Industry 4.

0: Opportunities and Challenges for Operations

Management

Tava Lennon Olsen

The University of Auckland Business School, 12 Grafton Road, Auckland 1142, New

Zealand

Brian Tomlin

Tuck School of Business, Dartmouth College, Hanover, NH 03755, USA

March 7, 2019

Abstract
Industry 4.0 connotes a new industrial revolution centered around cyber-physical systems. It
posits that the real-time connection of physical and digital systems, along with new enabling
technologies, will change the way work is done, and therefore, how work should be managed.
It has the potential to break, or at least change, the traditional operations trade-os among
the competitive priorities of cost, exibility, speed, and quality. This article describes the
technologies inherent in Industry 4.0 and the opportunities and challenges for research in this
area. The focus is on goods-producing industries, which includes both the manufacturing
and agricultural sectors. Specic technologies discussed include: Additive Manufacturing,
the Internet of Things, Blockchain, Advanced Robotics, and Articial Intelligence.

Key words: Additive Manufacturing; Internet of Things; Blockchain; Robotics; Articial

Intelligence.

1 Introduction
Mechanization, electrication, and computing each drove dramatic and disruptive progress in

the production of goods and services. Industry 4.0, a term rst coined by the German economic

development agency GTAI, is so-named to promote the idea that we are at the dawn of a

new industrial revolution brought about by the emergence, advancement, and convergence of

1
a number of technologies that enable an almost real-time connection between the physical and

digital realms. This digital-physical marriage, driven by Additive Manufacturing, the Internet of

Things, Blockchain, Advanced Robotics, Articial Intelligence, and other related technologies,

is gathering force [and will] be far reaching, aecting every corner of the factory and the supply

chain (McKinsey, 2015b). Individually and collectively, the technologies underlying the concept

of Industry 4.0 hold the promise of reducing costs, enhancing exibility, increasing speed, and

improving quality but more than that, Industry 4.0 oers the possibility of dampening the

tensions inherent between these key operational priorities.

The operations management (OM) academic community must engage with Industry 4.0.

From an educational perspective, we need to equip our students with the knowledge and skills

required to manage the new operations and supply chain realities that will emerge. From a

research perspective, we need to explore whether and how the technologies underpinning In-

dustry 4.0 challenge our current understanding of operations, and more than that, we need to

identify the novel and important operations questions that will emerge from the advancement

and adoption of these technologies.

In this article, we discuss a number of Industry 4.0's foundational technologies, with the dual

goals of (i) building awareness and understanding of Industry 4.0 in the OM community and (ii)

encouraging OM research in this area by identifying opportunities and challenges. With those

goals in mind, we will focus our attention on the operations implications of these underlying

technologies and intentionally omit consideration of other potentially important domains such as

medical and nancial applications. Furthermore, in keeping with the spirit of the term Industry

4.0, we will concentrate on tangible good production rather than service delivery. Of course,

with the rise of product servitization, a trend that may well be accelerated by Industry 4.0, the

distinction between goods and services is necessarily blurred at times. However, we explicitly

omit consideration of smart operations in the public and retail sectors because each is explored

by other articles in this issue; the public sector by Hasija and Teo; and the retail sector by

Martínez-de-Albéniz et al. Within goods production, we consider both the manufacturing and

agricultural sectors.

As will become apparent, Industry 4.0 is not simply an umbrella term for a collection of

disparate technologies. In fact, much of the promise of Industry 4.0 lies in the potential interac-

tions and synergies between subsets of these technologies; for example, advances in sensors and

articial intelligence has allowed the development of collaborative robots that work alongside

people. These synergies notwithstanding, we organize this article by technology and allude to

their potential interactions as necessary. In what follows, we give a high-level overview, along

2
with research implications, of the following technologies: Additive Manufacturing (Ÿ2), the In-

ternet of Things (Ÿ3), Blockchain (Ÿ4), Advanced Robotics (Ÿ5), and Articial Intelligence (Ÿ6).

Other relevant technologies are briey discussed in Ÿ7. We conclude in Ÿ8 with some remarks

on the impact of Industry 4.0 on operations strategy. We note that it is not the goal of this

article to survey the nascent academic OM literature related to Industry 4.0, and so any papers

referenced are in no way intended to be exhaustive or representative.

2 Additive Manufacturing
Additive manufacturing (AM), also known as 3D printing, is a process which takes a digital 3D

representation and produces the associated physical object layer by very-thin layer, joining the

layers as it goes along. Although there are many AM technologies (material jetting, powder bed

fusion, and vat polymerization, for example), the high level AM process ow is common across

these technologies. The initial 3D digital model is converted into a digital .stl le format that

is then virtually sliced into a set of at horizontal layer models. These digitally sliced layers

form the instructions for the AM machine to produce the object layer by layer. Oftentimes, a

nishing step is required after the object is removed from the machine. AM has been heralded

as a revolutionary technology with enormous consequences for operations management. Why?

Let us consider the operational priorities of quality, exibility, speed, and cost.

AM enables the production of complex shapes and internal geometries unattainable by tra-

ditional manufacturing methods. For products in which shape drives performance quality, AM

alleviates the manufacturability constraints facing design engineers. A by-now famous example

is the fuel nozzle for GE's LEAP aircraft engine. GE engineers developed a new nozzle-tip design

that was key for improved fuel eciency, but there was a problem. The tip's interior geometry

was too complex [and] was almost impossible to make (Kellner, 2017). GE was, however, able

to produce the new design using an AM technology, and now manufactures the nozzles using

production-scale AM at a factory in Alabama.

