Amelia in Action
A selection of stories
from organizations
adopting IPsoft’s
cognitive agent, Amelia
Meet Amelia
A glimpse into Amelia’s brain
Your first digital employee
Amelia is IPsoft’s virtual cognitive agent. Her mission is to deliver best-in-class service to
customers, fully automating human-to-human interactions and process execution.
Like a human, Amelia communicates using natural language and can respond to customers’
emotional states. Unlike a human, she can hold thousands of conversations in parallel.
Why is Amelia different?
She understands everyday language
Amelia stands out from other technologies through her ability to understand natural
language; not simply the words we use, but also their intended meaning. In contrast to
pattern-matching platforms, Amelia can comprehend like a human to get straight to the
point.
She learns quickly and gets smarter
Amelia can follow process maps created from her prior interactions. And like any smart
worker, she observes colleagues to discover the optimal course of action. Amelia can then
apply her learning to address similar future scenarios without human intervention. If she
cannot address an issue herself, she escalates to a human colleague.
She adapts to us
Whereas other technologies demand that humans adapt their behavior to interact with
“smart machines,” Amelia adapts to human behavior.
2
Episodic Memory
to understand what
your customer
wants in context,
and provide
immediate answers
Neural Ontology
to allow your
customers to have
a very natural
conversation with
Amelia
Process Ontology
to execute a process
for your customer
in order to address
their needs
EQ Ontology
to enable Amelia to
adapt her responses
to your client’s
emotional state
3
SEB, Nordic Bank
During the rollout, an SEB project team member
commented, “The response is always positive when
we introduce Amelia to the staff. As soon as Amelia is
demonstrated, the room fills with positive energy.”
IT Service Desk Agent
SEB, the leading Nordic corporate bank, has
completed a rapid deployment of IPsoft’s
cognitive virtual agent Amelia inside their
IT Service Desk. The Amelia pilot rolled
out in August 2016 and within three weeks
handled over 700 bank employees and 4000
conversations. The project met its targets
two weeks ahead of schedule.
4000
conversations with
700
3
employees
in
weeks
Amelia’s role covers two internal business
cases which make up 15% of the Service
Desk volume: Identity Access Management
and Knowledge Management. These were
chosen after prioritizing 90 tasks Amelia was capable of supporting. Amelia is engaging
directly with employees to:
•
Unlock Active Directory accounts
•
Unlock accounts for a mortgage application for home loans
•
Provide password guidance
•
Supply knowledge base answers to questions like ”How do I order remote access?”
44
Of her first 4200 conversations the majority were
solved by Amelia, so agents were freed from repetitive
employee queries. As for the queries Amelia cannot
resolve, she observes the employee-agent interaction,
learning how to deal with similar variations. Once her
learnings are approved by her human supervisors she
can answer these queries herself. The pilot phase
also measured the ability to automate requests endto-end by integrating Amelia with IPcenter, IPsoft’s
service delivery platform. SEB had already deployed
IPcenter through a long-standing contract with IPsoft,
so new automations were quickly deployed. Amelia integrates securely with SEB back-end
systems via IPcenter’s autonomic engine, giving her a robust platform to engage with.
By adopting Amelia in its Service Desk, SEB can improve user experience and speed up
response for requests, while providing staff with time to dedicate to more complex requests.
The bank’s deployment highlights the potential of integrating digital labor, autonomics,
people, processes and technology into a single system. Amelia’s performance has inspired
SEB to continue the journey. The bank plans to extend Amelia on the Service Desk and
incorporate support centers and customer-facing channels.
5
Enfield Council
Public Service Virtual Agent
In a move to enhance customer service for more than 330,000 residents, the north London
borough of Enfield is adopting Amelia.
Enfield is one of London’s largest boroughs and its population is growing by four to five
thousand each year. Demand for service is growing all the time and each month the
council receives 100,000 visits to its website and takes 55,000 telephone calls. Sustaining
consistently high quality customer service in order to meet rising expectations 24 x 7 is
challenging. This is particularly difficult when set against a backdrop of central government
spending cuts. By introducing Amelia, the council expects to increase the volume of queries
it manages; Amelia will be able to absorb time intensive routine requests while freeing up
the time of council employees to focus on more complex issues. In short Amelia will help the
council deliver more with the same resources.
6
In the first instance, the council plans to
implement Amelia to answer general queries
coming to the website — answering requests in
an intelligent, non-scripted way. In addition, the
council will explore how far Amelia can help in
managing application processes for specific areas:
for example, pre-screening planning applications
and providing self-certification for those building
plans that fall within specific parameters.
