Hive Ventures - State of Taiwan Enterprise AI 2021 - EN
Hive Ventures - State of Taiwan Enterprise AI 2021 - EN
Hive Ventures - State of Taiwan Enterprise AI 2021 - EN
HIVE VENTURES
Taiwan was ranked 4th on Innovation Capability by the World Economic
Forum (WEF) World Competitiveness Report in 2020 and is recognized
in the world for its industry leading developments in the semiconductor,
electronics, and hardware manufacturing sectors.
But, as the paradigm continues to shift towards a software driven world,
powered by machine intelligence, where does Taiwan stand in this new
wave?
Through our engagements with start-ups and grown-ups in Taiwan and
around the world, in our current capacity and previous ones, we’ve
seen firsthand how economies have ridden tides during these
paradigm-shifting times, coming from behind and leap-frogging ahead
of the race. Will Taiwan be able to ride this wave? Are Taiwan’s
enterprises ready to ride this tsunami, to enlarge the gap ahead of their
competitors?
One of the key building blocks for Taiwan’s early economic success in
the semiconductor industry and the previous Made In Taiwan (MIT) era,
is Taiwan’s strong and steady supply of technological talent. Will this
talent pool cascade down into a vibrant AI powered startup ecosystem?
Can Taiwan incubate and become a hotbed for AI startups?
We hope this survey, and more to come in the coming years, will serve
to help answer these questions, helping start-ups, scale-ups, and grown
ups alike to understand where they stand, and how they can progress
to the next level to harness this wave.
1
COMMENTARY RICHIE TSAI
Acting CEO,
TAIWAN AI ACADEMY Taiwan AI Academy
Since the epidemic, we’ve accelerated into a new era of The third wave will be in upgrading at the industry level.
AI. Traditional industries will begin to catch up, and AI
adoption by “invisible champions” within mature and
In the next decade after the epidemic, we will see four special sectors will move become an industry wide
waves of opportunities for AI to enhance transformation movement. The breadth and effectiveness of
of enterprises, powered by different "domain knowledge applications will be greatly enhanced, and many will
x information maturity" matrices: start maturing into standardized products.
The first wave will be in single instance efficiency The fourth wave will be in the overall improvement of
improvements, such as defect detection, intelligent the ecosystem. Industry clusters will proliferate and
scheduling, predictive maintenance, and optimization of diversify. Qualitative changes will materialize at scale.
raw material mix at the manufacturing end, which will New applications will be built upon the existing
mean the replace of humans by machines. We have foundations, and the exponential efficiency
seen rapid progress in specific applications by start-ups, improvements will further drive industry transformations.
such as voice assistants, facial recognition, medical Industries will be restructured, the global division of
imaging interpretation, etc. labor will reshuffle.
The second wave will be the enhancement of enterprise The Taiwan AI Academy has been promoting industrial
processes. When all single-instance applications are AI since 2018 and is now in its fourth year. It has been
popularized, the efficiency of the reorganization process TAIWAN’s vision to let the world think of AI when they
will be improved, and changes in the line of production, see TAIWAN, and to gasp “WA” when they see
service procedures and human-machine collaboration TAIWAN's achievements in AI. We are pleased to work
will bring about the reorganization and restructuring of with Hive Ventures and several outstanding start-up
the management process. companies to produce this report: STATE OF TAIWAN
ENTERPRISE AI 2021, especially in an
internationalized English format, so that together we
can make Taiwan's soft power in Artificial Intelligence
shine upon the global stage.
2
TABLE OF CONTENTS
5 OVERVIEW 14 METHODOLOGY
3
AI has been the raging buzzword globally and across industries over
the past few years. From governments to small medium businesses,
from North America to Australia, organizations have been pushing
towards digital transformation and even the adoption of some form of
artificial intelligence for the purpose of automation and deeper
applications.
However, whilst many have yet to take that bold leap, many others
have tried, paused, become disillusioned, and only few have managed
to cross that chasm.
“Where exactly am I in that cycle? What have others who have
succeeded done? What do I need in order to make this a successful
conquest? Where do I go next?” These questions arise in every board
room and meeting we’ve had with enterprises, both large and small, in
4
• CLOSE TO 25% OF SURVEYED COMPANIES HAVE ENTERED
DEPLOYMENT OF AI/ML
According to this survey, most companies have already entered the
development and deployment phase of introducing AI/ML into their
enterprise. Additionally, 24.8% of respondents have already
entered Level 4 and beyond on our Maturity Cycle.
• THE “COST DOWN ECONOMY” LOOKS BEYOND ADOPTING
AI FOR “COST OPTIMIZATION”
Taiwan is renowned for maintaining cost leadership in our contract
manufacturing industry globally. It would have been safe to assume
that most Taiwanese enterprises are adopting AI in pursuit of
maintaining their cost leadership position, further streamlining
costs in production. However, of all respondents, “Improving
Organizational Efficiency” is considered the core objective they
wish to achieve when adopting AI, whereas “Cost Cutting” is
considered the least important objective.
