The past decade, Artificial Intelligence has made rapid strides from boardroom discussions to fueling multi-billion dollar success stories. Now, more and more businesses are realizing the potential benefits and are aiming to make this a top priority for investments.
As we are entering the new decade, we predict that this will be the decade when AI will come of age, and this will mean that it will be adopted to improve day-to-day processes.
Presenting our second part of this series, where we aim to ready enterprises to ride this technology-led wave of growth.
We have also included some case studies to share crucial insights into how successful enterprises are moving on this complex but rewarding journey.
Scaling API-first – The story of a global engineering organization
4 Pillars for an AI Ready Enterprise
1. Polestar Solutions & Services India Pvt Ltd
Steps for an AI Ready Enterprise
4 PILLARS FOR
CREATING A
WINNING AI
STRATEGY
FOR BUSINESS
2. Introduction
Therefore it will be necessary to
constantly revisit the whole program.
A fine balancing act needs to be done
to devise a flexible roadmap.
Companies also need to figure out
how their competitors are using AI.
This will help them better understand
how AI can actually improve their
business as well as discover the right
IT partners who can help them in
their journey.
Companies must avoid pitfalls that
may jeopardize the initiative.
Business leaders must take measures
so that their employees are educated
on the benefits from AI.
In order to derive optimal benefits
from the technology, companies need
to audit their whole IT set-up.
They need to analyze if they need to
upgrade their infrastructure to cope
up with the increased memory &
processing
Businesses are rapidly preparing
themselves for this new paradigm.
But for this to work effectively, they
need get their house in orders.
Lay out crystal-clear specific and
general objectives from the
program, plan the roadmap, set
milestones, gauge the hard and soft
costs associated with the program.
Identify the stakeholders who will
be impacted and figure out how they
are going to measure the ROI of the
whole initiative.
Incorporate the expectations from
stakeholders with regard to the
whole AI program by taking into
account their feedbacks in a
structured manner.
It might get difficult to set the
project roadmap in stone at the
beginning since the program might
need to incorporate multiple
feedback as the program progresses.
3. Companies need to consider few
important points to ensure the success
with their AI Strategy
Data - AI systems will require a large quantity of
clean, usable and relevant data input - does the
infrastructure have the capability to integrate
more data sources
The Need to Start Small - Analyze a specific use
case & KPIs rather than investing heavily into
multiple areas is going to be a good way to start
with the initiative- It will convince the
stakeholders by giving them quick visibility into
the results and benefits
Cost Analysis - Calculating ROI from the
initiatives, understanding budget constraints and
the total cost of ownership
4. Nowadays, AI is the talking point across
every industry.
But, should this be the reason why you are
jumping into AI?
The answer should be 'No’.
According to a McKinsey Global
Survey On Artificial Intelligence,
companies cited the most common
constraints to AI as the following –
1) Lack of clear AI strategy
2) Lack of appropriate talent
3) Functional silos that constrain end-
to-end AI solutions
4) Lack of leaders who demonstrate
ownership and commitment to AI.
6. Like any other solution,
the first thing for you is to
identify a 'Need' for your
business. It can be related
to a problem in your
operations or optimizing
your workforce or finding
patterns in unstructured
data sets or something
entirely new for your
business.
The road to an AI
implementation would
seem rocky and many
times the whole effort
would come into question.
A defined outcome works
as a guiding light at the
end of a dark tunnel. The
turnaround time or payoff
time with AI is going to be
longer.
Mostly because AI is often
linked with complex outcomes.
But this is not true, and it
would be a good practice
starting with a smaller
implementation.
Like automating a repetitive
task that is prone to human
error. Once you show the value
that AI solutions can deliver,
you will have buy-in from the
senior executives.
After gaining your
organization's trust, you can
chalk out an AI roadmap with
all the stakeholders in your
company. It should be aligned
with achieving the long-term
goal of your business.
1. Finding a Business
Use Case
7. Carlsberg, the fourth-largest brewing company in the world
adopted AI. They always aim to bring in innovation and
disruption in the brewing industry. In AI, they have found the
right technology to keep pushing their boundaries and push
new beer flavors to market faster.
