an explosion in AI-enabled business applications. shift toward the top 15 per cent of personalization companies
As always, this new era will have winners and losers. But our
research suggests that if current patterns continue, the separa- Many best practices of successful personalization have
tion between the two could be especially dramatic and unforgiv- emerged in fast-moving retail environments. One global re-
- tailer, for example, used a loyalty app’s smartphone data — in-
ing and spreading AI within large organizations. cluding location, time of day and frequency of purchases — to
gain a deep understanding of its customers’ weekly routines. By
Artificial Intelligence in Action combining millions of individual data points with information
- on general consumer trends, the retailer built a real-time mar-
cant adoption of AI in business remains low. According to our sur-
vey, only one in 20 companies has extensively incorporated AI. In some sales-and-marketing organizations, AI augments
Nevertheless, every industry includes companies that are ahead rather than automates processes. For example, a multiline in-
of the pack. The following uses demonstrate how pervasive AI surer relied on machine learning to segment its customers in
rotmanmagazine.ca / 61
intersection of a customer’s needs and the insurer’s objectives repair frequency — to optimize designs. AI is especially helpful
— to the company’s sales agents. To accomplish this, the insurer in developing designs for additive manufacturing, also called 3D
built a model of the insurance needs of customers as they pass printing, because its algorithmically-driven processes are un-
through various life stages. The model relied on complex algo- constrained by engineering conventions.
rithms that crunched more than a thousand static and dynamic Aggressive forms of data collection should be a key ele-
variables encompassing demographic, policy, agent tenure, and ment in the design of AI pilots in R&D. It may be necessary to
sales history data. As a result, the insurer could match particular collaborate with universities, digitize old records, or even gener-
policies with individual members of specific clusters. The system ate new data from scratch. Given the knowledge and expertise
has the potential to increase cross-selling by 30 per cent. The in- required to engage in R&D, useful turnkey AI solutions will rare-
surer can also use the machine learning system to improve sales ly be available. Instead, scientists must rely on systematic trials
optimization efforts by processing geographic, competitive, and for guidance in building the data inventory they need for future
agent performance data. AI applications.
Such examples demonstrate the effectiveness of AI in de-
centralized settings — such as retail or financial services sales— OPERATIONS. Operational practices and processes are naturally
that benefit from a rich supply of contextual and specific cus- suited for AI: They often have similar routines and steps, gener-
tomer data. Properly constructed pilot projects can generally ate a wealth of data, and produce measurable outputs. Many AI
validate a proof of concept within four to six weeks and help de- concepts that work in one industry will work in another. Popu-
termine the data infrastructure and skill building necessary for lar current uses of AI include predictive maintenance and non-
a full rollout. linear production optimization, which analyzes a production
environment’s elements collectively rather than sequentially or
RESEARCH AND DEVELOPMENT. R&D problems are complex, deeply in isolation.
technical, and bounded by hard scientifi constraints. Even so, An oil refinery wanted to predict and avoid breakdowns
AI has high potential in this fi For instance, in the biophar- of an important gasifier unit responsible for converting resid-
maceutical industry, where R&D is the primary profi driver, AI ual products of the refining process into valuable synthesis gas
could reverse the trend toward higher costs and longer develop- used in generating electricity. An unplanned outage of that unit
ment times. Citrine Informatics, an AI platform designed to forced a costly suspension of electricity generation for a month.
accelerate product development, illustrates one way to meet the Although the refinery had accumulated plenty of data about on-
challenge of limited data in R+D. going operations, it had no clear understanding of what specific
Most published studies have a bias toward successful ex- factors drove the unit to break down. Conventional engineering
periments and, potentially, the interests of the funding organiza- models could not fully describe the complex interdependencies
tion. Citrine overcomes this limitation by collecting unpublished that existed among more than a thousand variables that might
data through a large network of relationships with research in- lead to failure.
