There is no doubt the universe of artificial intelligence extends far beyond the world of robotic process automation. AI is actually an umbrella covering a broad set of methods, algorithms and technologies that make software ‘smart’. Machine learning, computer vision, natural language processing, robotics and related topics are all part of AI. They collectively form a subset of AI broadly termed ‘cognitive technologies’.
RPA and AI impact on Banking - 4th Annual Back Office Operations Forum, Vienna
1. RPA/AI and its Impact on Banking
2016 Back Office Operations Forum
Guy Kirkwood, COO, UiPath
guy@uipath.com
2. I think, therefore RPA/AI
“I propose to consider the question ‘can machines think?'” - The words of Alan Turing
introducing his plans for what is now known as the Turing Test, in which it would be verified
whether a machine can develop the ability to exhibit intelligent behavior equivalent to that of a
human being.
AI means many different things to many different people because the way we think about and
define intelligence varies fundamentally. For our current purpose we will treat AI as technology
which is used to do tasks that require some level of human intelligence to accomplish.
Like AI, robotic process automation (RPA) is a term that often generates as much confusion as
it does clarity.
In order to assume the role of a human equivalent for automating business processes, RPA
robots -or more accurately, automation software - must, as an essential criterion, be able to
mimic two human actions.
- First, robots must be able to find the screen element where some process action takes
place. If a choice in a drop-down box has to be clicked, the robot’s worthless unless it
consistently can find where the drop-down box is on the computer screen.
- Second, a robot must be able to take the correct action when it finds the screen element.
It has to be able to apply decision-making rules to perform the same action as a person
would.
3. The Fit between RPA & AI Technologies
There is no doubt the universe of artificial intelligence extends far beyond the world of robotic
process automation. AI is actually an umbrella covering a broad set of methods, algorithms and
technologies that make software ‘smart’. Machine learning, computer vision, natural language
processing, robotics and related topics are all part of AI. They collectively form a subset of AI
broadly termed ‘cognitive technologies’.
And within cognitive technology there are specific disciplines particularly relevant to RPA:
- Computer vision: the ability of robotic software to identify objects, scenes, and
activities.
Think of how Facebook automatically tags photos by: using facial recognition technology
on the photos; then analyzing FRT results against an analysis of your image history,
along with those of your friends.
- Machine Learning: businesses use this cognitive technology to gain the value of
software capable of acquiring knowledge through supervised or unsupervised learning.
Take, for instance, software generated credit scores. The software begins by analyzing
targeted data, arriving at a proposed credit score, and submitting it to an employee who
either accepts or rejects it.
Drawing lessons from these acceptances and rejections, the software improves
knowledge of how to analyze the target data and arrive at better outcomes. Eventually it
learns to be as accurate as employees performing the same credit score task. When it
reaches that point, direct supervision is unnecessary.
Unsupervised learning is more challenging. It’s also a larger part of our lives than credit
scores. For instance, how many of us have had Amazon or Netflix display
“Recommendations because you watched Breaking Bad” - or some such show? That’s
an example of unsupervised learning. The software infers classifications or boundaries
based on analysis of past a customer’s previous viewing behaviors and patterns.
Unsupervised learning, without the need for humans to evaluate its work, is a highly
scalable solution. Error consequences are also much lower in the Netflix scenario.
- Natural Language Processing: the point of NLP is to enable software to process highly
specific written content just as a person would, thereby identifying valuable information
such as: topics; purpose; meanings; specific requests or questions.
4. To appreciate the consumer value of NLP, consider the possibilities it
brings to Live Chat alone. With NLP software able to pull those various types of
information from a Live Chat exchange, customers could easily avoid having to ring up a
call center again - ever.
To understand the business value of NLP, imagine how useful the scalability of a
NLP-based customer help feature will be for processing returns on the day after
Christmas.
Cognitive technologies have such obvious value for RPA solutions the better question is: what’s
the best fit for RPA and cognitive; not, do they fit? Andrew Anderson, CEO of Celaton,
commented on the question:
“One thing I do see happening in the automation world is a convergence of technologies. Take
robotic automation, cognitive automation, content analytics – all of which you could describe as
being part of that convergence. I see them converging into solutions that will address all these
areas for the customer.”
Intelligent Robotic Process Automation
What do we mean by “Intelligent Robotic Process Automation”? For this discussion the term
describes how robotic process automation, infused with elements of cognitive technologies,
significantly increases the power and value of automation solutions.
To substantiate the potential for greater performance and value, consider this example of an
extensible platform-based solution used to seamlessly integrate robotic software with cognitive
technologies for the purpose of structuring data prior to automation.
In fact, this integrated solution perfectly reflects what David Poole, CEO and co-founder of
Symphony Ventures, Ltd. recently described in an interview as “future-proofing”.
5. Challenge: A customer receives over 70,000 invoices per year in various
formats incompatible with their SAP system. Currently these invoices are placed in a SAP
mailbox with an initial work item and status; a BPO provider accesses them and extracts the
necessary information to index and post them into the SAP system. Further cost reductions will
be difficult, human errors are an issue and scalability is limited.
Solution: UiPath robots pull invoices from the mailbox, captures work item and status for log
files, then places them in a work queue. The queue transfers them to Celaton’s inStream
cognitive product which extracts the unstructured data, structures it into a standard format and
checks it against a SAP validation form. Failed invoices are placed by UiPath Orchestrator into
an exceptions queue which logs and returns them to the customer.
With successful invoices, the inStream places the formatted data into work files tagged with the
invoice status and unique work item. Orchestrator then places these work files into an index
queue where robots index them into the SAP system; upon successful SAP indexing,
Orchestrator places the work files into a posting queue where robots post them as invoices into
the SAP system, then send them for customer archiving.