Use Case: AI to transform their brewing
process and predict flavors.
Carlsberg have embedded AI as an integral part of their
brewing research process. Traditionally, identifying flavors
required samples to be created for tasting which required
resources and time. Carlsberg has significantly optimized this
process by installing sensors.
They brew microliters of hundreds of different beers every
day. This small volume is not testable by the human senses.
Case Study
8. They use Microsoft AI Solutions with machine learning
algorithms, to analyze the signals coming from these sensors
to measure the flavors and aromas created by the yeast and
different varieties of ingredients.
The timeline for this project is three years and is still into
progress, but they have already realized some great results.
The system is already able to differentiate between lagers.
Researchers are trying to make it friendlier for technicians
to use it in their day-to-day work. Once, the system gets
completely optimized, it will help Carlsberg reduce the time
to research flavors by a third. Ultimately it is aligned with
the long-term goal of - more distinct beer to market faster.
“So we realized that if we had sensors that could tell us at the
outset if the yeast is really going to be usable later in large-
scale beer production — and that could recognize the
chemicals and flavor compounds to predict what a beer will
taste like — that would really help our research a lot.” – says
Jochen Förster, the director and professor of yeast and
fermentation for Carlsberg Research Laboratory.
9. Defining a proper governance
framework is going to be
extremely crucial to oversee a
meticulous change
management programme.
A program management
office (PMO) comes useful
here that can define the roles
& rights, provide maker-
checker role and strong
monitoring of the entire
process.
It is important to have well-
defined governance
framework in place for
regulating the AI-usage.
The accuracy of AI system
depends upon the quality of
data input. It is important to
have active support from all
the stakeholders.
A governance system
manages and defines the rules
for data access and
collaboration, allowing
organization to maintain
control over their data assets
in a proactive and responsible
manner.
Does your organization have
the capability to assess the
risks? Organizations must
have a coherent and clear AI
strategy.
Many companies are still not
mature in identifying and
formulating a well-defined
governance mechanism
around this technology.
Data that is shared must not
get misused and the integrity
of the data is maintained.
2. Governance
10. Rapidly evolving regulatory frameworks require
companies to constantly rethink their data protection
policy.
Many times, everyone at the company may be
unaware about the nuances of such regulations and
this may prove dangerous. Hence, it is important to
have effective guardrails and to communicate these
details effectively throughout the organization.
And this is bound to affect the effectiveness of AI
algorithm which demands the-greater-the-better access
to data for making more accurate predictions.
"As organizations move from optimization and
automation to innovation and beyond AI, they
may want to ask different kinds of data-centric
questions such as, ‘What are the boundaries
for using data to perform analytics?’ ‘How are
those boundaries impacted by AI?’ ‘Who are
the stakeholders in these conversations?’,”
- Deborah Adleman, Director, EY LLP
11. For example, when the civic administration of Espoo,
Finland’s second largest city, wanted to devise an intelligent
social welfare system that would deliver the right benefit to the
right citizen at the right time, they took on the help of
Microsoft to build an AI system that would process over 500
millions of rows of data and not only deliver the civic
authorities a holistic overview of the system, but also help it
optimize the utility, infrastructure, educational, cultural,
health, and social welfare services for its constituents.
But before developing the AI system, Espoo needs to address
significant security, data privacy, and regulatory concerns.
Current GDPR rules require data to be anonymized to protect
citizen privacy, but that would reduce the effectiveness of the
AI algorithms to make predictions.
The city is also working with international organizations and
government and business partners in Finland to establish
globally recognized standards and rules for the ethical and
transparent use of AI in the public sector.
“Our citizens want to know how their information is used, so
we have responded by creating and communicating ethical
rules for AI,” says Päivi Sutinen, Director for City as a
Service Development at the City of Espoo.
Case Study
12. AI strategy will be successful
if it is linked to real business
use case for the company. For
this, company would require
input and active support from
its business leaders.
Companies also need subject
matter experts to work
closely for successful
development of AI expertise.