• SIZE MATTERS, BUT AGILITY MATTERS MORE IN
ENTERPRISE AI ADOPTION
KEY FINDINGS
SUMMARY
Companies falling into the mid-large enterprise segment (between
200-500 employees) have the resources and ambition to adopt
enterprise AI with more agility. They are poised to lead the new era
if they continue the momentum.
• HIGH MODEL DEPLOYMENT RATES CHARACTERISTIC OF
TAIWAN’S CULTURE
Whilst globally, only 13% of models developed eventually enter
production, Taiwan’s enterprises has exhibited over 20% (and even
50% in certain categories) model deployment rates, fully
characteristic of Taiwan’s history and culture of pursing efficiency
and high ROI.
5
OVERVIEW
6
LEVEL OF URGENCY TO ADOPT AI
OVERVIEW OF
35.0%
30.0%
29.6% ENTERPRISE ADOPTION
25.0%
The notion of enterprise AI adoption has become
20.0% widely popularized thanks to the incessant
14.1% 14.1%
15.0% promotion by the government, industry think tanks,
10.2%
10.0% 7.8%
9.2% and media. Most people already understand that
3.4%
4.9% adopting solutions powered by artificial intelligence
5.0% 2.9%
1.9% 2.4%
is important and urgent for the organization.
0.0%
0 1 2 3 4 5 6 7 8 9 10 In our survey, while 52.5% of respondents already
Not Urgent Extremely Urgent believe it is relatively urgent, 29.6% are still sitting
on the fence. In our past interviews and
YEARS OF EXPERIENCE ADOPTING AI conversations with business owners, most of these
fence sitters are unfamiliar with which facet of their
35.00% 33.33%
businesses should be and could be transformed,
30.00% and the actual impact that can be brought forth with
the transformation.
25.00%
0.00%
1.15%
For those who have taken that leap, adoption
0.00%
0 1 2 3 4 5 6 7 8 9 10 11
became a wave circa 2015-2016, with a strong
No. of years since AI adoption wave culminating in 2017-2018; Partially due to the
establishment of government initiatives pushing
towards Industry 4.0.
7
URGENCY OF AI ADOPTION
THE BIGGER THEY ARE, THE MORE URGENT IT While it seems that the majority in Manufacturing
SEEMS and Healthcare are cold towards the urgency of AI
adoption, part of these responses are due to the fact
Overall, size does matter when it comes to AI adoption that they've already begun adopting, and are
urgency. considered advanced in their adoption, hence the
lesser urgency.
Across the respondents, larger companies seem more
eager to transform. This is due in part because of the
available resources at hand, and the ever higher need to
maintain continuous momentum to drive their business
forward.
B2B VS. B2C: EAGERNESS DISPARITY APPARENT URGENCY: BY COMPANY SIZE
Retail and FMCG lead the pack in terms of their 100%
8
ADOPTION IN VARIOUS
FUNCTIONAL UNITS
SALES AND MARKETING LEADS ADOPTION
Sales and Marketing represents one of the most
developed function that has applied AI in part of their
operations. Applications and services that offer intelligent
customer data segmentation, personalized or
REVENUE CENTERS LEAD ADOPTION
programmatic advertising, and customer service chatbots
have made it simple and alluring for the sales and
Manufacturing 48.55% marketing teams to adopt these applications in the form
Sales & Marketing 47.40%
of Software-as-a-Service subscriptions.
User 31.21% Globally, the same trend exists, as the Sales and
Marketing function is generally the most outward facing
Security 28.32% business function, and hence, has the most liberty to
Backoffice Ops 25.43%
engage external solutions in their operations.
Retail 17.34% In our opinion, while there are already a myriad of
MarTech and SalesTech companies in the region offering
Logistics & WMS 12.14% simple integration and analytics, more savvy users are
0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% already seeking more sophisticated solutions that are
able to extend the application services across more
business functions and data silos, enabling better
decisional intelligence across the organization, instead of
single point solutions solely for the sales and marketing
function.
9
FUNCTION UNIT ADOPTION
BY INDUSTRY
MANUFACTURING AND USER FOCUSED customer experience personalization, and digital
APPLICATIONS PERVASIVE ACROSS VARIOUS assistants are also on the rise and widely adopted
INDUSTRIES across industries for both internal and external
users.
While Sales and Marketing is generally regarded as
more critical by more consumer facing industries, AI
for manufacturing efficiency is pervasive across all
industries.
Thanks to a strong talent base and increasingly FUNCTIONS ADOPTED BY INDUSTRY
vibrant startup community, advanced optical
inspection (AOI), predictive maintenance, and even
machine learning for production parameter
optimization and management have become
commonplace.
These solutions are not exclusive for the
Manufacturing and Oil and Gas industry only, but
also widely adopted in the Healthcare and Telco
sectors as well.
