Using machine

learning
on Vishal
Sikka

Ankit Pande, CFA +91 22 4088 0380

Saurabh Deshpande +91 22 4088 0385
October 23, 2017 Using machine learning on Vishal Sikka
October
Monday,23,October
2017 |23,For further enquiries, contact ankit.pande@quantcapital.co.in
2017 1
 WHAT WE SET OUT TO TRY TO CAPTURE Unstructured segmented data (Dr Sikkas remarks
segmented into (1) generic, (2) strategic, (3)
operations & financial constructs

Pre-processing
Dr Sikka has always been excited about
deploying machine learning and Learning
automation. This is our attempt to logarithm
training set (Q2FY15-Q1FY18)
deploy machine learning onto his public
statements through his time at Infosys. testing set
Dr Sikka has been speaking publicly
regularly since 14th Jun, 2014, the day Features
We set out to deploy text analytics on
of his appointment as CEO Designate of Dr Vishal Sikka using Natural Language
Infosys, taking over as CEO & MD from Processing (NLP) techniques. This gets
Mr S. D. Shibulal on August 1, 2014 to the heart of the matter of how to Sentiment (+ve/-ve)
Prediction/
make a computer respond to common
Classifier
languages. Text analytics needs to be Impact (+ve/-ve)
prepared with the help of what is
called a training dataset, a veritably
large database of inputs and outputs.
The new set of inputs for the machine Pre-processing of data is carried out
A summation can be captured by an analysis of algorithm, namely Dr Sikkas opening with the help of certain procedures
positive sentiment in Dr Sikka's strategic constructs remarks in the earnings calls of Q1 the data is cleansed, numbers and
punctuations are removed, the stems of
As positive sentiments rose, on average, from 28% in FY18, constitute the testing dataset
complex words are generated & the data
Q4FY15 to 70% in Q4FY16, the stock prices reacted
is stored as a data frame. We use three
favorably and sharply, staying in the leading quartiles
machine learning algorithms: (1)
for the period under study. However, through FY17,
Logistic regression, (2) Regression
even as Dr Sikka's sentiments were on song, the
trees, and (3) Random forests. Their
Company's financial performance dipped markedly,
features, pros & cons follow on the
and the stock price reacted negatively, dipping to
next slide
Rs874 in late August after Dr Sikka's resignation
down a full 45% from early June, 2016

October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 2
 SUMMARY VIEW

Q2FY15-Q1FY16 Q2FY16-Q1FY17 Q2FY17-Q1FY18

Summary of NLP output*

Somewhat positive sentiments in Much more positive sentiment in Positive sentiment peaks in
strategic constructs (36%). strategic constructs (52%) as strategic (63%) and operation &
Noticeably non-impactful across also in operations and financial financial constructs (50%).
all constructs constructs (24%). Impactful Impactful positive constructs
constructs show: strategic (12%) rise somewhat in strategic (from
and operations & financial (13%) 12% to 15%), but drop elsewhere

Summary of financial performance

Dollar revenue CQGR of 1.41%, EBIT Dollar revenue CQGR of 2.61%, EBIT margin Dollar revenue CQGR of 1.47%, EBIT
margin of 25.6%, INR PAT CQGR of 1.21% of 25.0%, INR PAT CQGR of 3.21% margin of 24.7%, INR PAT CQGR of 0.34%
Dollar revenue growth guidance of 7-9% Co finished FY17 with CC dollar revenue FY17 guidance revised downwards from
was met at the lower end in FY15, but was growth of 13.3% from initial guidance of 10- 10.5-12.0% to 6.5-8.5%. Annual policy of
bumped up to 10-12% for FY16, marking 12%, but FY17 initial guidance of 11.5-13.5% dividends & buyback worth 70% of FCF;
the first guidance for Dr Sikka is revised down to 10.5-12.0% in Q1 FY17 announcement of US$2.0 bn worth of
dividend/ buyback during FY18
Stock market and analyst reaction^
The stock closed at Rs985 on 30th June, The stock closed at Rs1,171 on 30th June, The stock closed at Rs936 on 30th June,
2015, 25% higher vs. a year earlier 2016, 19% higher vs. a year earlier 2017, 20% lower vs. a year earlier
64 analysts covered the stock, of which at 55 analysts covered the stock, lower by 9 53 analysts cover the stock. 39 had BUY
least 6 analysts (Net) revised their rating y-y. 48 had BUY recommendation (Up by recommendation (down by 9 y-y), 10 had
downward 5 y-y), 4 had HOLD and 3 had SELL HOLD and 4 had SELL recommendation
recommendation FY18/FY19 EPS revisions were
FY16/FY17 EPS revisions were
-3.7%/-4.6% through the year FY17/FY18 EPS revisions were -15.1%/-16.6% through the year
+1.8%/+0.6% through the year
*Average figures in parentheses, Random forest algorithm; ^Source: Bloomberg consensus estimates

October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 3
 GENERIC CONSTRUCTS (547 constructs)

What constitutes generic constructs?