When data scientists and
subject matter experts work
in tandem, they are better
able to prioritize the right
business use cases to solve,
understand opportunities and
its relevance for the company
and consider all the
stakeholders - take into
account the data spread
across departmental silos,
and provide context to the
solution.
3. Subject Matter Experts Need to be
Involved in AI
Usually the data scientists
tend to have less experience
of working on the actual
business concerns and
priorities for the company.
It will require a well-defined
mechanism to capture
feedback from users.
Companies will require a
program management team
which will prioritize and
incorporate the feedback in a
structured manner.
Many tools today make it
easy for non-technical people
can try their hands at the data
science wizardry.
Gartner coined the term
“citizen data scientists” to
refer to people from non data
background who are
groomed with the data skills
13. However, this does not mean
doing away with a dedicated
data science team altogether.
These experts, with thorough
grounding in math and
statistics are going to be
indispensable.
Subject matter experts might
not have the technical and
mathematical acumen to
assess the intricacies &
capabilities of AI models.
Data scientists will be able to
outline the bounded realities
of AI to subject-matter
experts.
They will be able to set
expectations for the features
of a machine learning model:
what it can and can’t do for a
business. This intricate
balancing act will only be
successful if both sides work
together. For example, a
marketer may expect that by
deploying
a machine learning model, they
would be able to optimize the
bid amount, time and targeting
of a particular ad on social
media
A data scientist would be able
to assess the feasibility of this
analysis by studying the
available data, technical
expertise available and what
they would need in order to
make it possible.
Once the AI system is up and
running, the role of subject
matter experts don’t end.
Subject matter experts can
provide valuable context to
determine if the AI system is
leading to insights that can
result in improved processes.
Taking feedback from these
subject matter experts in order
to understand what their
expectations will lead to
adoption
14. For manufacturing companies, having continuous machine
health monitoring and pre-empting catastrophic failure of
equipment, are very important use case that can be done with
“intelligent machines” resulting in minimized machine
downtime, maximized productivity.
Pumps, fans, and motors in plant environment often have
problems such as lubrication, cavitation, looseness,
imbalance, as well as bearing issues. In this case,
accelerometers are used to monitor vibration and ultrasonic
energy. These critical measurements can sense problems in
rotating equipment.
For example, when problems occur on equipments, a
maintenance professional will create a work order and then
it will be queued up to be serviced. The question then
becomes whether or not that analysis can be leveraged on a
continual basis.
For example, in manufacturing predictive maintenance, an
equipment expert could identify what parameters are
fundamentally related to observing and making predictive
analysis..
Case Study
15. A process expert could relate how different contextual
environments and their combinations of data such as
temperature, pressure, humidity, ambient temperature, etc.
might result in equipment failure.
An analytics expert could determine the best algorithm(s)
to model and predict the trend given the data size,
parameter set, dynamics, etc. The data scientists might
believe that the system is making accurate predictions on
the optimal maintenance schedule on the machines, but
they lack the necessary context about those machines.
A data scientist can’t interpret the numbers to figure out if
in fact one machine should be maintenance more often
than the predictions are calling for.
A domain expert at the manufacturing company who has
experience overseeing the machine maintenance has that
context. This would be valuable to the data scientist in
making the right adjustments in the weight of the features
that they include
Case Study
16. Any strategy is incomplete
without estimating the
budget it will require and the
corresponding returns you
expect to realize.
In the preceding steps, you
have already discussed the
business use cases of AI,
gaining SMEs support and
the importance of setting up
a robust governance
framework.
Each of these stages has a
cost attached to them. For
obtaining a clear picture you
need to quantify the benefits
in terms of numbers.
Calculating the ROI during
the pilot is a mandatory step
and helps in getting the
green light for the
deployment.
4. ROI – Cost Benefit Analysis
For example, in case of a
pilot of predictive
maintenance of machine
setup, you can easily
compare the change in
uptime of the machine post-
AI implementation.
Only if the investment made
against the underlying
technology underweights the
benefit accrued from
reductions in maintenance
cost, you should go ahead
with the deployment.
However, in the case of AI,
calculating ROI is not
always straightforward.