stitutions. “Negative data are almost never published. But the Working closely with data scientists, the refinery’s engi-
corpus of negative results is critical for building an unbiased neers turned to artificial intelligence to determine the cause of
database,” says Bryce Meredig, Citrine’s co-founder and chief the breakdowns, feeding six years’ worth of operational data
science officer. This comprehensive approach has enabled the and maintenance information through a machine learning al-
company to cut R&D time for its customers by one-half for gorithm. The AI model successfully quantified the impact of all
specific applications. key factors (including feedstock type, output quality, and tem-
Within the industrial goods sector, leading manufacturers perature) on overall performance. Engineers were then able to
combine AI, engineering software, and operating data — such as gauge whether the unit would continue to run between episodes
62 / otman Management Spring 2018
of planned maintenance. bots; semi-automated contract design and review; and sourcing
Relying on insights that the machine learning algorithm recommendations based on analysis of news, weather, social
generated, engineers designed a transparent, rules-based sys- media, and demand. Signifi augmentation or even automa-
tem for adjusting key operational settings for variables such as tion of sourcing is only now emerging.
steam and oxygen to enable the unit to keep running between Supply chain management and logistics are a different
scheduled maintenance periods. This system minimizes the risk story. Historical data is readily available for these processes,
of unplanned shutdowns of the unit and reduces the number of making them a natural target for machine learning. One global
short-term changes in the maintenance schedule, yielding sig- metals company recently built a collection of machine learning
nificant economic benefits. engines to help manage its entire supply chain, as well as to pre-
Predictive maintenance solutions can also work for hu- dict demand and set prices. The company integrated more than
mans. A U.S. insurer receiving fixed payments from Medicare 40 data warehouses, ERP systems, and other reporting systems
wanted to use AI to reduce avoidable visits to the doctor or to into one ‘data lake’. As a result of these changes, the systems can
hospitals by Medicare patients. The insurer fed data from medi- now identify and predict the way complex and opaque demand
cal histories, such as adverse reactions to drugs, and case man- patterns ripple throughout the supply chain. For example, a shift
agers’ notes into a machine learning system. The system devised in the U.S. corn harvest by a single week has global repercus-
an intelligent segmentation of customers and provided useful sions along the supply chain for aluminum, a common packag-
insights into preventive action. For instance, the recent loss of ing material for corn.
a patient’s spouse proved to be highly predictive of future medi- The company’s initiative helped improve its customer ser-
cal intervention and the need for preventive care. These insights vice levels by 30 to 50 per cent. It is also set to achieve a two-
allowed the payer to redesign programs to achieve potential an- to-four per cent increase in profit margin within three years and
nual savings of $650 million. a reduction in inventory of between four and ten days within
Moving beyond maintenance, a smelter operator used AI two years. This example highlights the importance of data, data
and non-linear optimization to improve the purity of its copper, preparation, and data integration in bringing AI to life. It takes
which engineers had spent years trying to do. Working with a far more time to collect data and build the data infrastructure
team of data scientists, the engineers fed five years of histori- than to build a machine learning model.
cal data into a neural network. The system suggested production
changes that resulted in a two per cent increase in purity—an SUPPORT FUNCTIONS. Companies often partially outsource sup-
improvement that doubled the smelter’s profit margins. The ex- port functions, which tend to be similar across organizations.
ercise took six weeks and did not require additional capital or But soon, they may be able to buy AI-enabled solutions for these
operational spending. processes. Heavy AI development is underway at outsourcing
giants such as IBM, Accenture, and India’s Big Four players
PROCUREMENT AND SUPPLY CHAIN MANAGEMENT. In procurement, (HCL, Infosys, Wipro and Tata). These companies are shifting
structured data and repeat transactions are common and AI’s focus from emphasizing lower labour costs and scale to building
potential is substantial but largely unrealized. Machines today intelligence and automation platforms in order to offer higher-
can beat the top poker players in the world and trade securities, value services.
