software robots and machine learning - aalto...

12
8.6.2016 1 Software robots and machine learning Petri Karjalainen Twitter: @PetriKarj Linkedin: www.linkedin.com/in/PetriKarj Facebook: www.facebook.com/PetriKarj Confidentiality Intelligent work 2030 “In the new workforce of 2030, the most successful organizations will optimize the usage of all their resources, both human and machine, for competitive advantage. “An increasing portion of your workforce will not be human,” Mr. Prentice said. However, while machines are very good for consistency, performance, predictability, efficiency, and safety; they can’t match humans’ skills in ingenuity, novelty, art, creativity, emotion, and to address variability and provide context.” –Gartner Summary of the top news from Gartner Symposium/ITxpo 2015 http://www.gartner.com/smarterwithgartner/technology-and-business-in-2030/?cm_mmc=Eloqua-_-Email-_-LM_EVT%20EMEA%202016%20ESC27%20E10%20- %20Newsflash5%20%2816.11.15%29%20Non%20attendees-_-0000

Upload: vannhi

Post on 16-Mar-2018

214 views

Category:

Documents


1 download

TRANSCRIPT

8.6.2016

1

Software robotsand machine learning

Petri KarjalainenTwitter: @PetriKarjLinkedin: www.linkedin.com/in/PetriKarjFacebook: www.facebook.com/PetriKarj

Confidentiality

Intelligent work 2030

“In the new workforce of 2030, the most successful organizations will optimizethe usage of all their resources, both human and machine, for competitiveadvantage. “An increasing portion of your workforce will not be human,” Mr.Prentice said. However, while machines are very good for consistency,performance, predictability, efficiency, and safety; they can’t match humans’skills in ingenuity, novelty, art, creativity, emotion, and to address variabilityand provide context.”

–Gartner Summary of the top news from Gartner Symposium/ITxpo 2015

http://www.gartner.com/smarterwithgartner/technology-and-business-in-2030/?cm_mmc=Eloqua-_-Email-_-LM_EVT%20EMEA%202016%20ESC27%20E10%20-%20Newsflash5%20%2816.11.15%29%20Non%20attendees-_-0000

8.6.2016

2

Youtube video here

February 25, 2015 | OpusCapita INTERNAL 3

OpusCapita Software Robots

Confidentiality

Megatrends shaping the future

Cloud, IoT and Mobile internet• Economic impact largely dependent

on establishing suitable ecosystemsas growth drivers

Automation of Knowledge Work• One of key disruptive technologies

within next 10 years• Supported by advanced machine

learning and artificial intelligence

8 June, 2016 4

MobileInternet

Automation ofknowledge work

The Internet ofThings

Cloudtechnology

Advancedrobotics 1.7-4.5

1.7-6.2

2.7-6.2

5.2-6.7

3.7-10.8

Range of sized potential economic impacts

Low High Impact from other potentialapplications (not sized)X-Y

8.6.2016

3

Confidentiality

Gartner: Postmodern ERP era is alreadyhere moving applications to Cloud

8 June, 2016 OpusCapita Internal 5

Factors driving Postmodern ERP era• B2B buyers: Buying behaviour is driving cloud services• SAP: moving to cloud reduces core ERP functionality allowing competitors to bid for cloud solutions• Market: By 2018, at least 30 percent of service-centric companies will move the majority of their

ERP applications to the cloud

Confidentiality

Automation of Knowlede work happensas we speak

8 June, 2016 OpusCapita Internal 6

Automation ofknowledge workhas started withRPA automatingclerical tasks

With statistical analysis andmachine learning the impact ofknowledge work automation will growto the next level

With components and technologiessuch as artificial intelligence,machine learning, natural userinterfaces andbig-data technologies

2014-2015

Economicimpact ofknowledgeworkautomation

2016 2-10 years

8.6.2016

4

Confidentiality

OpusCapita Robots in the Cloud

Image source: http://www.brainstormmarketingproductions.com/

• Software robots are using a computer on behalf of aperson or with a person and they can:

• See and interpret text and pictures.• Move and click the mouse.• Write text and numbers.

• Work 24/7 without breaks.• Use several IT systems just like a human would use

them.• Implementation does not require changes to the

existing IT systems.• Are programmed with ”work shadowing” and defined

logic for exception handling.• Work according to preset rules.• Currently operate as rule based, in the future utilize

statistical analysis, machine learning and artificialintelligence.

Confidentiality

AnalyticsEngine

MachineLearning

Software Robots Architecture

Robots, Win Server 2008 R2Virtual Desktops

RDSMaintenance

Transactionlog

VirtualSupervisor

CustomerApplications

BACKENDengine

ROBOTICnetwork

CUSTOMERnetwork

SECUREconnection

InformationSecurity Policiesare based on ISO

27002requirements,

approved by OCSecurity

Committee.

Robots act asremote userswith own user

accountsRDS

Implementation

8.6.2016

5

Confidentiality

0 €

20 000 €

40 000 €

60 000 €

80 000 €

100 000 €

120 000 €

140 000 €

160 000 €

180 000 €

200 000 €

0 1 2 3 4 5

Manualwork

year

RPA Setup

5 000 - 20 000 €per process

The setup time forinstalling RPA robot in

production environmentvaries between two and

six weeks.

RPA Operating

5 000 - 10 000 €per year

RPA production,monitoring,

licenses and IT.

Manual work

40 000 €per year

Typical FTE cost perannum.