Also, as enterprises raise their awareness on
User/Customer experience (UX/CX), intelligent
solutions geared towards users, including
automated user identification, facial recognition,
10
THE ENTERPRISE AI
MATURITY CYCLE
11
LEVEL 0 LEVEL 1 LEVEL 2 LEVEL 3 LEVEL 4 LEVEL 5 LEVEL 6
Based on the survey, we’ve scored all respondent companies according to their data, team, and operational readiness
to derive their overall maturity levels.
Across various industries, its encouraging to see that the Banking and Finance industry is skewed towards a higher
level of maturity as a whole, whereas the Manufacturing sectors, which represents a core economic driver behind
Taiwan’s economy, see a more even distribution with varying degrees of maturity. Of course, if we look at the absolute
leaders, the high-tech manufacturing sector occupies the majority of Level 4 and 5 companies, leading the overall
cohort when it comes to representation in the most mature classes, whereas the Retail sector, another major contributor
to Taiwan’s economy, trails behind the rest of the pack.
13
MATURITY DISTRIBUTION: OVERVIEW
MANUFACTURING AND USER FOCUSED THE ENTERPRISE AI MATURITY CYCLE DISTRIBUTION
APPLICATIONS PERVASIVE ACROSS 40%
VARIOUS INDUSTRIES 35%
34.29%
30%
20%
10%
0%
Banking & Manufacturing Oil, Gas & Healthcare Internet & Media
Financial Services Chemicals
Maturity Level 0 1 2 3 4 5
14
MATURITY DISTRIBUTION: DOES SIZE MATTER
15
OBJECTIVES FOR ADOPTING AI
16
DATA
READINESS
17
DATA READINESS
DATA ACCESSIBILITY
ACCESS TO QUALITY DATA ACCESS TO QUALITY DATA: BY INDUSTRY
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
YES NO
18
DATA READINESS
CONTINUOUS OPTIMIZATION
CHALLENGES IN MODEL TESTING CHALLENGES IN MODEL SCALING: BY INDUSTRY
26.9
39.1 %
%
34.0
%
We see a continuous trend towards solutions that can work at the heart of manufacturing execution systems
(MES), bringing operational technology (OT) with information technology (IT) together in tandem, channeling
data from the front lines of execution to the back office war room seamlessly. While many of these solutions are
still focused on single pain points, solutions aiming to be the "brains of the operations" are emerging and battling
for dominance from big tech and startups alike.
Nonetheless, the desire and expectations for high quality data is ever changing. As enterprises and teams
continue to advance in their quest for faster, deeper insights and more advanced AI applications in production,
there will be a continual need for data access optimizations. Companies who are more advanced in its adoption
journey will continue to see ever higher needs, and should continue to optimize its infrastructure to meet the
needs of the enterprise and market alike.
19
DATA READINESS
DATA INFRASTRUCTURE
20
DATA READINESS
VARYING INDUSTRY LIMITATIONS
21
DATA READINESS
TOOLS & SERVICES
data streams and requirements their business is 0% 20% 40% 60% 80% 100%
facing without upgrading their infrastructure.
22
DATA READINESS
TRENDS: DATA TOOLS & SERVICES
CannerFlow 23.95%
RStudio 32.84%
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
Data preparation and data virtualization tools that enhance data transformation, processing, and enrichment have
also picked up its pace in its presence in the data infrastructure.
These tools enable better integration, flexibility, and analysis of large datasets when working with advanced
analytics solutions and machine learning applications. Its flexible nature also allows enterprises to adjust itself
quickly to changing business and data environments without large scale infrastructure adjustments and investments.
Data preparation tool wise, tools with open source versions, Rstudio and KNIME, are the most widely adopted,
whereas enterprise user driven Rapidminer and Alteryx are also represented amongst the top solutions.
Data virtualization is relatively new, and enterprises have been experimenting between overseas offerings such as
Denodo and AtScale, as well as local player CannerFlow*, in search of solutions balancing performance, value, and
support.
BI tools
Data Lakes
Data Marts
Data Warehouses
Data Pipeline
Data Preparation
Data Visualization
ETL Tools
Operational Data Store (ODS)
24
TEAM
READINESS
25
TEAM READINESS
AI LEADERSHIP: AN EVOLVING TEAM GAME
26
TEAM READINESS
AI LEADERSHIP: AN EVOLVING TEAM GAME
MOVING TOWARDS AI DEMOCRATIZATION (Level 6)
As observed in the composition of Level 5 stakeholders, we see a clear trend, as organizations move towards a
culture where AI is already in its DNA.
There is no longer a need for a solitary leader to lead an AI initiative, characterized by a lower CEO dominance,
lower Dedicated Project Office presence, and even a higher percentage of organizations without a dedicated AI
leader.
In these organizations, AI is becoming a part of everyday life, with no need to emphasize the "introduction" of AI,
but a return to pure business operationalization.
