ILLUSTRATION:^ MAÏTÉFRANCHI^The last few years have been auspicious
for artificial intelligence, which has been
embraced as a tool for business sectors as
wide ranging as cybersecurity, customer
service and analytics. Data scientists are
now ubiquitous in most industries.
As AI’s influence continues to grow, one
potential pitfall has emerged: the data. The
vast troves of data created by companies
enables machine learning, the technique
whereby algorithms are trained.
However, this data is not always usable,
explains Fernando Lucini, AI lead for
Accenture UK and Ireland. Different kinds
of data can be scattered throughout a
business, “which presents two questions:
how to get it together, then how to use
it in a way that’s valuable?” Lucini says.
A big issue is the data’s “dirtiness”:
missing information, inconsistencies,
errors. “It’s problematic if you try to use
machine learning on a dataset where some
entries are empty or inconsistent,” Lucini
explains. “You can’t count on this data.”
“Businesses do have a sense of the
value of their data. They have a notion, an
educated instinct, that there are answers
somewhere within it,” Lucini adds. “There
often are, but they will need data scientists
to work with the data and help realise the
value. But if the insight is hidden behind
too many issues within the data, maybe
they won’t be able to sort it out, making
it harder to identify value the next time.”
How to proceed? Good data hygiene
would help, but Lucini says companies
don’t often clean and de-silo their data
because they don’t see the use in devoting
time and resources to it – a catch 22.
Research suggests that 79 per cent
of businesses base critical decisions on
data without properly investing in its verifi-
cation. This is risky: take United Airlines,
which missed $1bn in revenue – its pricing
models were built on outdated information.
“Examples like this, as well as instances
within your business,” are how Lucini says
we can convince sceptical executives.
Want AI? Fix
the data first
Smart algorithms are only as good as
the inputs they receive – data hygiene
is key to programming for success
Finding that value relies on focused use-cases. “Things
go very badly when your effort is open-ended,” Lucini says.
In his opinion, one of the key trends we are likely to witness
in 2019 is the progressive “industrialisation of AI”.
Until now, many companies seeking AI capabilities
have been acting unsystematically. “One or two years have
been spent on AI experiments, and many companies feel
that they haven’t seen the return they might have expected,”
Lucini says. “This is the year things need to change.”AI industrialisation needs a framework to help a company
discover AI’s value; a plan to involve all relevant parts of the
company; an approach that aids data scientists in getting the
data; and a reliable data structure. “We can dream up many
ways that AI can make a business hugely successful, but
[companies] need to look at their data through a new lens,”
Lucini says. “Without strong data foundations, there’s no AI.”‘Without strong data
foundations, there’s no AI’- Fernando Lucini, Accenture
WIRED PARTNERSHIP | ACCENTURE
To help businesses, AIs need a clear view of all the data