but they have not yet shown the ability to outsmart vendors in Many service organizations are starting to recognize the
corporate purchasing — at least publicly. Companies may be benefits of combining AI with robotic processing automation
using AI-enabled procurement systems but not telling their (RPA). They are using rules-based software bots to replace hu-
suppliers, or anyone else, in order to maintain a competitive man desk activity and then adding flexibility, intelligence and
edge. The known examples of AI in procurement involve chat- learning via AI. This approach combines the rapid payback of
rotmanmagazine.ca / 63
The AI Building Blocks
DATA P OCESSING ACTION
AI Building Blocks
Machine Vision Information Processing Image Generation
Detecting faces and objects easoning by analogy Modifying faces in a picture
in images to related concepts
Speech Generation
Speech ecognition Learning from Data Providing speech to a virtual
Transforming spoken words Learning to drive a car from assistant
into text recorded driving data Handling and Control
Picking up novel items
Natural-language Processing Planning and Exploring Agents
Detecting the intent in Building a map while exploring Navigating and Movement
a text-based command the environment Avoiding obstacles as an
autonomous vehicle
AI Platforms and Services
AI Infrastructure and Algorithms
Source: BCG Analysis
FIGU E ONE
RPA and the more advanced potential of AI. To replace human In the auto industry, for example, Bosch is investing €300
tasks, one Asian bank installed RPA and AI systems that learned million over the next five years to establish AI facilities in Germa-
on the fly. These systems routed cases to human workers only ny, India and the U.S. “Ten years from now, scarcely any Bosch
when they were uncertain about what to do, enabling the bank product will be conceivable without artificial intelligence. Either
to reduce costs by 20 per cent and decrease the time devoted to it will possess that intelligence itself, or AI will have played a key
certain processes from days to minutes. role in its development or manufacture,” said Volkmar Denner,
the company’s CEO. At the same time, automation creates new
PRODUCT AND SERVICE OFFERINGS. Unlike most of the prior exam- business models. Insurers and manufacturers, for example, will
ples, AI applications that involve advanced product and service be able to use AI to predict risk with greater accuracy, allowing
offerings — digital personal assistants, self-driving cars, and them to price on the basis of use, care or wear.
robo-investment advisors, for example — tend to receive a lot of
attention. Companies that offer AI-enabled services are eager How Industry Value Pools Could Shift
to demonstrate to the public the competitive performance and Collectively, use cases and potential scenarios will infl ence
features of these offerings. Because their products and services entire industry structures. Self-driving cars, for example, will af-
and potentially their entire business models are at stake, compa- fect not just car manufacturers but also drivers, fl t owners and
nies must build strong internal AI teams. This helps explains the traffi patterns in cities. The city of Boston has determined that
fi ce competition for AI talent among technology vendors, car self-driving vehicles could reduce both the number of vehicles in
manufacturers and suppliers. transit and the average travel time by 30 per cent. Parking needs
64 / otman Management Spring 2018
How Value Could Shift Within Healthcare
Med- Potential Value Shifts
6
tech
Improved operations or clinical decision making benefits most
6 1 companies
2
Increased diagnostic demand outweighs reduced treatments
Tech 2
Ven- 6 Providers 5 Payers 5 Consumers Enhanced diagnostics reduce the need for treatments
dors 3
Demand increases for self-service sample tests and preventive
4 medications
3 4 1 5
1
Consumers benefit from newly achieved savings through payers
5
6 Most players buy at least some solutions from tech vendors
Biopharma 6
1
Net impact of AI: Positive Neutral Negative
Note: Arrow width is proportional to the estimated value impact of AI in five to ten years. The size of each circle reflects the approximate EBIT
value for the corresponding sector in the U.S.
Source: BCG Analysis
FIGU E TWO
would fall by half, and emissions would drop by two-thirds. traction in diagnosing illnesses. Visual AI agents already out-
Healthcare offers another dramatic example. It consists perform leading radiologists at diagnosing some specific forms
of several sectors, including medical technology, biopharma- of cancer, and many start-ups and tech giants are working on
ceuticals, payers and providers, each with distinct and often AI-enabled methods to detect cancer even earlier and to pro-
competing interests. The industry is the scene of rampant AI vide ever more accurate prognoses. In the primary-care setting,
experimentation across the value chain, particularly in the areas AI can improve or replace some physician interactions. Mean-
of R&D, diagnostics, care delivery, care management, patient while, remote diagnostics can eliminate or drastically reduce
behaviour modification and disease prevention. the number of patient visits to the hospital for some conditions.