AI LEADERSHIP: BY INDUSTRY
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
CEO CTO CDO (Chief Digital Officer) CIO CMO Project Office Others IT Team None
27
TEAM READINESS
DATA SCIENTISTS: THE ARSENAL
NO. OF DATA SCIENTISTS IN ENTERPRISE: US DATA SCIENTISTS IN ENTERPRISE: 2018 vs. 2020
US VS. TAIWAN
70% 63% 70%
58%
60% 60%
50% 50%
50% 50%
40% 40%
30% 30%
20% 20%
17% 14% 20% 14%
20%
8% 9% 7%
10% 5% 6% 10% 2% 5%
0% 0%
1-10 11-50 51-100 >100 1-10 11-50 51-100 >100
29
TEAM READINESS
DATA SCIENTSTS: EXPERIENCE LEVELS
30
TEAM READINESS
THE DIGITAL TALENT SKILLS GAP
AVERAGE DAYS OF RESKILLING NEEDED, 2018-2022 The World Economic Forum forecasts that by 2022,
54% of the world's workforce will require re-skilling
and up-skilling in order to keep up with the
dynamic digital nature of their existing roles.
Asia represents economies with the highest need
and willingness of investing into re-skilling in the
world. 6 of the top 10 countries are Asian, and
centered around North and South East Asia.
As businesses continue to transform, new
technologies are required to be learned to improve
job quality and efficiencies. The technological
change is also changing the demand for skills at a
faster pace than before. Hence, driving a strong
need for up-skilling and re-skilling of the work force,
especially in developing economies.
Taiwan's strong supply of technological talent,
supported by machine learning education at all
levels, from secondary to tertiary education and
vocational training institutes such as the AI
Academy and ALPHACamp, positions Taiwan
favorably to offer a technically-ready talent pool to
Source: “Future of Jobs Survey”, 2018, WEF
fill the gap across the region.
31
TEAM READINESS
AN AI FRIENDLY CULTURE
ARE YOU AI FRIENDLY? INDUSTRIES WITH AI FRIENDLY CULTURES
YES! We are AI Friendly! NO! We're still building it. YES! We are AI Friendly! NO! We're still building it.
32
TEAM READINESS
A CROSS DISCIPLINARY COLLABORATION AI TEAM LINE-UP
When implementing AI within the enterprise, it is most effective to bring business operation leaders together with
data experts to work side by side, to ensure organization-wide issues are addressed instead of singular objectives,
and to predict the operational issues involved with every implementation decision. These issues might relate to
the IT infrastructure overhaul, but also impact fundamental manufacturing process and software development
processes.
The manufacturing management teams are already seen to play a prominent role in the Oil, Gas and Chemicals
sector as well as the Manufacturing sector, while the software development teams are the indispensable leaders
to experiment and implement these initiatives within the enterprise across all industries.
Involving operational stakeholders can also help build trust in the system and its recommendations, beyond
personal subjectivity and intuition from past experiences, and can also alleviate any anxiety of AI adoption
disrupting future work.
IT
Software Development Team
Dedicated Data Engineers
Dedicated Data Scientists
Manufacturing Manager
33
OPERATIONAL
READINESS
34
OPERATIONAL READINESS
TECH STACK ADOPTION
AN INTERCONNECTED FABRIC OF
TECH STACK ADOPTED TECHNOLOGIES
NLP/NLU
Computer Vision
AutoML
Machine Learning
Deep Learning
AI Acceleration
Deep Neural Networks/
Deep Learning
Data Labeling & Annotation
Services
Smart Robots/CoBots
Decision Intelligence
Robotic Process Automation
(RPA)
Digital Twin
Anomaly Detection
36
OPERATIONAL READINESS
TECH STACKS VS. ADOPTION OBJECTIVES
From the perspective of delivering organizational objectives, natural language processing (NLP) technologies
matched with machine learning are most widely adopted in enterprises seeking to increase revenue and raise
customer loyalty. These can stem from providing personalized user experiences (i.e. personalized marketing and
incentives) or in automated customer support and conversational commerce.
A noticeable trend is towards the rising adoption of automated machine learning (AutoML) services to drive
increasing revenue and organizational efficiency. Global unicorns such as DataRobot and H20.ai, and the
recently public C3.ai, have been delivering these services to various industries. Home grown players such as
ProfetAI* have also helped leading manufacturers in SMT parameter optimization and drive increased efficiencies
and revenues.
38
OPERATIONAL READINESS
FROM LAB TO PRODUCTION: DEPLOYMENT
39
OPERATIONAL READINESS
MODEL DEPLOYMENT: TAIWAN VS. US
Breaking down the respondents into tiers vis-à-vis 30.00% 28% 28%
0.00%
Yet, Taiwan’s ecosystem shows a bulk mid 0-7 Days 8-30 Days 1-3 Months 3-12 Months >1 Year Don't
know/Unsure
performers segment, who can continuously Taiwan US
40
OPERATIONAL READINESS
HIGH DEVELOPMENT TO DEPLOYMENT RATES
MODEL DEPLOYMENT RATES: MEASUREMENT The move from the sandbox to production is challenging,
OF EFFICIENCY and we encourage companies to build out efficient
workflows and environments as they go, to further
When we compare models developed against accelerate deployment rates.
models actually deployed, it is encouraging to see
that most companies who have developed more
than one model have deployed at least part of
these models into actual production. Despite the
long and challenging deployment time, this signifies
an effectiveness in moving beyond the sandbox into
MODEL DEPLOYMENT RATES
real world environments.