Generic constructs are general comments & basic courtesy on the
conference call or while attending pressers. They also include
personal insights, charity and learning initiatives at the personal
and Company level

What constructs are sentiment-positive? Constructs on topics such

as machine learning, AI, transformation, automation, AI, design
thinking and co-creation, spoken of in a positive vein
What constructs are impact-positive? There can be veritably limited
stock impact of generic constructs as highlighted above. In this
exercise, therefore, we could hardly uncover an impact. Simply put,
none.

Quotes of generic sentiment-positive constructs

Q2FY15: Having studied AI at Stanford and Syracuse and other
places at the feet of the pioneers of this domain, I see the world
fundamentally differently
Q3FY17: The (Infosys) Foundation also renewed its partnership with
Code.org one of the most active CS Education Advocacy organizations
worldwide

Pattern: Concentration of generic sentiment-positive constructs in Q2FY15, H2FY17

In Q2FY15, Dr Sikka was clearly excited about his new home at Infosys, with most positive
constructs being related to automation, AI, machine learning, agile and other new
concepts
In H2FY17, Dr Sikka expressed considerable satisfaction in relation to charity, foundation
work and about Infosys crossing the $10 bn mark in annual revenue

October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 4
 STRATEGIC CONSTRUCTS (929 constructs)

What constitutes strategic constructs?

Strategic constructs are specific comments tied to the outlook on
future outcomes. These include corporate strategy & initiatives
(e.g. Zero Distance), guidance, clients, employee wages &
incentives, deal announcements

What constructs are sentiment-positive? Naming prominent clients,

higher bracket of wage hike, retaining employees, deal pipeline,
momentum in EdgeVerve, Skava (Digital e-commerce platform), IIP
(Infosys Information Platform) & Panaya businesses
What constructs are sentiment-negative? Sluggishness in
transformation initiatives, soft guidance, client-related issues,
other business headwinds
What constructs are impact-positive? Deal conversion, Finacle &
Edge go-lives, large deals & TCV momentum, client bucket expansion

Quote of a strategic sentiment-negative construct

Q4FY17: Any large-scale transformation takes time with the early
years requiring a delicate balance between the dual tasks of
protecting and growing the existing businesses and simultaneously
exploring and entering new frontiers

Pattern: Sentiment-to-impact ratio peaking in Q1FY17 tells its own tale

In Q1 FY17, Dr Sikka was being sentimentally more positive in FY17 than in FY16, without
having a proportionate level of impact. This is captured in the sentiment-to-impact ratio,
which is akin to an RSI indicator in the overbought zone
The strength of the random forest algorithm comes to the fore here. It was able to catch
the correct pattern of a spike in positive sentiment in Q4FY16 and then a drop in Q1FY17

October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 5
 OPERATIONS & FINANCIAL CONSTRUCTS (500 constructs)

What constitutes operations & financial constructs?

Operations & financial constructs are specific comments related to
financials, segment internals, top client growth, margins, pricing,
efficiency (e.g. Zero Bench), attrition & employee engagement,
realizations and productivity

What constructs are sentiment-positive? Strong financial

performance, strong processes leading to productivity
What constructs are sentiment-negative? Soft financial performance,
increase in costs, inefficiency
What constructs are impact-positive? Productivity gains quantified,
above average growth in top clients
What constructs are impact-negative? Termination of projects,
changes in growth patterns

Quote of an impact-negative operations & financials construct

Q3FY16: Blended per capita revenue decreased by 2.5% as a result
of lesser price realization
Q1FY17: Lower revenue from some client engagements, especially in
our Life Sciences, Energy and Utilities portfolios
Pattern: Positive-impact constructs peak through FY16, negatives through FY17
Uniquely, the operations and financial impact constructs are disciplined in their time
period. That is to say, the positive-impact constructs peaking through FY16 and negative
constructs peaking through FY17 actually reflect reality in the respective time periods
Similar to the case in strategic constructs, Dr Sikka was sentimentally more positive in
FY17 without having the proportionate level of impact in operations and financial
constructs. The sentiment-to-impact ratio peaks throughout FY17, perhaps foretelling a
change in pattern of growth, or even a decline in fortunes for the Company in FY18