An AM machine is indierent to the shape it is instructed to produce, and therefore does

not require a new set up when switching between the production of objects that dier in their

geometries. Subject to space constraints, some AM machines can simultaneously produce dif-

ferent objects. As such, AM technologies are inherently exible and, in their theoretical limit,

eliminate manufacturing diseconomies of variety.

New product design is an iterative process whereby provisional designs are rened based on

prototype testing. Commercial production begins after the design is nalized. With traditional

manufacturing methods, production of prototypes is time consuming and commercial production

3
may require the development of new product-specic tooling and equipment. Because AM is

exible, prototyping is rapid and commercial production does not need to wait for product-

specic resources. Therefore, speed to market is greatly enhanced.

For similar reasons, the upfront cost of production is also reduced because product-specic

manufacturing investments are not required. All else equal, a lower upfront cost reduces the

sales volume required to break even, thus allowing protable production of low volume products.

Assembly-related variable costs can be reduced by designing products that require fewer com-

ponent parts. In a dierent AM application, GE engineers developed a new turboprop engine

that was assembled from a dozen parts rather than 855 components as was the case in the older

version.

The promised quality, exibility, speed, and cost benets of AM may lead to fundamen-

tal changes in operations strategies. Instead of mass production of a limited variety of prod-

ucts concentrated in a small number of factories, geographically-distributed low-volume produc-

tion of highly customized products becomes increasingly attractive. Taken to its natural limit,

geographically-distributed low-volume production may evolve to personal fabrication whereby

consumers locally print their purchased design on demand. It is not inconceivable that for some

products, the operations strategy of mass production of limited variety may give way to one of

customized production by the masses. That is, the entire architecture of the supply chain may

change.

Service parts for uptime-critical industrial assets, turbines, for example, oer a compelling

case in point for geographically-distributed low-volume production. Service part demand is

highly unpredictable in the timing, location, and part required, and therefore inventory storage

is expensive. Locating AM machines close to assets will enable rapid on-demand printing of the

required service part, with a resulting uptime improvement and inventory reduction. This vision

is becoming a reality, with Siemens claiming to be the rst to commercially print spare parts on

demand for large gas turbines (Müller, 2016).

The promise of AM notwithstanding, a number of signicant disadvantages currently limit its

applicability. Complex geometry is not the sole determinant of product quality. Strength, size,

materials, product uniformity, and many other characteristics matter. AM still faces challenges

on these characteristics. Dierent materials require dierent AM technologies, and this limits

the exibility of AM machines and their ability to produce multi-material products. Although

the new product development time is fast, the AM production cycle time is typically slower than

traditional manufacturing methods, with post-process nishing sometimes being the bottleneck.

And while the upfront cost of production is low, the per-part variable cost of AM production

4
is often high because of high input material costs. Though it should be noted that AM can be

attractive from a sustainability perspective because the quantity of material wasted in production

can be substantially lower.

These speed and cost limitations matter a great deal at scale production, with the result that

AM is currently more attractive than conventional methods only at low production volumes. As

a case in point, HP announced a new AM technology intended for mass production and entered

a partnership with auto component manufacturer GKN in 2018 to deploy HP Metal Jet in their

factories to produce functional metal parts for auto and industrial leaders including Volkswagen

(HP, 2018). However, according to Volkswagen, the sweetspot of 3D printing technologies is

not in giant numbers of vehicles [and] there's a better use case for specialty parts (SCDigest,

2018). Indeed, the HP printers are currently intended only for production of cosmetic pieces

such as customizable car key rings and name plates (SC Digest 2018).

Perhaps because maintenance and repair operations (MRO) are seen as a natural early

application of AM, it has also been an area of recent OM research. Song and Zhang (2018)

model a hybrid multi-part MRO system in which spare parts can be stocked or 3D printed on

demand, and nd that the value of 3D printing is increasing in part variety (as measured by

the number of parts) and decreasing in the part criticality (as measured by the part's outage or

backorder cost). Motivated by a Royal Netherlands Army peace-keeping mission and the fact

that AM-produced parts in the eld may be of lower quality, Westerweel et al. (2018) explore a

dual-sourcing service part system in which regular spare part orders can be lled after a lead time

but AM can be used to on-demand print temporary replacement parts. Knous et al. (2019)

explore how the part consolidation potential of AM inuences overall lifecycle costs when spare

parts management is taken into account. AM aside, the MRO space has undergone signicant

change over the past twenty years, with OEMs such as Siemens and GE eager to take on an

increasing role in the protable after-market service business. The implications of AM for MRO

contracting and supply chain structures may be profound and, therefore, merit future research.

Away from the MRO space, the exibility benets of AM in terms of upfront product-line

design have been the subject of recent research. Dong et al. (2016) examine the impact of

exibility economics on assortment planning, i.e., which product variants to oer, and nd that

as compared to traditional exible production technologies whose variable capacity cost grows in

product variety, the adoption of AM, whose variable capacity cost is invariant to variety, leads

the rm to provide more variety. Sethuraman et al. (2018) explore the personal fabrication

potential of AM in the context of a monopolist rm that sells a product design to customers

who can then use AM to produce the product at a quality of their choosing.

5
 D'Aveni, a strategy scholar who has studied and written extensively about AM, claims that:

We're headed toward an assembly-less world where there are no supply chains because

everything's made together at the same time in one product build-up or printing job.

Such a future is not without its challenges. For starters, companies will face dicul-

ties adjusting to the new manufacturing reality, where purchasing components and

sub-assemblies disappears, and the R&D department will reign supreme. (Blanding,

2018).

Whether or not such a future comes to pass any time soon, this vision presents opportunities

and challenges for the OM academic community.

3 The Internet of Things

and translates its input into an electrical signal that can be processed by electronic circuitry.