100%
90%
According to various industry reports, globally, only 80%
less than 20% of models progress beyond model
Models Deployed
70%
60%
development and experimentation. Our survey 50%
shows a significantly higher deployment rate than 40%
30%
this global reference, with the majority of 20%
companies deploying over 20% to 50% of their 10%
0%
developed models. 1-3 4-10 11-50 51-100 101-1000 >1000
Models Developed
As companies grow through the maturity cycle, the
deployment rates (no. of models developed / no. of 0 1-3 4-10 11-50 51-100 101-1000 >1000
models deployed) increases as the experience
level rises and workflows are improved, thus
encouraging higher development velocities with the
aim of achieving even more business outcomes.
41
OPERATIONAL READINESS
IMPROVING WITH EXPERIENCE
DEPLOYMENT VS.
DEPLOYMENT SPEEDS BY INDUSTRY
ENTERPRISE AI EXPERIENCE
100%
90% Banking & Financial Services 銀行金融服務
Consulting 諮詢公司
No. of models deployed
80%
70% Healthcare 醫療保健
60% Internet & Media 網路 & 媒體
50% Manufacturing 製造
40% Marketing 行銷
30% Oil, Gas & Chemicals 石油 & 化工
20% Others 其他
10% Public Sector 公共部門
0% Retail 零售
1 2 3 4 5 6 7 8 10 Telecommunications 電信
No. of years AI introduced into enterprise
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
1-3 4-10 11-50 51-100 100-1000 >1000 0-7 days 8-14 days 15-30 days 1-3 months 3-6 months 6-12 months >1 Year Don't know/Unsure
42
OPERATIONAL READINESS
POST DEPLOYMENT CHALLENGES
LIFE AFTER DEPLOYMENT While the 5G era may solve connectivity issues,
other critical issues, such as edge model design
Model deployment is not the end of the road for the and acceleration still pose challenges, leading to
life cycle of a model. In reality, it is the beginning of heightened research in the field of high power and
a new life cycle for the developed model, as it takes low power inference engines to bring together the
in new data and learns even more in the real world. best of both worlds.
This inadvertently leads to deviation and
degradation of the model due to variances in the
data stream in actual production environments, data
entry errors, glitches in the database, and errors in
the data pipeline. MODEL MANAGEMENT CHALLENGES
43
CHALLENGES & CONCERNS OF
ENTERPRISE AI ADOPTION
44
OPERATIONAL READINESS
GROWING PAINS OF AI ADOPTION
CHALLENGES IN AI ADOPTION
Data Collection
Data Applicability
EARLY GROWING PAINS STILL A MAIN
Data Accessibility CHALLENGE
Business Case POC
Model Development for production
Scale Up 82% of Taiwan’s adoption woes are early stage
Duplicating organizational efforts growing pains, with the top concerns being in data
Organizational Alignment
collection, proving business case, and developing
Cross Programming Language
support useful models for production. In comparison, US
ML Versioning and reproducibility counterparts are more concerned with scaling up
Talent Supply
Model performance monitoring
their ML operations, versioning, and reproducibility
Data Scientists Performance of their models; yet approximately 69% of US
enterprises are still bothered by early stage
growing pains of proving business case and
90%
getting organizational buy-in.
82%
80%
70%
69% While the distribution of these challenges
60%
suggests that most of Taiwan’s enterprises are still
50%
in its nascent stages of AI and ML adoption, these
40% 31%
concerns are common growing pains shared by
30% enterprises all around world, and no less easy in
18%
20% more developed economies.
10%
0% We expect continued shifts from early stage woes
Early Growing Pains Operational Challenges towards more advanced operational stage
Taiwan US challenges in the years to come as companies
grow through the maturity cycle.
45
OPERATIONAL READINESS
GROWING THROUGH THE LIFECYCLE
Model performance monitoring and ML versioning issues rise as more models are developed and deployed, and the
teams get larger, with more projects and more specialized teams. As these teams start working in smaller pods,
platforms that allows for better collaboration becomes more critical to avoid duplicated efforts, creating different
versions of a model simultaneously, and other issues that arise from disconnected data resources.
Software platforms such as MLOps platforms have risen in popularity over the past 2 years to alleviate these issues.
Many companies currently use tools such as GitHub or MS Visual Studio, which are geared towards software
developers for these purposes. Open source tools such as Kubeflow or enterprise software such as Anaconda,
Algorithmia, and InfuseAI* are also picking up in popularity as more data scientist specific platforms are required to
bridge the divide between software developers and data scientists workflows.
