ESCP ALUMNI magazine - #204 - Novembre 2023 - 40

Campus ❙ École
identify the pitfalls in existing processes.
This helps optimise inventory levels, production
planning and distribution - ultimately
achieving a better balance between
supply and demand.
AI can also generate entirely new insights
that would otherwise have remained hidden
in the data, our expert adds.
But there are challenges and barriers that
companies commonly face when implementing
AI in their digital transformation
strategies.
When building
the AI roadmap,
it's important to
ensure the data
is accessible, that
we have the right
quantity and
quality of data.
Uncover latent insights
Consider hyper-customisation, or the ability
to create products, services or experiences
that are tailored to the needs and wants
of individual consumers. AI systems can
pour over customer data - including preferences,
purchase and browsing history, plus
social media activity - to serve them up the
most relevant options from a menu of customisations.
" We can identify the best way
to satisfy their needs, " Dvaretskaya says.
In a similar vein, she reminds us of an example
in life sciences, where AI has accelerated
the drug discovery process - by
identifying therapeutic targets and drug
candidates, as well as predicting the efficacy
and safety of the latter. AI algorithms can
scan through chemical, biological, and clinical
data in order to estimate how a patient
may respond to a certain drug candidate.
" Thus, AI helps optimise clinical development
and thus, bring treatments to patients
faster, " our expert says.
40 ❙
Amass diverse, high-quality data
Firstly, to build robust and accurate AI
models, Dvaretskaya points out that a substantial
amount of diverse and high-quality
data is required. This data serves as the
foundation on which AI algorithms are
trained to spot patterns, understand relationships
and make predictions or even
decisions. Without high-quality data to
feed these machines, they cannot deliver
valuable insights. Garbage in, garbage out
is the common expression.
" When building the AI roadmap, it's important
to ensure the data is accessible, that we
have the right quantity and quality of data, "
she says. " Quite often in an organisation, the
data is scarce, siloed or incomplete. And it's
complicated to get the full picture because
the data is spread between different functions
or difficult to complete with external
sources. So it is vital to get the data foundations
right and make data compatible. "
The second major challenge is access to data
scientists with the right skills and expertise
to extract valuable insights from massive,
complex datasets - especially when that is
combined with other in-demand competencies,
such as software engineering. " Those
profiles are crucial to be able to not only
create a proof of concept but actually bring
to life new processes, products and services
using AI, " says Dvaretskaya.
Outline precise business goals
Prior to addressing the data and talent obstacles,
however, it is vital for organisations
to define the specific business objective
behind their digital transformation efforts,
according to her.
" Do not try and implement an AI solution
for the sake of doing so, but rather start
by analysing the actual business needs, "
she explains. " This way you will naturally
prioritise the solutions that have the biggest
impact, and they are more likely to be
accepted by everyone in the company. "
She adds that by setting clear objectives and
a step-by-step process to achieving them,
you can improve the overall efficiency of
the project.
ESCP Alumni Magazine numéro 204 ❙ NOVEMBRE 2023
Minimise bias in algorithms, boost
transparency
Additionally, our expert identifies two primary
ethical considerations linked to the
use of AI in digital transformation. One
concern is the potential for human bias
to become embedded in the data used to
train AI algorithms, and subsequently replicated
by the AI system. " There are some
very well-known examples of biased data
sets leading to discriminatory results, "
Dvaretskaya says, including some companies'
" sexist " hiring algorithms that showed
bias against women.
The second concern is that the algorithms
and decision-making processes used by
the AI system are not easily explainable or
interpretable by the humans using them.
" Some of the AI systems are black boxes, "
she explains, adding the absence of transparency
can result in individuals misinterpreting
the results.
Additionally, Dvaretskaya warns against
" concept drift " , when the patterns in the
data being modelled by an AI system evolve
over time. This can mean the assumptions
made by the system no longer hold, leading
to inaccurate or unreliable predictions and
decisions. " If the situation changes drastically,
the results become false, " she says,
underlining the importance of continuously
monitoring data for changes and updating
models using new data.
Through addressing these ethical
concerns and overcoming challenges
around data access and quality, access
to talent and skills, as well as taking systematic
approaches to AI, companies can
ultimately unlock the potential for this
technology to accelerate digital transformation
- a strategic imperative.
❙
This article was originally published on
ESCP's media, thechoice.escp.eu.
http://thechoice.escp.eu
ESCP ALUMNI magazine - #204 - Novembre 2023

Table des matières de la publication ESCP ALUMNI magazine - #204 - Novembre 2023