According to a report from Deloitte University Press (Holdowsky et al., 2015), the average price

of a sensor fell by over 90% between 1992 and 2014 (from $22.00 to $1.40) and microprocessor

clockspeeds increased by a factor of 991 (from 29 million Hz to 28751 million Hz) over the same

time frame. This conuence of cost reduction and processing speed, along with advancements

in measurement and communications technologies, has enabled the vision of a vast array of

interconnected sensors on machines, people, and products coupled with intelligent controllers,

broadly dened, that can take actions based on real-time sensor readings; in other words, the

Internet of Things (IoT).

Whereas in the past, decision makers operated somewhat in the dark, lacking full information

on the current state of their relevant environments, communication between sensor-enabled

devices promises to enhance the information quality and information completeness available

to decision makers (Saghaan et al., 2018). By quality, we mean the accuracy or precision

with which some nominal property is measured. By completeness, we mean that by combining

sensor measurements of dierent local attributes one can build system-wide state information.

Crucially, the vision of real-time communication in the IoT will enhance the timeliness associated

with this information quality and completeness. Indeed, the strategic signicance of the IoT is

born of the ever-advancing ability to break that [darkness] constraint, and to create information

without human observation, in all manner of circumstances that were previously invisible,

(Holdowsky et al., 2015, p.5).

A McKinsey report estimated that by 2025 the economic impact of the IoT would be in

6
the trillions of dollars, with the majority of this impact attributed to operations applications

in the areas of inventory, maintenance, worker productivity, and optimization opportunities

(McKinsey, 2015a, p.111-112). McKinsey is far from alone in projecting profound implication

for operations. It is the information completeness, quality, and timeliness promised by the IoT

that is the foundation of such projections.

There is an echo of recent history in this heralding of information-driven improvements

to inventory management. Radio-frequency identication (RFID) tags and readers promised a

revolution in the tracking and monitoring of inventory as it owed through a supply chain because

RFID enables high quality (i.e., accurate) and timely (no periodic-auditing delay) inventory

records in a way that barcoding cannot. Industry's excitement about the potential of RFID was

soon followed by a surge of OM research that explored the benets of inventory record accuracy

and real-time visibility of the inventory pipeline progress; see, for example, Heese (2007); Lee

and Özer (2007); Gaukler et al. (2008); and others. Evaluating the benet of perfect inventory

information - a common, if often implicit, assumption in the traditional inventory literature

 can be done by developing a policy that accounts for information inaccuracy or delay and

then comparing it to a perfect-information benchmark policy; see DeHoratius et al. (2008) for

example. If accuracy and timeliness of inventory information were the only operational benets

of the IoT, then prior RFID-related research has already answered many of the related research

questions.

Although RFID technology plays a role in the IoT, sensors

to be gathered. Not only can a rm have timely and accurate data on inventory levels, ubiquitous

sensors allow it to have timely and accurate information on the state of the local environment

in which any particular unit of inventory or processing asset resides. Moreover, the real-time

connectivity associated with the IoT means that this rich local information can be translated

into rich system-level information. In the food industry, IoT-enabled monitoring of temperature,

humidity, and other environmental conditions of perishable foods across the supply chain is seen

as an opportunity to improve buer inventory (Shacklett, 2017).

The sizing of inventory buers  although a dominant focus in the history of OM research

 is far from the only concern of an operations manager. The real-time execution, continu-

ous improvement, and upfront design of the operating system are equally important, and the

completeness, quality, and timeliness of information envisioned by the IoT have the potential

to dramatically impact these aspects. General Motors, for example, uses sensors to monitor

humidity to optimize painting; if conditions are unfavorable, the work piece is routed to another

1
For ease of exposition, we will use the term sensor to describe both the sensor and any associated controller

that drives action based on the sensor reading.

7
part of the plant (McKinsey, 2015a, p.68). GE sees the IoT as enabling real-time optimization

not only within a facility but also across networks of assets and facilities, allowing optimization

. . . to nd the most ecient system level solution (Evans and Annunziata, 2012, p.11). In agri-

culture, multiple sensors deployed across a eld enable precision-application of pesticides and

water based on the local conditions of individual parcels within the eld or even at the individual

plant level. Condition-based maintenance (CBM) of operational assets, whereby repair and re-

placement decisions are made based on current asset conditions rather than planned inspections

or unplanned outages, has received signicant practical and academic attention long pre-dating

the emergence of the IoT, e.g., Wijnmalen and Hontelez (1997). However, the ubiquitous and

connected sensors in the IoT  coupled with advanced predictive analytics  is making CBM a

key IoT application area (McKinsey, 2015a).

The recent emergence of digital twins oers a glimpse into the potential of IoT-enabled op-

erations. A digital twin is a sensor-enabled digital model of a physical object that simulates

the object in a live setting (Parrott and Warshaw, 2017). The crucial aspect here is the live

setting; that is, it is not simply a static digital representation of an object before use; it is a

dynamic representation of the object that evolves in use. Real-time sensor measurements across

a range of elements enable a system-level representation of a complex asset, e.g., a turbine, or in-

deed an entire production system. The physical world has a dynamically-coupled representation

in the digital world, and this data-rich environment can enable operations managers to uncover

previously-unknown relationships between system conditions and outcomes. This understand-

ing can drive continuous improvement in defect reduction, uptime, and other key performance

indicators of the existing system while also generating improved manufacturability-design ideas

for future products and factories.

Digital twins exist at the nexus of physical engineering, data science, and machine learning

(GE, 2016, p.2), and this holds true for IoT-enabled operations more broadly. Distilling the vast

data generated from sensors into actionable knowledge will be a formidable challenge. Real-

time, data-rich, system-level optimization within and across networks of factories and assets

will require algorithms that can solve extremely large problems almost instantaneously. Such

diculties oer ample research opportunities in the realm of data analytics and optimization

methods.