Value: The customer received the benefits of both technologies: Celaton’s cognitive software
turned unstructured data into structured; UiPath’s robots indexed and posted the structured
invoice data into SAP with low-cost, scalable robots devoid of human error.
Here is another example of how an extensible platform-based approach was used; in this
instance it integrated the capabilities of IBM Document Management with UiPath robotic
software to produce an automation solution neither technology could have achieved on their
own.
Another equally valuable way robotic process automation can possess qualities of cognitive
technologies is to actually incorporate those technologies into the product itself. By doing so, the
product will deliver advanced capabilities and superior outcomes to the customer in every
automation solution. UiPath has incorporated cognitive attributes into three product features:
Computer Vision: our robots have intelligent eyes to “see” screen elements using contextual
relationships - just as humans do, bringing unrivaled accuracy and precision to automation.
Other robots, blind by comparison to ours with computer vision, are limited to locating screen
elements with various methods- e.g. configured searches or coordinates (think ships using
longitude and latitude).
6. UiPath robots with computer vision give two very significant advantages: implementations are
two to four times faster and production is much more stable, because computer vision instantly
accounts for any screen changes that might occur while processes are running.
Our computer vision advantages are multiplied in Citrix environments, a large part of RPA
automation solutions with ERP systems in scope.
In Citrix environments, our implementation speeds are twenty times faster than those for
providers using blind robots. Why? Because Citrix hides screen elements behind the customer’s
firewall, and lengthy and custom configurations are the only option blind robots have for getting
past this problem.
You can try it out yourself by visiting our Automation Challenge site.
Unattended Automation: how can 24/7/52 robots be optimized if tethered to 8/5/50 people?
They can’t. Which is why UiPath has invested heavily in product innovations that automate the
management of robots. Only by using automation for robots as well as business processes can
RPA become a highly effective automation and operational tool.
7. Machine Intelligence: this is different than machine learning, which enables
software to develop new capabilities as it works.
UiPath’s machine learning enables our robots to recognize process outcome exceptions, then -
rather than simply escalating them for human resolution - the robot takes prescribed actions to
resolve the exception by themselves - avoiding reliance on employees.
While some employee involvement with exception resolution is a necessary part of process
automation, even modest levels of robot success with curative responses has been shown to
noticeably increase performance, particularly during periods of high transaction volumes.
Transforming finance operations with Intelligent
Automation
Traditionally there is such an exigency to maintain a high level of engagement with the
customer, that much of the investment to date has been made in transforming front-end
systems, leaving most of back-office operations to drag along and accumulate heat.
8. The good part, however, is that in this quest for digital transformation, banking
organizations and other financial institutions have managed to establish themselves as
frontrunners in the adoption of smart technology.
Most of the processes banks offshore could be given to smart robots. It will certainly be less
politically treacherous than shifting the jobs overseas as Unions and Works Councils support
the idea of members doing higher-value work.
RPA is a transformational force helping banks differentiate, providing lower costs without
compromising quality, compliance or effectiveness.
Firstly, it’s transformational because it creates a predictable and transparent environment for
data management. Banks are heavily reliant on the many legacy systems and different
applications being used to manage and retrieve information. This tangled web of legacy is
obviously not going to be phased out anytime soon and it places a premium on efficiency.
With RPA none of your IT systems or databases are impacted - they can’t be since nothing
touches them. This is why RPA implementations are faster than IT automation projects, much
less expensive and can create triple digit ROI’s in the first year of operation. One of our financial
clients has achieved over 850% ROI
9. Secondly, it provides better financial performance through insights from data
analytics. RPA is not only a tool for non-invasive extraction of legacy data for wider and deeper
analytics, it provides granular data points for each process it executes. It can identify transaction
patterns and help detect fraud. And what’s best, it’s not an extra service or application you pay
for separately, it comes wit the automation - for free!
Thirdly, it creates a massively improved customer experience by providing services that are
flexible in scope, highly scalable and very efficient. Robots are easily configurable for a wide
range of processes, can be deployed within minutes and are intrinsically more efficient than
humans.
It’s also a type of “RegTech”, offering a direct and indirect regulatory compliance value
proposition.
A direct compliance value proposition would be having a robot join a compliance office as a
virtual member of the staff - but one responsible for all repetitive compliance tasks, freeing up
the human staff members to focus on compliance responsibilities requiring reasoning and
judgement. Additionally, the robot archives log files on every activity - so those repetitive tasks
are well documented for any audits.
This is the answer to the KYC requirements that the FMA, FCA and SEC are beating down your
doors for.
10. Examples of indirect compliance value include:
High compliance levels in RPA transaction processes - for the reason robot actions cannot vary
from those prescribed by the rules and business logic set by the automation software.
Complete audit trails and evidence in RPA transaction processes - because each robot creates
log files documenting all its actions and activities. These log files are archived in a central
repository and can be easily mined by ElasticSearch (our big data analytics platform that is built
right into our robots) for compliance reports, resolution of compliance issues, audit responses
and compliance improvements.
Reduces exposure to remediation actions - the underlying cause of most faultless compliance
issues and remediation actions is errors and omissions caused by human error during manual
data entry. RPA simply eliminates human error.
Too good to be true?
In order to compete in a market that is becoming highly disruptive, banks need intelligent
automation if they want to:
● improve quality and control over operations
● reduce human input errors & improve compliance and governance
● strike a better balance between the front and the back office
● deepen analytical insight
● expand the value-added tasks in people’s roles
● scale operations easily
● stimulate innovation and differentiate
Too good to be true? When you think about it, what business - hearing about modest
investments, quick implementations and compelling benefits - wouldn’t reach the same skeptical
conclusion? Remain skeptical, it shows our ability to think. Then take action and start your
intelligent automation journey. It’s basically now or never.