Figure Two illustrates one potential scenario for how over- These changes are likely to primarily benefit medtech compa-
all healthcare value pools may shift with the increased adoption nies, while possibly hurting biopharma companies and to some
of AI. Of course, value shifts for individual players within sectors extent, providers, as better, earlier diagnoses and methods of
will vary, and there will be winners and losers in each sector. Ini- prevention reduce demand for treatment.
tially, most companies will benefit from the incorporation of AI With its intrinsic performance metrics, artificial intelli-
into internal operational processes. Biopharma companies and gence will likely accelerate the trend toward value-based health
payers are likely to gain the most from these efforts because they care — the practice of paying for outcomes rather than volume.
can take advantage of R&D efficiencies, personalized market- This trend should benefit consumers as payers pass along sav-
ing, and streamlined support functions. ings and set new rates for providers and biopharma companies.
Over the next five years, we expect AI to gain significant Finally, most companies are likely to buy at least some of
rotmanmagazine.ca / 65
An in-depth understanding of developments in the AI building blocks will
be critical for systematically incorporating technology advances.
their AI solutions from technology vendors, including tradi- zation has selected a fi set of pilots, it should run them as test-
tional tech players that enter the healthcare space. This possible and-learn sprints, much as in agile software development. Since
scenario — which would occur against a backdrop of increas- most pilots will still have to deal with kludgy data integration and
ing demand for healthcare — could improve health outcomes, processing, they will be imperfect. But they will help correctly
but biopharma companies could feel the heat. Alternatively, prioritize and defi the scope of data integration initiatives, and
biopharma companies might make bolder moves in diagnostics, identify the capabilities and scale needed for a fully operational
and personalized medicine might take off, opening up new profi AI process. Each sprint should concurrently deliver concrete cus-
pools. Furthermore, payers could themselves develop remote tomer value and defi the required infrastructure and integra-
diagnostics, while providers start incorporating AI into their pa- tion architecture.
tient treatment protocols. In almost any scenario, medtech and
technology vendors will profit. . SCALING UP. The last phase consists of scaling up the pilots into
solid run-time processes and offerings, and building the capabili-
How to Get Started ties, processes, organization, and IT and data infrastructure. Al-
We recommend that executives divide their AI journey into three though this step may last 12 to 18 months, the ongoing rhythm of
steps. agile sprints should maximize value and limit major, unexpected
course corrections.
1. IDEATION AND TESTING. At this stage, companies should rely on
four lenses: customer needs, technological advances (especially In closing
those involving the AI building blocks shown in Figure Two), Leaders across industries need to familiarize themselves with the
data sources, and decomposition (or systematic breakdown) of basics of AI and build an intuitive understanding of what is pos-
processes — to identify the most promising use cases. sible. The good news is, at their core, algorithms are simple; and
Customer needs offer crucial guidance in discovering valu- beyond the mysterious jargon, the fi is quite accessible.
able AI uses. The customers may be external or, in the case The even-better news: What is hard but doable today will
of support functions, internal. An in-depth understanding of likely be easy within a few years, and the impossible today may
developments in AI building blocks will be critical for system- be possible within three to fi e years. Make no mistake, AI will
atically incorporating technology advances. Rich data pools, fundamentally transform business. Your best chance to succeed
especially new ones, provide another important lens, given AI’s is to tune out the hype and do the necessary work. There is no
dependence on them. Finally, by breaking down processes into substitute for action.
relatively routine and isolated elements, companies may uncov-
er areas that AI can automate. Aside from customer needs, these
lenses are quite distinct from those that companies must use to
identify digital opportunities.
For companies with limited AI experience, we strongly rec-
ommend including a second, parallel testing stage based on a
use case that is likely to deliver value, is reasonably well defined,
and is only moderately complex. This test will help the organiza-
tion gain familiarity with AI and will highlight data or data inte- Philipp Gerbert is a Senior Partner and
gration needs and organizational and capability hurdles — criti- Managing Director in The Boston Consulting
Group’s Munich office. Martin Hecker is
cal inputs for the next step. Senior Partner and Managing Director in
BCG’s Cologne office. Sebastian Steinhauser
2. PRIORITIZING AND LAUNCHING PILOTS. Pilots should be prioritized is a Principal in BCG’s Munich Office. Patrick
based on each pilot’s potential value and speed of delivery. The Ruwolt is a consultant in the Munich office.
testing done in the fi st step will provide insight into the time re-
quirements and complexity of potential pilots. Once the organi-
66 / otman Management Spring 2018