Organizational Alignment
46
OPERATIONAL READINESS
CONCERNS OF AI ADOPTION
RISING COSTS
Adopting machine learning and deep learning doesn’t come cheap. And as more models are developed, and
consume more data in their self-learning process, they demand more GPUs and power to continue to learn and grow.
Despite cloud computing, costs continue to burgeon. As such, cost concerns come in at the top at 19.2% of our
respondents concerned about rising costs to keep the system running going forward.
TRUST THROUGH EXPLAINABILITY AND RESPONSIBILITY
As we advance towards autonomous systems, we begin to rely on the systems to make decisions on our behalf, that
can be explained and attributed to. Algorithms today work in black boxes, without its creators fully appreciating the
model’s rationale beyond pure math. 16.9% of respondents are concerned with the explainability of their AI, and
10.5% have concerns fully trusting their systems to make autonomous decisions on their behalf. This is a global
phenomenon, and data scientist teams around the world are now seeking solutions, in hopes of lifting this obstacle in
our next step towards full autonomy.
47
LOOKING TOWARDS THE
FUTURE
48
LOOKING FORWARD TO THE FUTURE
TRENDS IN APPLICATION ADOPTION
FUTURE APPLICATION ADOPTION SHIFTS TOWARDS THE BACKEND, AND TOWARDS PROLIFERATION
We observe that overall AI/SaaS adoption has started from more revenue generating functions, but will move towards
back-office functions in the coming years.
Foremost functions such as sales and marketing will stay largely as a staple for adoption in Taiwan. But
manufacturers and retailers alike will shift investments towards logistics and warehousing to speed up organizational
efficiency, including areas such as warehouse automation, inventory forecasting, and pick-up automation.
Buyers are also expected to ramp up on investments into security, especially in light of the heightened level of cyber
attacks over the past few years. Coupled with government incentives for small medium businesses and select
industries to invest in cybersecurity, this has become another hot sector for adoption in the coming years.
Retail
Logistics/Warehousing
Backoffice Operations
Users
Security
Manufacturing
Sales/Marketing
49
LOOKING FORWARD TO THE FUTURE
TRENDS IN TECHNOLOGY ADOPTION
FUTURE TECHNOLOGIES ADOPTION MOVING TOWARDS SPEED AND DEPTH
We see a shift in tastes towards deeper applications and faster experimentation.
Computer vision, which has been popular, especially in the fields of quality inspection and anomaly detection, will
slow down as most major players already have smooth operating solutions, and investments will come from late
adopters of these technologies, most likely in medium sized manufacturing and chemicals players.
Pureplay machine learning will slow down, especially as infrastructure and SOPs have been established.
Instead, adoption of AutoML for faster experimentation and solving many more business problems simultaneously
and driving decision intelligence will increase in spending.
Digital twin investments are expected to increase as we move beyond the innovators, towards the early adopters of
the technology.
Anomaly Detection
Digital Twin
Robotic Process Automation (RPA)
Decision Intelligence
Smart Robots/CoBots
Data Labelling & Annotation Services
Deep Neural Networks/Deep Learning
AI Acceleration (GPU/FPGA)
Deep Learning
Machine Learning
AutoML
Computer Vision
NLP/NLU
0 20 40 60 80 100 120 140
50
RECOMMENDATIONS TO SHIFT
THE MATURITY
51
RECOMMENDATIONS TO SHIFT THE MATURITY
THE ENTERPRISE VISION AND KEY OBJECTIVES
FOR ADOPTING AI
What do we as an enterprise want to achieve from
adopting AI? This should not be for the purpose of
SHIFTING TO LEVEL 1 adopting a fancy buzzword, but helping enterprises
achieve their overall organizational vision, and
ALIGNMENT
retention, and cost reductions in certain processes,
amongst many.
IDENTIFYING AND ALIGNING THE KEY
STAKEHOLDERS
Who will these new initiatives benefit and impact? Who
are the stakeholders that will best be involved to
LEVEL 1: ORGANIZATIONAL ALIGNMENT impact these changes? These should include top level
management down to key working level stakeholders,
As we act upon the urgency of AI adoption into in order to get organization-wide buy in and alignment
the enterprise, the leadership should consider for execution purposes.
and decide upon the following questions in order
to realize a successful kickoff. IDENTIFYING THE KEY OPPORTUNITIES AI CAN
OPTIMIZE
What are the pain points and friction of our target users
in their workflow? How can our automation and AI
initiatives help with that? A good vision and objective
will be rooted in real world operational environments
and will ensure the initiatives developed will see
smooth deployment and actual business impact.