OM research has already extensively explored the value of inventory information (see the

RFID discussion above) and the value of forecast-information sharing in a supply chain, e.g.,

Lee et al. (2000) and Aviv (2002). Therefore, viewing IoT-enabled operations as information-

rich ones, one might conclude that the interesting research questions have already been explored.

8
This would be an unfortunate conclusion. IoT-enabled operations will deliver a much broader

array of production- and logistics-relevant data, opening up questions as to what types of data

are most useful, how should the data be best used, and how should data ownership in inter-rm

production systems (e.g., servitization of processing assets) be best congured? That is, the

IoT will lead to a wide variety of research questions related to the design of the processes and

procedures for managing operations and supply chains.

4 Blockchain
With extensive data now available through the IoT, there is a desire to store such data in an

accessible yet secure fashion. One possible solution to this storage problem is blockchain. While

blockchain technology has been surrounded by signicant hype, there do appear to be solid

operational use-cases for the technology, e.g., Babich and Hilary (2019).

A blockchain is a distributed and secure ledger. It is distributed in the sense that it can be

accessed and written-to from any (possibly authorized) location and its data is stored on a peer-

to-peer network (i.e., not in a central location). It is secure because once a block has been added

to the chain, it cannot be altered unilaterally. It is a ledger because it stores information. Many

people are familiar with digital currencies, such as bitcoin, which use blockchain technology.

However, it is blockchain's other applications that will likely be of more use to operations

managers.

Blockchain allows information to be kept on the entire history of a product as it travels along

the supply chain. While this can also be done through a central database, such as those provided

by enterprise software companies (e.g., SAP and Oracle), its distributed nature provides greater

exibility. With a blockchain, there is no need to allow third party access to commercially

sensitive systems. Further, anyone in the supply chain can upload information to the chain and

need not be directly connected to some centralized owner.

Although the information is secure once it is in the chain, there is of course no magic

guarantee that only factual information will be uploaded. Blockchain suers from the same

garbage-in-garbage-out danger of any information system. However, there is one important

dierence in this regard. Blockchain information can be entirely digitized. That is, information

can be uploaded automatically from IoT sensor data. So long as the sensors are accurate, so

will be the information in the blockchain.

To make this concrete, consider the following possible future for the kiwifruit supply chain

from New Zealand to China. Suppose each fruit is automatically tagged with a micro-RFID

tag upon picking. Every time the tag passes a reader, location and timestamp information

9
is uploaded to the chain. In addition, suppose that, once packed, a temperature monitor is

attached to the tray the kiwifruit sits in. Then, every time a reader is passed, e.g., when loading

and unloading from the shipping container, the full history of the temperature all of the fruit in

the tray have experienced is uploaded to each of the fruit's chains. A buyer in China might scan

the tag and then have access to a website that translates the blockchain information into useable

information. Where was the kiwifruit grown? How old is it? What temperature extremes has

it been exposed to? How likely is it to be at its optimal eating point?

All of the technology to make the scenario in the previous paragraph a reality currently

exists. The only stumbling blocks would be the insertion of the micro-RFID upon (manual)

picking and the reading of the chip by the Chinese consumer. However, currently fruit are

tagged at packing with individual stickers and pallets of kiwifruit are given RFID tags, so it is

not a very large jump to individualized RFID tags. Also, while scanning of tags might involve a

ramping up of phone technology, Chinese consumers are used to scanning Quick Response (QR)

codes (2D-barcodes) to learn more information about a product (or to pay for that product).

Tray-level QR codes linked to a blockchain could be implemented without delay.

The level of information described above is unprecedented. It would allow for product

recalls to be handled extremely eciently. It would also increase product (particularly food

and pharmaceutical) safety and decrease fraud. So long as eective methods could be devised

for inputting the information, it could signicantly increase the accountability of the supply

chain. In particular, it should allow visibility on whether ethical standards have been adhered

to in the production and distribution of the product.

Another important application for blockchain, beyond information storage, is the availabil-

ity of so-called smart contracts. These are contracts that are automatically triggered based

on some externally veried event. For example, payment could be authorized as soon as the

container reaches customs. More interestingly is if smart contracts are combined with detailed

supply chain information. For example, the kiwifruit transporter's payments could depend on

the maximum and minimum temperature experienced by the fruit during transit. While eective

contracting is a relatively mature area within OM, smart contracts would seem to oer a whole

range of coordination mechanisms not previously considered.

A nal important benet of blockchain is the elimination of intermediaries. Much of the

current role of distributors and freight-forwarders is information-based. With blockchain, many

of those roles can be eliminated. Further, if governments embrace the technology then customs'

processes may be able to be signicantly simplied. However, all of these eciency gains rely

on a level of standardization within the industry. If multiple competing blockchains develop

10
(as they likely will) and if they are made proprietary (as could easily happen), many of the

eciencies may be lost.

We see a number of research opportunities stemming from blockchain's likely impact on

the processes and procedures surrounding the supply chain. First, if blockchain fundamentally

changes the structure or the power relationships in supply chains, then these new congurations

will need to be analyzed. Second, as discussed earlier, smart contracts would seem to provide

a signicant opportunity for more eective and sophisticated coordinating contracts. Finally,

and as mentioned for the IoT, the quantity of data available through blockchain is a level of

magnitude not considered in most operations models.