How much savings does RPA offer?Cost of 1 FTE mnual work vs. robotized process

RPA

Typical payback time is less than six months

Confidentiality

> 95 % automation

Old manualprocess: 10-20

minutes

Complex rules

Trigger

List of new employee relationships is sent fromcustomer’s ERP system to customer’s payroll

system nightly.

Automated Activities

Robot fetches the list of new employeerelationships. It navigates in payroll application,

runs dozens of different validations and fixesobvious deviations. Some deviations are

reported to payroll specialists.

Case Example:Validating & Correcting Employee Data

8.6.2016

6

Confidentiality

Other Case Examples in Payroll

Constant Salaries Update• Validating new and updated salaries

• Entering salaries and other monthly payments in payrollapplication

Employment Experience Calculation• Fetching relevant data about employee’s employment history

• Calculating employee’s work experience and updating relevantdata in payroll application

Sick Notes Handling• Ensuring that working hour data is in line with sick note

• Entering sick leave information in payroll application• If required, filling in and sending an application to Kela or sick fund

Confidentiality

Youtube video here

February 25, 2015 | OpusCapita INTERNAL 12

OpusCapita Robots I

8.6.2016

7

Confidentiality

Youtube video here

February 25, 2015 | OpusCapita INTERNAL 13

OpusCapita Robots II

Confidentiality

Macrorecorders– Simple, runs on one

computer– Requires refreshing– Not for complex

operations

Application macros– Automates a single

application, such asExcel

– Available for multipleusers in the organization

Evolution from macros toMachine Learning

RoboticProcessAutomation– Front-end level automation– Any set of applications– Centrally monitored

Machine Learning– Learns from what has been

done in past– Uses e.g. neural network to

build a solution model– Can predict a solution

8.6.2016

8

Confidentiality

Machine Learning application areasinvestigated by OpusCapita• Sales forecasting

– Case: sales forecasting using weather data

• Anomality and fraud detection– OpusCapita is piloting this with Haaga-Helia to develop algorithms for payments deviations

• Cash flow forecasting– OpusCapita is planning to start piloting how to build cash flow forecasts using machile learning

• Purchase Invoice preposting automation– OpusCapita is having a proof of concept study for own invoices and with some selected

customers

Confidentiality

Case: Improve invoice posting quality andreduce lead time with Machine Learning• OpusCapita has tested that a Artificial Intelligence and Machine Learning

solution can prebook the invoice data: Inspector, Approver, Cost Center, G/LAccount

• The solution consists of three modules:1. Training the AI: The prefilled information is based on real accepted invoice data. This data is

used to build and improve the Machine Learning algorithm.2. Prediction by AI: When a new purchase invoice comes to the process, the Machine Learning

model prefills the invoice data to the posting view of OpusCapita Invoices. The probability of theprediction is shown as well.

3. Verification reporting: The change log of posting data is used to measure the quality of theMachine Learning algorithm (rate of no correction needed)

• The solution shorten the lead time, boost efficiency and improve quality

Jan 29, 2016

8.6.2016

9

Confidentiality

Invoice preposting process with artificialintelligence and machine learning

8 June, 2016

Invoiceread-in

Supplierdefinition +

OCI ruleengine

AI/MachineLearning

JSON call forAzure ML Webservice API with

XMLconversion

InvoiceWorkflow

Invoicedata

Logdata

PredictingAccount,

CostCenter,

Inspector,Approver

TrainingMachineLearningAlgorithm

Acceptedposting datais used totrain and

improve thealgorithm

VerificationReporting

Automationquality

measuredby

monitoringthe numberof human

correctionsneeded

Self learning loop

All thecustomer

invoices areread in to

OCI

Known rulesapplied andinvoices with

PurchaseOrders

separated.

Rest of theinvoices

assigned forArtificial

Intelligence

Confidentiality

Invoice workflow approval user interface withMachine Learning based preposting

Jan 29, 2016 18

Invoice is sent straight forthe best known inspector

Account, Cost Center andApprover are prefilled withthe estimated probability

Inspector verifies thepredictions before

sending the invoice to theApprover

8.6.2016

10

Confidentiality

Prediction accuracy verification

• The quality of the Machine Learning algorithm is tracked by analyzing the change loginformation: prefilled vs. approved invoice data

• % =

Jan 29, 2016 19

7580

8590

85 8792

0102030405060708090

100

Intelligent invoicing quality (%)

week1 week2 week3 week4 week5 week6 week7

Confidentiality

What does machine learning look like ?

20

8.6.2016

11

Confidentiality

Learninig machine learning:”Read the book and just start using”

Commercial platforms• Microsoft• Microsoft seems to have the best functionality

and also be the easiest to learnand best documented.

Open source platforms• Weka• Not so easy to start using but completely open source

Confidentiality

Summary

• RPA is mainstream and suitable for processes with simpleclearly defined rules

• Machine Learning is available for more complex decisionsand it works already today

• Processes where Machine Learning could be used areeasy to identify with these three rules:1. Humans are processing the data with complex rules, too complex for programmer to encode2. Dataset exists as a result of human processing so that machine learning can be used to build a

model3. Users should still review the data that machinelearning has prefilled but processing time and

quality improves drasticly

8.6.2016

12

Petri KarjalainenTwitter: @PetriKarjLinkedin: www.linkedin.com/in/PetriKarjFacebook: www.facebook.com/PetriKarj

Confidentiality

Buzzwords to follow for future trends

8 June, 2016

Social Mediafor business

BlockchainAI

Machine Learning

Consumer UX