52
RECOMMENDATIONS TO SHIFT THE MATURITY
IDENTIFY HIGH VALUE DATA
Experts advise that a data audit be conducted to
identify which data points are most relevant to tackle
the identified challenges and opportunities. Very often,
SHIFTING TO LEVEL 2 some required data sets might not be readily available,
so teams may need to devise ways and retool
ACCESSIBILITY
BUILD UP THE DATA INFRASTRUCTURE TO
COLLECT THE RIGHT DATA FROM THE RIGHT
PLACES AND DELIVERED TO THE RIGHT PEOPLE
For many large enterprises, data collection has already
been at the heart of their operations. However, these
legacy systems might not be keeping up with the latest
data streams and data structures, or are stored in
LEVEL 2: TOWARDS DATA ACCESSIBILITY various different data silos across the organization.
These will need to be broken down, so that data flows
As identified in our survey, access to quality data seamlessly into data lakes, warehouses, or a
that are valuable towards the objectives and virtualized repository for users to access and work
applications to be achieved are the bedrock of the upon these data sets.
initiative.
Enterprises should also build flexibility into this
infrastructure as you continue to experiment with the
enterprise’s objectives and applications, as these might
change over time with your experimentation and as
objectives evolve.
53
RECOMMENDATIONS TO SHIFT THE MATURITY
EXPERIMENTATION
initiative and can be built and expanded upon to drive
momentum. Good pilot use cases should also tactically
be projects that have a high probability of success with
minimum impacts to current business operations. This
is likened to the minimum viable product (MVP)
BUILD THE TEAM concept in the product management school of thought.
Picking these issues to experiment on also allows the
Bring the identified stakeholders into the team at newly established team to break into the new tasks and
the start. Team members can be generalists with scope and also build rapport across the team.
broad skill sets who are familiar with the subject
matter, complemented by technology experts from EXPERIMENT WITH TECHNIQUES AND TOOLS
the IT and software development teams to
support the experimentation. Data scientists can While a scalable process and state of the art tools will
be gradually added to the team to run the enable scaling much faster later on, companies in this
experiments. As unveiled in our research, higher current stage should focus on delivering the MVP and
maturity teams have included members from the proof of concept as quickly as possible.
operations to data science under one roof to
expedite the discovery process. Teams may also find it helpful to work with external
partners who can help speed up the experimentation
process with their experience and solutions within your
field.
54
RECOMMENDATIONS TO SHIFT THE MATURITY
TOWARDS
development teams towards the core enterprise
objectives, and more data engineers and scientists can
be brought on to continuously develop and optimize
55
RECOMMENDATIONS TO SHIFT THE MATURITY
PROLIFERATION
capabilities to these business units, disseminating their
expertise and experience across the organization.
These pods should become more specialized in certain
interconnected business processes or business units,
developing solutions that have the enterprise’s overall
architecture in mind.
WIDER TEAMS AND MORE SPECIALIZED
PODS STABILIZING YOUR TECH STACK
As the fruits of operationalization have been As processes have shaped into SOPs in the previous
promoted and recognized across the enterprise, phase, the infrastructure and tech stack should see a
more business units and functions will want to certain stability in this phase.
participate and leverage on the organizational
A truly stable tech stack should consider modularity in
resources to accelerate their goals. This will spark
its design, allowing for experimentation and adoption of
a renewable momentum within the enterprise.
new software and solutions as the landscape evolves.
The overall space is still highly dynamic with new
offerings emerging and older ones being retired at light
speed. Hence, it is imperative to keep an open eye on
new relevant technologies to ensure the entire system
stays up to date.
56
RECOMMENDATIONS TO SHIFT THE MATURITY
TRANSCENDENCE
While none of our respondents have currently arrived
at this phase yet, we envision that in this stage,
enterprises will no longer see AI as an independent
SHIFTING TO LEVEL 6 initiative. AI is already part of the organizational DNA,
with all employees, top to bottom, harnessing the
DEMOCRATIZATION
enabling management to rise above their competitors
and achieve high levels of business efficiency.
Businesses seeking to achieve this stage should start
deploying easy to use (no code/low code) software and
applications for everyday users, that allow for a smooth
and intuitive human-system interfacing.
Business leaders should continue to build and
evangelize an AI embracing culture, continually
facilitating open cross-disciplinary collaboration.
Enterprises should also continuously help employees
improve their competency to harness these
technologies to enhance their work efficiency and
workflow.
57
In this report, we hope we have provided a glimpse into
the current state of enterprise AI adoption in Taiwan,
and the challenges posed to companies across each
different stage of the AI maturity cycle.
We are in the early days of technological developments,
and in the harnessing of these technologies. But we
believe that these findings are encouraging signals to
A FINAL WORD
our fledgling state, that promises to mature quickly into
a stronger system for growth in the near future.
We strongly believe that the AI adoption maturity is a
cycle. And the ecosystem of technologies and solutions
continue to evolve dynamically. Even the most
developed companies should keep an eye on the latest
industry developments to sustain their edge, and
maintain the most current technologies to stay relevant.
We look forward to the survey in years to come, and
continue to chronicle the developments of the industry
going forward.
58
METHODOLOGY
59
ABOUT
Hive Ventures is a global venture capital firm
committed to accelerating technological advancements
from Taiwan to build a smarter, hyperconnected world.