5 Advanced Robotics
Manufacturing automation is not new and production-system robotics are already highly so-

phisticated. The concept of a lights-out factory in which processing is almost entirely carried

out by robots has been around for decades. Over fteen years ago, in one of Fanuc's 40,000-

square-foot factories near Mt. Fuji, robots [were] building other robots at a rate of about 50

per 24-hour shift [running] unsupervised for as long as 30 days at a time (Null and Cauleldt,

2003). Nevertheless, robotics technology is rapidly advancing in ways that may have profound

implications for factories, agricultural production, eld service management, and distribution

logistics.

Conventional robots in a factory reside within protective metal cages that ensure physical

separation from workers under normal operation. This is not only because the nature of the work

carried out by robots, welding, for example, is often inherently dangerous. It is also because we

have not been able to equip robots with sucient intelligence to dynamically adapt and adjust

to an ambiguous and rapidly-changing local environment. The motion of human workers is not

fully predictable and, therefore, separation is required to maintain a tightly-controlled operating

environment for the robot. Recent advances in sensor technology and articial intelligence are

enabling a new generation of robotic technologies that can be deployed alongside human workers.

These collaborative robots, often referred to as cobots, are being rolled out in real pro-

duction settings. BMW, for example, now has robots working alongside human workers on its

assembly line in Spartanburg, South Carolina, with the motive for one such deployment being

to automate a manual task that could cause repetitive strain injury. Interestingly, existing

industrial robots could perform this work, and do it much more quickly, but they could not

easily be slotted into a human production line because they are complicated to program and set

up, and they are dangerous to be around (Knight, 2014). Exoskeletons are also being used to

11
further blur the lines between human and robotic workers.

Ongoing advancement and development of collaborative or interactive robotics is, of course,

an active eld of research in computer science and engineering. Studying the adoption, accep-

tance, and application of cobots in production (and indeed service) settings would seem like a

natural line of inquiry for the OM eld, especially for those interested in behavioral operations

management. The vision  which is to some extent already a reality  of humans and robots

working alongside each other raises questions of how should such systems be designed and man-

aged. Answering those questions will require an understanding of how workers perceive and

interact with robots in repetitive processing settings.

Although perhaps less interesting from a novel OM research perspective, advancements in

robotics are also changing the economic rationale for investments in factory automation. Pur-

chasing and installing robots has long been expensive and time-consuming: robots deployed on

factory oors today require teams of specialists who have in-depth expertise for installing (the

robots as well as the safety systems), calibrating, and programming them to perform the man-

ufacturing tasks. Typically, this setup takes weeks and costs a multiple of the purchase price of

the robot itself  (Intel, 2015, p.1). Traditionally, automation of a processing activity was driven

by a desire for variable cost reduction (when labor is expensive), quality improvement (when a

robot can repeatedly execute the task with more precision than can a person), or safety (when

the processing environment is dangerous or the task physically onerous). The signicant upfront

cost and conguration time meant that robots were attractive for high-volume repetitive opera-

tions but were not attractive for low-volume operations or for operations requiring exibility in

terms of volume or mix of work (because robot capacity was xed and not easily recongured

or repurposed). Newer generations of robots, from ABB, iRobot, Motomon, and others, are

becoming cheaper, and perhaps more important, more easily congured for a range of dierent

tasks, thereby alleviating their conventional exibility disadvantage. The factory of the future

will look very dierent as cheap, congurable, and collaborative robotics become increasingly

available.

Robotics are also taking o in the production of agriculture. According to Jordan (2018)

about 60 percent of the romaine lettuce and half of all cabbage and celery produced by Taylor

Farms are harvested with automated systems. While automatic harvesting of more delicate

fruit, such as berries, is further o, advanced robotics are also being used to assess ripeness,

pack produce, and hoe weeds. The trade-os involved in such technologies are likely to be very

similar to those found on the factory oor.

The eects of advances in robotic technology are being felt outside of the farm and factory

12
as well. Great strides are being made in the automation of vehicles and the use of drones. Au-

tomated guided vehicles (AGVs) have been in use in factories for decades. However, sensors and

related technologies are now reaching the point that such vehicles can become autonomous and

freed from predened paths. For example, thousands of students at George Mason University

will have another dining option at their disposal: on-demand food delivery via an autonomous

robot on wheels (Holley, 2019). Autonomous vehicles are projected to have a large impact on

transportation networks and supply chains, e.g., Olsen and Parker (2019).

Drones, or unmanned aerial vehicles, are another robotic technology that is fast becoming

ubiquitous. They can be used to survey remote locations, deliver small items such as pharma-

ceuticals, monitor assets, and spray crops. Larger drones that carry passengers also appear to be

on the horizon (AssociatedPress, 2016). Drones are being used for lights-out inventory counting

and for inspecting oil pipelines, to name just two common operational tasks. They may lead to

new transportation optimization formulations, e.g., Agatz et al. (2018). The opportunities for

OM research seem similar to those for autonomous land vehicles.

In summary, the cost, exibility, speed, and/or quality of various robotic technologies are

improving, with collaborative robotics being a key trend. Further, robotic transportation options

have the potential to signicantly aect supply chain design.

6 Articial Intelligence
Like Industry 4.0 itself, Articial Intelligence (AI) is a term that does not have one precise

denition. Traditionally, AI has meant the mimicking of human intelligence using computers,

but recently the term has begun to encompass analytics and big data also. Indeed, it is often

used simply to indicate that a computer rather than a human is engaged in problem solving; a

perspective that readily ts the OM mindset. Borrowing a framework from analytics, we can

categorize AI industrial applications as being descriptive, predictive, or prescriptive.

