CASE STUDY
T
oday, every business has an email
address. And if you’re a financial
services organisation with many
business sections and over a million clients,
that channel can get crowded.
Nedbank Insurance covers clients’ life
and other insurance needs and receives
thousands of emails about claims, policies,
address changes, complaints, queries and
potential new clients. Some emails are long;
others are short. Many have attachments
that must be categorised correctly to
simplify processing for back-end teams.
There are also unpredictable peaks when
for example heavy storms cause inboxes to
be bombarded.
Indranil Bandyopadhyay, Head of Business
IT Enablement for Nedbank Insurance,
suspected that an automated system could
solve the problem and increase operational
efficiency and client satisfaction.
“With the old system, a single person can,
on average, process one email every 60 to
300 seconds (three minutes on average),”
he said.
“That’s about 160 emails per dedicated
resource a day. With email volumes
growing, our backlogs were mounting, and
we didn’t want this to influence our client
service. We knew technology could help us
create a more sustainable solution, but we
needed the right partner.
“Nedbank Insurance had to try a couple of
times to get this right. The challenge was
that they [other companies] didn’t try to
solve the real problem and looked towards
their overseas counterparts to solve for this
problem. Some of them went half-way and
gave up.
“I also engaged with international vendors
who mentioned they couldn’t solve the
end to end problem. This made me feel we
are really trying this for the first time in the
world. I did not give up because I believed
this could be solved.
“Grit and tenacity along with the right
partner were the reasons we were successful
with Synthesis.”
GRIT AND TENACITY ALONG WITH
THE RIGHT PARTNER WERE THE
REASONS WE WERE SUCCESSFUL
WITH SYNTHESIS.
Solution
Following various methods to analyse
email and attachment data, Synthesis
built predictive Machine Learning (ML)
models with natural language processing
algorithms to understand the intent of
the email. The models first learnt from the
employees at a rapid pace, refining the
process through feedback. Then, when
the models performed reliably in terms of
www.intelligentcio.com
predicting where emails should go, they
were routed automatically.
“I have been thinking about ML for the past
three years,” said Bandyopadhyay. “The
benefits are huge. Especially in an admin
orientated industry like insurance.
“While RPA can deal with repetitive stuff, ML
can be used where cognitive ability is needed
to complete a task. The main challenge
of ML and automation in general is the
introduction of ‘positive feedback loop’. This
has a leg for a process to go rogue and can
have huge negative impact in a very short
period of time.
“Sometimes, the amount of data to teach
the ML algorithms is also huge challenge and
the quality of data is also a major constraint.
But what struck me is that Synthesis used
Design Thinking to truly understand what
the problem was.
“In a very short time they created a
prototype, tested it with users and then
adapted it. The solution they created
revolved around genuine client satisfaction
and not just completing a task, and that
worked for us.
“Design Thinking assists in understanding
the real problem and not the superficial
issue. After asking 10 whys, one gets to the
core of the problem.
“The issue for us was to find a solution
to deal with a huge number of incoming
emails. Our emails were in the Microsoft
Inbox and the attachments needed
downloading on to a local PC.
“Once a human being understood what the
emails were all about, they needed to go to
our policy admin system and then categorise
with the documents attached.
“While we thought about using a ML/AI
solution, Synthesis was quick to understand
the real problem via DT. It quickly created
a front end that pulled emails and
attachments and, once the human being
understood the content of the email, the
front end automatically attached the
document in the right category.
“This immediately gave 80% efficiency
though no ML was used at this stage.
INTELLIGENTCIO
51