As Amelia works alongside the existing service
channels, residents will be able to select the way
of interacting with the council that best suits their
personal needs.
Rather than requiring diverse visitors to be
technology-literate, Enfield Council will require
that their technology be “people-literate.” Given the
fact that Amelia interacts using natural language,
the expectation is that she will be well placed to
support everyone.
Enfield’s pioneering adoption of cognitive
technology is expected to set a trend for other
public sector bodies both in the UK and across
other regions.
“Our approach to
transformation
embraces digital
technology to find
completely new ways
of supporting residents,
which, in turn, frees up
valuable resources for
reinvestment in front
line services. Deploying
IPsoft’s world-leading
artificial intelligence is
another major milestone
in this journey.”
—James Rolfe
Enfield Council Director
of Finance, Resources &
Customer Services
77
Global Bank
Mortgage Broker Agent
One leading global bank intent on seizing first-mover advantage in its digital strategy is
working with IPsoft to transform key operations by incorporating Amelia. In order to test her
flexibility and breadth of understanding, the bank set out a project that could demonstrate
the depth of her capabilities by incorporating her into the mortgage broker advisory team.
Around the clock external mortgage brokers request information about bank products and
policies. Their questions need to be answered correctly to ensure that subsequent mortgage
applications are compliant and can lead to a successful approval. A fast response can
make the difference between a broker selling the bank’s product or that of a competitor. By
equipping Amelia to respond to queries, a bank could be sure to satisfy both requirements.
Just as one would with a human agent, the first step in Amelia’s training was to assemble the
data to give her the knowledge required to answer 160 of the most common queries raised by
mortgage brokers. The project team then began to test her ability to manage the questions.
The questions themselves were highly challenging. For example, “My customer is self
employed and has land and property income, do you take this into account?” is an interesting
question for Amelia to answer as there isn’t enough information shared by the customer to
match all the parameters defined for providing a response in her training.
Amelia was able to understand the underlying intent, which is to confirm the acceptable
income types for a mortgage application. However, to provide an accurate response, Amelia
8
also needed to interpret “land and property income” as a synonym
for rental income and seek clarifying information to see how this
attribute might impact the mortgage types available. As a result,
Amelia would ask, “Is your mortgage type residential or buy to
let?” Only once this has been answered is she able to determine
the answer on the basis of having established employment status,
additional incomes available and purpose of mortgage.
Over the space of just two weeks, Amelia’s ability to comprehend
the questions, however they were phrased, and provide the correct
response, rose dramatically. At the end of the training period Amelia
could answer 120 of the full 160 with an 88% success rate. The
speed with which Amelia’s performance improved has now led to the
bank to extend the scope of potential scenarios in which she could
impact operational efficiency and revenue growth.
At the end of the
training period,
Amelia could
answer 120 of the
full 160 questions
with an 88%
success rate.
9
Oil & Gas Company
Invoice Query Agent
Working together with Accenture, IPsoft
conducted a project for a large oil and gas
company to assess whether Amelia could help
provide a prompt and more efficient way of
answering invoicing queries from suppliers.
Amelia was trained in the process to respond
to the top 25% of questions received from 500+
suppliers.
To provide the most effective response, Amelia
was integrated into a supplier self service
portal. A secure log in is required, so whenever a
supplier initiated a conversation, Amelia would
instantly be aware of whom she was speaking
with and which organization they represented.
The single sign-on approach opened up the
possibility of providing personal context to
every interaction and the ability to eliminate
time wasting basic questions. It also allowed
for varying levels of access to information to
10
be applied depending on the individual in the conversation. Upon
contact using a live chat function, Amelia reconfirmed the identity
of the supplier, before answering questions in the same way as a
human service desk agent. Typical questions included:
•
•
•
I received a payment of $512. What’s it for?
I would like to know the status of an invoice.
When will my invoice be paid?
As you would expect of a human agent, Amelia asked questions to
establish exactly which invoice was being referring to. For example:
•
•
•
What is the invoice number?
What date was the invoice issued?
What amount was the invoice for?
Amelia’s
estimated
resolution rate
was 72% in less
than 8 weeks.
A warm “handover” process was established to provide the full context of the conversation
taking place to a “live” agent whenever Amelia needed to escalate the query. Importantly,
when a live agent joined the conversation, Amelia stayed engaged to observe the dialogue
and learn what she should do in future. Although she would subsequently put forward
updates to the process based on what she had observed, an SME check point was in place to
allow the company to ensure that Amelia would only learn new steps that they had approved.