There are multiple factors
acting at the same time
making it complex to
quantify the benefits from
an AI solution.
17. with governance framework
need for AI and cost incurred
in setting up a data platforms
on cloud or on-premise,
equipment or software costs,
energy and data center cost,
training and extra manpower
cost as well.
If even after this exercise the
ROI is positive, it is indicative
of the right path of
deployment.
AI solutions tend to deliver
some intangible returns which
companies often realize in the
long term.
The AI champions should work
with the finance team to
determine the ROI projections
in the long run. There is always
an inherent risk involved while
working on emerging
technologies. Keeping this in
mind, it is always advisable to
factor in this risk and adding a
cushion to the cost projections
while calculating ROI.
In these cases, it becomes
difficult for the business
stakeholders without the
empirical results of
investment paybacks to air
their opinion in front of the
top management.
To calculate the ROI of an
AI implementation, the
stakeholders are required to
provide the returns from the
entire business unit.
Since the pilots are done on
a limited set of operations
like if the packaging
assembly line is automated
but the rest of the processes
remain the same, the
benefits of AI will be lost.
So, in the initial phases, it is
important to set the
company’s expectations for
a single process only.
However, in the long run, it
becomes important to factor
in the cost associated
18. SR-Bank was looking to ramp up their customer service
need owing to growing customer base. They wanted to
reduce the cost associated with handling customer queries
and impart better experience.
Use Case: A secure Conversational AI Solution backed by
NLU & ML – Banki
SR-Bank wanted to achieve better control and provide a
superior experience to their customer in an efficient way.
They set up systems like Natural Language Understanding
(NLU) and Machine Learning (ML) to set up virtual agents
for customer queries redressal. The virtual agents were
able to resolve most of the queries, but advancements of
conversational AI presented them with some very exciting
opportunities.
Banki – an AI-enabled conversational solution was
developed. The results with Banki are so ground-breaking
that SR-bank has prioritized a ‘Chat First” strategy. Out of
approximately 25,000 monthly conversations, Banki is able
to recognize around 2,000 intents along with response time
and level accuracy that isn’t possible from human
personnel.
Case Study
19. Also, the volume of customer queries handled by SR-Banks has
dramatically increased by nearly 150%. As per their
calculations, its capacity is close to 20 full-time employees. It
frees up human resources from the repetitive tasks allowing
them to focus on more creative tasks.
Banki has really delivered some great ROI in error handling
encountered while authorizing their e-banking solution. Users
typically get an error during a network outage; in such
incidents, the error code is redirected to Banki. It then pops up
and guides the customers by telling them to either try again in a
couple of minutes or contact customer service.
In March of 2018, Norway’s primary electronic identification
system, BankID, suffered some technical glitches disabling
customers from login into their accounts. During the outage,
Banki successfully fielded more than 4 000 conversations in a
single day. It couldn’t have been possible without the help of
Banki.
Case Study
20. Wrapping Up
This has been a decade where;
businesses have realised the potential
that AI holds. As we are entering a
new decade, technology gurus are
predicting it be a decade where we
will see AI coming of age and getting
adopted in day-to-day processes.
In light of these developments, this
series aims to ready enterprises to
ride this technology-led wave of
growth.
In part -1 of this series, we walked
you through the basics of AI. Then,
we dwell into the key aspects of
building an AI strategy from scratch.
We break it down for you into:
❑ Organizational
❑ Data Management
❑ Infrastructure
❑ Analytics
to measure your enterprise readiness
and formulate a comprehensive AI
strategy.
This part of the series presents you
with typical challenges companies
face while starting with their AI
strategy.
So we have broken it down into four
key pillars that could help you to
tackle them and garner the support of
business users. These are:
❑ Finding a Business Use Case
❑ Governance
❑ Subject Matter Expertise
❑ ROI – Cost Benefit Analysis
We have also included some case
studies to guide you along this
complex but rewarding journey.
You can also take some outside help
from companies who have
successfully done this for other
organisations.
At Polestar, we have helped
companies in their technology-led
transformation of business operations
and building a culture of innovation.
Link to Part 1: AI Strategy
for Business