October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 6
 STUDY FEATURED 3 ESTABLISHED ALGORITHMS

Features, pros and cons

Logistic regression Regression trees Random forests

The target variable is
continuous and tree is
Logistic regression is a used to predict it's value.
statistical method for The algorithm is structured
Creates many regression
analyzing a dataset in which as a sequence of
trees from bootstrapped
there are one or more questions, the answers to
training set data, K-fold cross
independent variables that which determine what the We were able to deploy
Features validation algorithm is
determine an outcome. The next question, if any, the Random forests
available to optimize the
outcome is measured with a should be. The result of algorithm to our
parameters of Random
dichotomous variable (in these questions is a tree- dataset, given the
Forests highly structured and
which there are only two like structure where the
possible outcomes) ends are terminal nodes at stylized nature of Dr
Sikkas remarks. The no.
which point there are no
of structures were also
more questions copious, (>=500 for
(1) Simple model, (1) Good results with (1) Floor to accuracy, each type of construct),
which helped shore up
(2) Good accuracy with negations and complex (2) Handles data with low
accuracy
Pros closely correlated data, inputs correlation with impressive
(3) Just one parameter (2) More customizable to accuracy,
vector suit different types of data (3) Easy to optimize

(1) More complex/more (1) Slow,

(1) Accuracy decreases parameters (2) Requires a lot of RAM,
Cons sharply with correlation in (2) More extensive (slow (3) Prone to failure under
data & requires more RAM), high uncertainities
(3) Ceiling to accuracy

Source: Data Science Central, quant Global Research

October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 7
 Disclaimer
RATINGS AND OTHER DEFINITIONS
STOCK RATING SYSTEM SECTOR RATING SYSTEM
BUY. We expect the s tock to deliver >15% abs olute returns . Overweight. We expect the s ector to relatively outperform the Sens ex.
ACCUMULATE. We expect the s tock to deliver 6-15% abs olute returns . Underweight. We expect the s ector to relatively underperform the Sens ex.
REDUCE. We expect the s tock to deliver +5% to -5% abs olute returns . Neutral. We expect the s ector to relatively perform in line with the Sens ex.
SELL. We expect the s tock to deliver negative abs olute returns of >5%.
Not Rated (NR). We have no inves tment opinion on the s tock.
DISCLOSURES AND DISCLAIMER
quant Broking Private Limited, Mumbai, India (Hereinafter referred to as QBPL) is Registered Member of National Stock Exchange of India Limited and Bombay Stock Exchange Limited. QBPL is also
registered as Depository Participant with NSDL. QBPL is an registered entity as Research Analyst under SEBI (Research Analyst) Regulations, 2014. QBPL or its associates has not been debarred/
suspended by SEBI or any other regulatory authority for accessing /dealing in securities Market.

ANALYST(S) CERTIFICATION
We, Ankit Pande, CFA and Saurabh Deshpande, research analysts, hereby certify that all of the views expressed in this Research Report accurately reflect our personal
views about all of the issuers and their securities and no part of our compensation was, is, or will be directly or indirectly related to the specific recommendations or views
expressed in this research report.
As per the declarations given by the Research Analyst(s) and/or any of his family members do not serve as an officer, director or any way connected to the company/companies mentioned in this
report. Further, he/they has/have not received any compensation neither for investment banking/merchant banking/brokerage services or other products/services nor for benefit of him/them or
third party in connection with research report and also have not managed or co-managed public offering of securities from the company or its associates in the preceding twelve months. The
Research Analysts or company or its associates do not have actual/beneficial ownership of one per cent or more securities of the company. Our entire research professionals are our employees and
are paid a salary and they are not engaged in any market making activity. The Reserch Analyst or company or its associates do not have any financial interest in the company. They do not have any
other material conflict of interest of the Research Analyst or member of which the Research Analyst knows of has reason to know at the time of publication of the research report or at the time of the
public appearance.
The Research Analyst(s) are bound by stringent internal regulations and also legal and statutory requirements of the Securities and Exchange Board of India (hereinafter SEBI) and the analysts
compensation are completely delinked from all the other companies and/or entities of quant Broking Private Limited (QBPL), and have no bearing whatsoever on any recommendation that they have
given in the Research Report.
This Research Report (hereinafter called Report) is meant solely for use by the recipient and is not for circulation. It does not constitute a personal recommendation or take into account the
particular investment objectives, financial situations, or needs of individual clients. Similarly, this document does not have regard to the specific investment objectives, financial
situation/circumstances and the particular needs of any specific person who may receive this document. The securities discussed in this report may not be suitable for all investors. This report is for
information purposes only and this document/material should not be construed as an offer to sell or the solicitation of an offer to buy, purchase or subscribe to any securities, and neither this
document nor anything contained herein shall form the basis of or be relied upon in connection with any contract or commitment whatsoever. This document does not solicit any action based on the
material contained herein.