Led by a team of entrepreneurs with deep expertise in
AI and big data, the firm invests in early-stage
companies developing the building blocks of AI, SaaS,
Software and other future technologies. It is Taiwan’s
leading fund focused on startups in these industries in
Taiwan.
With an extensive global network and a strong
entrepreneurial and technological background, Hive
Ventures aims to raise the profile of Taiwanese-led
startups and drive Made in Taiwan 2.0 internationally. It
has a track record of backing companies that break
data silos, accelerate development and enterprise
collaborations/processes, and enhance Taiwan’s brand
as the leading market in AI and SaaS innovations,
including successful co-investments with other leading
strategic and financial investors in the region.
60
Origin
As President Tsai’s inaugural address on May 20, 2016
indicated that “The new administration will pursue a
new economic model for sustainable development
based on the core values of innovation, employment
and equitable distribution”, the government will
promote the innovation economy on the basis of the
5+2 industrial transformation such as ASVDP and
Biotech and Pharmaceutical, jointly develop Taiwan’s
new economy development model, and implement
digital transformation of national development.
Vision
In order to connect with advanced technology R&D
capabilities globally for the development of next
generation industry, NDC, MOST, MOEA, and other
ministries will jointly promote the "Asia Silicon Valley
Development Plan," the government will build a R&D
oriented innovation and entrepreneurship ecosystem.
This plan was approved at the No.3514 Cabinet
Meeting by Executive Yuan on Sep 8th, 2016, and was
composed of two core elements —“Promoting IoT
Innovation and R&D” and “Optimizing Startup &
Entrepreneurship Ecosystem”—, linking with global,
local and future, and promoted with four major
strategies.
61
TAIWAN AI ACADEMY
Established by the Science & Technology Ecosystem Development Foundation and the Taiwan Data Science
Foundation, the Taiwan AI Academy is operated by the Artificial Intelligence Foundation with support from the
Institute of Information Science (IIS) and the Research Center for Information Technology Innovation (CITI) of
Academia Sinica.
At the Taiwan AI Academy, we endeavor to nurture generations of top AI talent through our intensive four-month
training programs. To cultivate a fertile ground for AI growth, we have marshalled the professional strength of the
industrial, academic, and governmental sectors in Taiwan and gathered leaders across disciplines under one roof to
serve as our faculty.
Our goal is to help industries in Taiwan gain a competitive edge in the AI revolution and open doors of infinite
opportunities for industrial development. We look forward to a future where professionals across fields can solve
problems in their domains through AI, ultimately becoming the driving force that takes industries to new heights.
62
ALPHA Camp is a tech and startup school present in Canner is a data management software company that
Singapore, Taiwan, and Hong Kong. Our mission is to provides data access control, pipelining, and
help talents in Asia build meaningful careers in collaboration tools to enterprises, maximizing data
technology. We help you develop future-ready skills, an intelligence control, efficiency, and minimizing cloud
entrepreneurial mindset, and the professional network
to succeed in the digital economy costs in one single virtualized database.
InfuseAI is a prime machine learning company founded Profet AI was established by a group of experts familiar
by elite industry experts in 2018. The company serves with operational systems across R&D, manufacturing,
corporations who are striving to offer related IoT, AI machine learning, and cloud native technologies,
development and data analysis results to their aiming to build high value-added auto machine learning
customers. With assistance from InfuseAI, corporations, solutions that solve business and factory management
organizations, and institutions are empowered to problems, thus helping customers achieve digital and
acquire, process, and obtain insights from data and intelligent factories.
quickly see the results through the tailor-made solutions.
63
Cacafly is a leading Taiwanese full service digital marketing group, serving thousands of Fortune 500 and SMB
clients for over an decade. Cacafly provides MarTech solutions and services spanning campaign planning, media
buy, analytics, and ROI optimization.
Its parent holding company funP group (BVI) was founded by a group of NCTU EECS alums in 2006, who
developed in-house AI-powered proprietary adtech platforms such as Tenmax.io, Adhub, and Lychee Acho.
Cacafly now operates offices in Taipei, Hsinchu, Tokyo, Vietnam, and Malaysia.
Please visit us at www.cacafly.com.
64
National Development Council 國家發展委員會 Richie Tsai Acting CEO, Taiwan AI Academy
Asia Silicon Valley Development Agency Ning Ning Yu VP, R&D, LA , Taboola
ASVDA亞洲·矽谷計劃執行中心
Edward Chen Strategy Consultant
Taiwan AI Academy 台灣人工智慧學校
Yves Huang CEO, Cacafly
ALPHACamp
Brian Yang Co-Founder, Cacafly
Canner, Inc.
Wufu Chen Chairman
InfuseAI
Joseph Yang Chairman
ProfetAI
Our sincere gratitude to all the friends, partners, and advisors who have painstakingly assisted and advised us in
designing, producing, distributing, and participating in this survey and report.
65
66
Report Download