Workers that inspect, pick, or sort products must continuously recognize dierent objects

and categorize them correctly. This is mentally fatiguing and errors can result in defects going

unnoticed or products being picked or placed incorrectly. AI that recognizes images or objects

can augment workers' categorization skills or alternatively enable robots to automate these

types of pattern recognition tasks. As discussed in Ÿ3, sensor-enabled operations will generate

vast amounts of data that can be mined using AI techniques to uncover previously unknown

relationships between processing conditions and outcomes, and this information can then be

used to improve process design and control. In agriculture, facial recognition is being developed

to allow farmers with large herds to know as much about the behavior of individual cows as

13
farmers with small herds do so that milk yield can be improved (Owen, 2018). Such applications

are descriptive in the sense that AI is describing an existing state (of an object or processing

relationships) and not necessarily predicting future outcomes or prescribing actions.

Of course, once cause-and-eect relationships are at least partially understood, then benets

can be gained by predicting future outcomes and prescribing current actions based on those

predictions. Predictive AI has become a key focus of durable asset manufacturers such as GE

and Siemens because condition based maintenance is dependent on accurate estimates of future

failure times. This requires sensors that deliver continuous data to a computer but also advanced

algorithms that predict remaining life based on current and past sensor readings. Prescriptive AI,

whereby operational actions (or at least recommendations for actions) are generated by software

is not especially new; manufacturing and supply chain decision-support systems (DSSs) have

existed for decades. What is new is the increasing adoption of machine learning techniques to

generate these prescriptions. (Advances in machine learning also underpin the increasing use of

descriptive and predictive AI.)

Whereas prescriptive OM models traditionally set up the objective and solved for the optimal

decision using some appropriate algorithm, with machine learning the computer uses training

data and statistical techniques to learn how to make good decisions without relying on a specic

underlying model. There are also hybrids to this approach where the algorithms are given some

guidance but not the entire model; see for example the dual sourcing application in Gijsbrechts

et al. (2018). Because machine learning in OM is a focus of the Perakis and Misic article in

this same issue (and also discussed somewhat by Song et al. in this issue), we do not elaborate

further here and refer the interested reader to those articles.

AI also goes hand-in-hand with advanced robotics. According to Roose (2019), while In

public, many executives wring their hands over the negative consequences that articial intel-

ligence and automation could have for workers . . . in private settings, . . . , these executives

tell a dierent story: They are racing to automate their own work forces to stay ahead of the

competition, with little regard for the impact on workers. Indeed the founder of the Chinese

e-commerce company JD.com has been quoted as saying I hope my company would be 100

percent automation someday. As noted earlier, human workers have traditionally been more

exible but slower than robots, but with advanced robotics and smart AI, that trade-o appears

to be changing.

From an OM research perspective, AI presents ample opportunities for algorithmic devel-

opment, either in general-purpose methods or in specialized applications. The complexity and

scale of the problem and the speed of solution are likely to be distinguishing features of future

14
work. From a behavioral perspective, just as with robotics, AI introduces an interaction between

humans and technology. In particular, workers will become more reliant on AI-generated recom-

mendations and this raises interesting questions as to how best to manage this worker-machine

interface. There is an emerging literature in organizational behavior on algorithm aversion (Di-

etvorst et al., 2015, 2016), and one in OM studying the interaction between human workers and

DSSs. For example, both Caro and Saez de Tehada (2018) and Sun et al. (2019) show that

adherence to such DSSs can be low but that it can be improved with changes to either how

it is presented or to the algorithmic prescriptions. In general, we would expect AI to have a

signicant impact on a large number of operations and supply chain processes.

7 Other Technologies
Although we have focused on the core technologies of AM, IoT, blockchain, advanced robotics,

and articial intelligence, other emerging technologies are also relevant to Industry 4.0.

Augmented reality (AR)  in which the physical world is enhanced with digitally-generated

visual or other sensory information  has applications not just in entertainment but also in

industry. GE is experimenting with AR as a means to drive productivity and quality by pro-

jecting the work instructions onto the parts and [using] sensors to monitor the assembly and

give feedback to the operator ... The system signals the operator immediately if an error oc-

curs or guides them to the next step Kellner (2018). Anecdotal case of AR use by Boeing,

GE and several other rms show an average productivity improvement of 32% (Abraham and

Annunziata, 2017).

In a similar vein, although Google Glass was not a consumer success, wearable AR glasses

have found applications in manufacturing: with Google Glass, [the worker] scans the serial

number on the part she's working on. This brings up manuals, photos or videos she may need.

She can tap the side of headset or say OK Glass and use voice commands to leave notes for

the next shift worker (Shamma, 2017). Other applications of industrial wearable technology

include devices aimed at reducing posture-and-movement induced injuries and those aimed at

tracking worker exposure to hazardous materials.

Along with progress in sensor technology and AI, innovations in material science (such as the

thin, strong, transparent and highly conductive material graphene and smart materials that

can adjust their properties in a controlled fashion in response to external stimuli) are likely to

drive advances not only in wearable devices but also in smart products that have embedded

intelligence that can be exploited during both production and use.

The term Agriculture 4.0 is sometimes used as an analog to Industry 4.0, e.g., De Clercq

15
et al. (2018). However, there is another dimension to Agriculture 4.0, namely agricultural-specic

technologies and high-tech foods . For example, vertical farming and genetic engineering are two

key technologies in this sphere. In addition, lab manufactured and bio-printed animal and plant

proteins are growing in popularity. Venture capital ooding into the agtech sector (protein, food,

and seed and crop technology) was US$25 billion in 2015 and the Agricultural Development Bank

of China has allocated at least 3 trillion yuan (US$435 billion) by 2020 to the modernization of

China's agriculture industry (Bosworth, 2016).