To roll Amelia into production, it will be necessary to give Amelia access to the company’s
ERP systems where invoicing data is held so she can retrieve answers and execute next
steps such as uploading an invoice. Based on the test phase Amelia’s successful resolution
potential was estimated at 72% with less than 8 weeks of on the job training refinement.
11
Insurance Company
Media Company
A large US-based insurance company approached IPsoft about digital service agent
integration with their support website. The company’s web support service already
has more than 3,000 users a week and is growing quickly. By using Amelia, the
increased demand for chat-based interactions can be met while improving customer
experience. More importantly, the company sees the potential to have a positive
impact on revenue growth when Amelia is involved in the initial quotes process.
Initial efforts to test Amelia’s ability to provide quotes focused on auto insurance
and integrating the process with the existing India-based chat support function and
relevant systems. Before providing a quote, Amelia queried and verified relevant
data: for example, customer’s zip code, car model and year, homeowner’s insurance,
etc. Early trials successfully integrated the client’s Amelia instance with existing
APIs for quote generation to provide accurate and compliant offers to customers.
A large US-based media services organization with millions
of customers across the country sought to explore the
potential for Amelia to raise the bar for customer service.
Together with IPsoft, the company wanted to increase
the speed with which customers could receive assistance
in resolving technical issues with internet, cable and
telephony services. It sought to test how Amelia could be
integrated into the team fielding more than 65,000 calls
a month. The project placed Amelia between the first line
call center agents and the third line agents with deeper
technical skills. This would allow Amelia to work alongside
her colleagues to shrink the time taken to respond to highvolume, repetitive questions.
In parallel, Amelia is being trained to answer 150 FAQs that come through on a
regular basis. Rather than have customers navigate unwieldy web pages in search
of information, Amelia will be able to provide an accurate, near immediate response.
Customer service improvement remains at the forefront of the initiative. With
trials going well, the insurer is already exploring future scenarios and considering
employing Amelia to assist in dental insurance applications.
Amelia was trained to manage common requests including
account unlock/reset, rate code investigations, porting a number and access requests.
It took 3 months to train Amelia to respond successfully to 64% of the queries on which
she was trained. Results of pre-production trials showed mean time to resolution (MTTR)
drop from 18.2 minutes per query to 4.5 minutes per query. Similarly, average speed of
answer waiting times fell from 55 seconds to 2 seconds.
Digital Service Desk Agent
12
Digital Service Desk Agent

Mean time to
resolution per query
dropped from 18.2
minutes to 4.5
minutes.
13
13
Insurance Company
Digital Service Desk Agent
This Fortune 100 insurance company is interested
in the potential of digital agents to improve
efficiency of the business whilst improving
customer service. After signing a contract to
deploy Amelia in March 2016, the company
decided to focus on training Amelia to support two
initial user groups: agents who sell the company’s
insurance out in the field and the firm’s existing
call center agents.

Amelia is being
trained to assist with
as many as 166,500
calls received
monthly.
Amelia will interface with insurance agents to ensure they remain productive and can
access all the technology needed to sell the company’s products. For example, Amelia will
guide one of the licensed agents through installing essential software. She will help the end
user work through a series of known steps to resolve the issue.
Amelia will also guide insurance agents through which forms need to be compiled and
submitted. As part of the pilot underway with just under 100 field agents, Amelia is resolving
common queries and integrating with other systems — most notably the company’s policy
and underwriting applications and IT service management tools including ServiceNow.
14
In parallel, Amelia is being deployed to
improve the customer service through
support for the company’s customer
service center and being trained in
the first phase to help deal with as
many as 166,500 of the monthly calls
the customer service center receives.
The overall goal is to improve support
quality, reduce the training time for
agents and reduce turnover. Amelia will
help the agents understand their role,
provide advice and assist their learning,
especially in terms of the new systems
they need to use. Overall operations
efficiency will also be impacted.
Amelia will route queries to the most
appropriate customer service agents.
The target impact is clear:
•
Reduce new customer service center unlicensed agent training from 14 weeks to 10
weeks
•
Reduce new customer service center licensed agent training by 4 weeks
•
Reduce the average call handle time for new analysts (those that have been employed
for fewer than 6 months) by more than a minute.
15
US: 212-708-5500
Email: [email protected]
Web: www.ipsoft.com
© Copyright 2016 IPsoft. All rights reserved.