October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 8
 Disclaimer
This Report has been prepared on the basis of publicly available information, internally developed data and other sources believed by QBPL to be reliable. QBPL or its directors, employees, affiliates
or representatives do not assume any responsibility for, or warrant the accuracy, completeness, adequacy and reliability of such information, opinions/views. While due care has been taken to ensure
that the disclosures and opinions given are fair and reasonable, none of the directors, employees, affiliates or representatives of QBPL shall be liable for any direct, indirect, special, incidental,
consequential, punitive or exemplary damages, including lost profits arising in any way whatsoever from the information/opinions/views contained in this Report. This Report is provided for
assistance only and is not intended to be and must not alone be taken as the basis for an investment decision. The user assumes the entire risk of any use made of this information. Each recipient of
this Report should make such investigation as it deems necessary to arrive at an independent evaluation of an investment in the securities of companies referred to in this report (including the merits
and risks involved), and should consult his own advisors to determine the merits and risks of such investment. Though disseminated to clients simultaneously, not all clients may receive this report at
the same time. The projections and forecasts described in this report were based upon a number of estimates and assumptions and are inherently subject to significant uncertainties and
contingencies. Projections and forecasts are necessarily speculative in nature, and it can be expected that one or more of the estimates on which the projections and forecasts were based will not
materialize or will vary significantly from actual results, and such variances will likely increase over time. All projections and forecasts described in this report have been prepared solely by the authors
of this report independently of the Company. This information is subject to change without any prior notice. QBPL or any of its affiliates/group companies shall not be in any way responsible for any
loss or damage that may arise to any person from any inadvertent error in the information contained in this report. QBPL is committed to providing independent and transparent recommendation to
its clients. Neither QBPL nor any of its affiliates, group companies, directors, employees, agents or representatives shall be liable for any damages whether direct, indirect, special or consequential
including loss of revenue or lost profits that may arise from or in connection with the use of the information. The information given in this report is as of the date of this report and there can be no
assurance that future results or events will be consistent with this information.
The price and value of the investments referred to in this Report and the income from them may go down as well as up, and investors may realize losses on any investments. Past performance is not a
guide for future performance. QBPL does not provide tax advice to its clients, and all investors are strongly advised to consult with their tax advisers regarding taxation aspects of any potential
investment. Opinions expressed are our current opinions as of the date appearing on this Research only. We do not undertake to advise you as to any change of our views expressed in this Report.
QBPL and/or its associates and/or employees and/or Clients may; (a) from time to time, have long or short positions in and buy or sell the investments in/security of Company(ies) mentioned herein
or (b) be engaged in any other transaction involving such investments/securities of company(ies) discussed herein or act as advisor or lender/borrower to such company(ies) these and other activities
of QBPL and its associates or employees may not be construed as potential conflict of interest with respect to any recommendation and related information and opinions. Without limiting any of the
foregoing, in no event shall QBPL and its associates or employees or any third party involved in, or related to computing or compiling the information have any liability for any damages of any kind.

Regn. No.:
CAPITAL MARKET SEBI REGN. NO.: BSE: INB011419430 EQUITY DERIVATIVES SEBI REGN. NO.: BSE: INF011419430
CAPITAL MARKET SEBI REGN. NO.: NSE: INB231419434 FUTURES AND OPTIONS SEBI REGN. NO.: NSE: INF231419434
CURRENCY DERIVATIVES SEBI REGN. NO.: NSE: INE231419434

Member (NSE and BSE)

Depository Participant (DP) NSDL DP ID: IN303614
Website www.qua ntca pi ta l .co.i n
Dri s hti Sha h
Compliance Officer Details:
Tel : (022) 4088 0100; Ema i l : dri s hti .s ha h@qua ntca pi ta l .co.i n

QUANT BROKING PRIVATE LIMITED (CIN: U67110MH2007PTC291657)

Registered Office 612, Maker Chambers IV, Nariman Point, Mumbai 400 021, India | phone 91 22 4088 0100 | fax 91 22 4088 0249-250

October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 9
 Title

Thank you

612, maker chambers IV, nariman point,

mumbai 400 021, india
phone 91 22 4088 0100, 3025 0100
fax 91 22 4088 0198, 3025 0198
October 23, 2017 | For further enquiries, contact ankit.pande@quantcapital.co.in Using machine learning on Vishal Sikka 10