8 Conclusions
One hierarchical view of operations and supply chain strategy is that a natural tension exists

between the competitive priorities of cost, exibility, speed, and quality; and whilst all are

important, a rm cannot be best-in-class on all four priorities and must therefore choose between

them. Overall business strategy (e.g., which customers to target and what do they value?)

helps the operations function make this choice. The ranking among priorities then drives both

the design of the tangible architecture of operations (i.e, the degree of vertical integration,

the capacity and type of production, transportation, and MRO assets, inventories, and the

geographic distribution of these various elements) and the design of the intangible processes

and procedures that orchestrate the ow of goods and information through this operations and

supply chain architecture.

As we have already discussed, the technologies underpinning Industry 4.0 can, either in iso-

lation or in combination, improve one or more of the four priorities. More intriguingly, Industry

4.0 may alleviate some of the tension between the priorities and thereby enable new customer

value propositions. Crucially, Industry 4.0 creates new design possibilities for the operations ar-

chitecture and its associated processes. Certain technologies, the IoT for example, will directly

impact the processes for managing the operations and supply chain architecture and thereby

indirectly inuence architectural choices. Others may directly inuence the architecture but the

processes only indirectly. Figure 1 categorizes technologies by their potential for direct impact

on architecture and processes. However, as reected in the gure, the key to Industry 4.0 is that

it is not an individual technology. Instead, it is a synthesis of these related technologies. We

can therefore anticipate that Industry 4.0 will have a profound impact on both the architecture

and process dimensions of operations strategy.

AM and cheap/congurable robotics reduce the minimum ecient scale required for eco-

nomic production and promote exibility and speed of production-launch. Cobots, AI, wearable

devices, and AR can elevate (or eliminate) the role of human assets. The IoT, blockchain, and

16
 Direct Impact on Ops and Supply Chain Processes
Artificial Intelligence
Industry 4.0
High Blockchain The set of these
and other enabling
technologies
Internet of Things

Additive
Manufacturing
Low
Advanced Robotics

Low High
Direct Impact on Ops and Supply Chain Architecture

Figure 1: Potential Impact on Operations Strategy

digitization enable remote and real-time monitoring, diagnosis, control, and optimization within

factories and across geographically dispersed assets. A globally distributed and agile network

of production assets that can dynamically and rapidly adjust and re-allocate activities becomes

possible.

It is promises such as these that make many believe the marriage of the physical and digital

worlds that is Industry 4.0 will have a revolutionary impact on operations and supply chain

management. The OM community should engage with this vision to both understand the

implications and limitations of Industry 4.0 and to help develop the concepts and techniques

that will further drive its potential and adoption. We hope that this article will provide useful

as OM researchers look for opportunities and challenges in this space.

References
Abraham, Magid, Marco Annunziata. 2017. Augmented reality is already improving worker
performance. Harvard Business Review .
Agatz, Niels, Paul Bouman, Marie Schmidt. 2018. Optimization approaches for the traveling
salesman problem with drone. Transportation Science 52(4) 965981.
AssociatedPress. 2016. First passenger drone makes its debut at CES.
Available https://www.theguardian.com/technology/2016/jan/07/
at
first-passenger-drone-makes-world-debut.
Aviv, Yossi. 2002. Gaining benets from joint forecasting and replenishment processes: The case
of auto-correlated demand. Manufacturing & Service Operations Management 4(1) 5574.
Babich, Volodymyr, Gilles Hilary. 2019. Distributed ledgers and operations: What operations

17
 management researchers should know about blockchain technology. Manufacturing and Service
Operations Management .
Blanding, Michael. 2018. Printable future. Tuck School of Business News (October 15). Available
at http://www.tuck.dartmouth.edu/news/articles/printable-future.
Bosworth, Rosie. 2016. In lament of the NZ farm. AgTech Available at https://pureadvantage.
org/news/2016/11/29/lament-nz-farm/.
Caro, Felipe, Anna Saez de Tehada. 2018. Believing in analytics: Managers' adherence to price
recommendations from a DSS. Working paper .
De Clercq, Matthieu, Anshu Vats, Alvaro Biel. 2018. Agriculture 4.0  the fu-
ture of farming technology. Oliver Wyman (February). Available at https:
//www.oliverwyman.com/content/dam/oliver-wyman/v2/publications/2018/February/
Oliver-Wyman-Agriculture-4.0.pdf.
DeHoratius, Nicole, Adam J Mersereau, Linus Schrage. 2008. Retail inventory management when
records are inaccurate. Manufacturing & Service Operations Management 10(2) 257277.
Dietvorst, Berkeley J, Joseph P Simmons, Cade Massey. 2015. Algorithm aversion: People erro-
neously avoid algorithms after seeing them err. Journal of Experimental Psychology: General
144(1) 114.
Dietvorst, Berkeley J, Joseph P Simmons, Cade Massey. 2016. Overcoming algorithm aversion:
People will use imperfect algorithms if they can (even slightly) modify them. Management
Science 64(3) 11551170.
Dong, Lingxiu, Duo Shi, Fuqiang Zhang. 2016. 3d printing vs. traditional exible technology:
Implications for manufacturing strategy. Working Paper .
Evans, Peter, Marco Annunziata. 2012. Industrial internet: Pushing the boundaries of minds
and machines. General Electric Report .
Gaukler, Gary M, Özalp Özer, Warren H Hausman. 2008. Order progress information: Improved
dynamic emergency ordering policies. Production and Operations Management 17(6) 599613.
GE. 2016. The digital twin: Compressing time-to-value for digital industrial companies.
Available https://www.ge.com/digital/sites/default/files/download_assets/
at
The-Digital-Twin_Compressing-Time-to-Value-for-Digital-Industrial-Companies.
pdf.
Gijsbrechts, Joren, Robert N Boute, Jan A Van Mieghem, Dennis Zhang. 2018. Can deep
reinforcement learning improve inventory management? performance and implementation of
dual sourcing-mode problems. Working Paper .
Heese, H Sebastian. 2007. Inventory record inaccuracy, double marginalization, and rd adop-
tion. Production and Operations Management 16(5) 542553.
Holdowsky, Jonathan, Monika Mahto, Michael Raynor, Mark Cotteleer. 2015. Inside the internet
of things: A primer on the technologies building the IoT. Deloitte University Press .
Holley, Peter. 2019. George Mason students have a new dining option: Food delivered by robots.
Washington Post (Jan 22).

18
HP. 2018. HP launches world's most advanced metals 3D printing technol-
ogy for mass production to accelerate 4th industrial revolution. HP Press
Release Available https://press.ext.hp.com/us/en/press-releases/2018/
at
hp-launches-worlds-most-advanced-metals-3d-printing-technology.html.
Intel. 2015. Simplifying the design process brings more robots to manufactur-
ing. https://www.intel.com/content/dam/www/public/us/en/documents/
Available at
solution-briefs/industrial-robotics-for-manufacturing-brief.pdf.
Jordan, Miriam. 2018. As immigrant farmworkers become more scarce, robots replace humans.
The New York Times (Nov 20).
Kellner, Tomas. 2017. An epiphany of disruption: Ge additive chief explains how 3d print-
GE Reports Available at https://www.ge.com/reports/
ing will upend manufacturing.
epiphany-disruption-ge-additive-chief-explains-3d-printing-will-upend-manufacturing/.
Kellner, Tomas. 2018. Game on: Augmented reality is helping factory workers
become more productive. GE Reports
https://www.ge.com/reports/
Available at
game-augmented-reality-helping-factory-workers-become-productive/.
Knight, Will. 2014. How human-robot teamwork will upend manufacturing. MIT Technology
Review (Nov-Dec).
Knous, Nils, Matthijs C van der Heijden, Willem HM Zijm. 2019. Consolidating spare parts
for asset maintenance with additive manufacturing. International Journal of Production Eco-
nomics 208 269280.
Lee, Hau, Özalp Özer. 2007. Unlocking the value of rd. Production and operations management
16(1) 4064.
Lee, Hau L, Kut C So, Christopher S Tang. 2000. The value of information sharing in a two-level
supply chain. Management science 46(5) 626643.
McKinsey. 2015a. The internet of things: Mapping the value beyond the hype. McKinsey Global
Institute .
McKinsey. 2015b. Manufacturing's next act. McKinsey & Company Report .
Müller, Bernt. 2016. How power plants can benet from 3d printed parts. Avail-
able at https://www.siemens.com/innovation/en/home/pictures-of-the-future/
industry-and-automation/additive-manufacturing-3d-spare-parts-gas-turbines.
html.
Null, Christopher, Brian Cauleldt. 2003. Fade to black the 1980s vision of lights-out" manu-
http://archive.
facturing, where robots do all the work, is a dream no more. Available at
fortune.com/magazines/business2/business2_archive/2003/06/01/343371/index.htm.
Olsen, Tava, Rodney Parker. 2019. Autonomous vehicles: A research agenda for operations
management. Working Paper .
Owen, David. 2018. Should we be worried about computerized facial recognition. The
New Yorker (Dec 17). https://www.newyorker.com/magazine/2018/12/17/
Available at
should-we-be-worried-about-computerized-facial-recognition.
Parrott, Aaron, Lane Warshaw. 2017. Industry 4.0 and the digital twin: Manufacturing meets
its match. Deloitte University Press .

19
Roose, Kevin. 2019. The hidden automation agenda of the davos elite. The New York Times
(Jan 25).

Saghaan, Soroush, Brian Tomlin, Stephan Biller. 2018. The internet of things and information
fusion: Who talks to who? Working Paper .
SCDigest. 2018. Supply chain news: Hp claims to have upped the ante with new 3d printer
that can use metals. Available at http://www.scdigest.com/ontarget/18-09-11-1.php?
cid=14661.
Sethuraman, Nagarajan, Ali Parlakturk, Jayashankar Swaminathan. 2018. Personal fabrication
as an operational strategy: Value of delegating production to customer. Working Paper .
Shacklett, Molly. 2017. How the internet of things will transform food supply chains. Food
Logistics
(Dec 18). Available at https://www.foodlogistics.com/technology/article/
12382105/how-the-internet-of-things-will-transform-food-supply-chains.
Shamma, Tasnim. 2017. Google glass didn't disappear. you can nd it on the factory oor. NPR
https://www.npr.org/sections/alltechconsidered/2017/03/18/
(Mar 18). Available at
514299682/google-glass-didnt-disappear-you-can-find-it-on-the-factory-floor.
Song, Jing-Sheng Jeannette, Yue Zhang. 2018. Stock or print? impact of 3D printing on spare
parts logistics. Working Paper .
Sun, Jiankun, Dennis J Zhang, Haoyuan Hu, Jan A Van Mieghem. 2019. Predicting human dis-
cretion and adjusting algorithmic prescriptions: A large-scale eld experiment in bin-packing.
Working Paper .
Westerweel, Bram, Rob Basten, Jelmar den Boer, Geert-Jan van Houtum. 2018. Printing spare
parts at remote locations: Fullling the promise of additive manufacturing. Working Paper .
Wijnmalen, Diederik JD, Jan AM Hontelez. 1997. Coordinated condition-based repair strategies
for components of a multi-component maintenance system with discounts. European Journal
of Operational Research 98(1) 5263.

20