The rapid change and constant disruption resulting from technological innovation is spreading to the workplace and it is demanding a new approach to the way we perform our daily jobs.
More than ever, businesses are challenged to find new workplace productivity solutions to increase the efficiency and flexibility of their workforce and to drive new opportunities, savings and profits. Understanding how a product is accepted and adopted by users is very important and has received a lot of academic attention resulting in a vast body of literature. The presented work investigates the impact of the interaction of technological, organizational, social and individual factors on the adoption of a ‘Digital Workplace’ solution in a big organization. Most importantly, it analyzes the characteristics of contexts leading to one of three outcomes: No adoption, normal utilization or strategic adoption.
The current study resulted in the productivity tool adoption model that can act as a guidebook to predicting adoption outcomes, by solely looking at a few elements. The binary nature of the framework, by which the presence or absence of a certain element leads to one of three adoption outcomes, makes it easy to apply for either analyzing a project, or predicting a project’s outcomes. It is synthesized as follows:
• Strategic Adoption occurs if the need for the product comes from the company, the product provides the user with improved performance, and the product is aligned with the company strategy. Additionally, and as a second priority, support from CXO, branding and marketing events, and training and communication are important elements.
• Normal Utilization occurs if the need for the product comes from the company and the product provides the user with improved performance.
• If these above two elements are not present, there will be no adoption of the tool implemented.
In the current dissertation, we show that the elements identified in the productivity tool adoption model as essential to adoption success received little attention in the literature. More importantly, the most essential element the need comes from the company has not been mentioned in the literature. Moreover, the third most important element product alignment with the company strategy, although highly researched, has not been studied in relation to user adoption. In fact, the role of the organization is reduced in the literature to the sole function of facilitating, encouraging, or driving the adoption, with little focus on the impact of the benefit that the company gets from the solution implementation on the actual adoption outcomes.
How can Merck use Machine Learning for demand forecasting, and more importantly, how can we enhance the usage of the Machine Learning in practice: those are the questions we will answer. The objective to enhance the usage of the Machine Learning forecast comes with two mains ideas. We want to reallocate the time spent by the demand planners on tasks which cannot be automated yet while improving the sales forecast accuracy.
Our action plan consists of 4 goals:
• Develop an in-house forecasting tool to challenge the existing one and conduct our own studies.
• Develop a new way to perform regression on time series by including information from similar products.
• Improve the stability of the current forecast over forecast cycles by defining a new criterion for the best fit algorithm selection.
• Enhance the usage of the Machine Learning forecast thanks to competitions and increased communication.
• A tool which allows to apply 7 algorithms on products’ history of sales was developed. It showed forecasting performances in line with the existing tool. Moreover it allowed us to conduct the following study.
• The clustered regression, which consists of applying a regression on a cluster of similar products to forecast a single product of the cluster was implemented. With more fine tuning of the parameters, we expect this method to be more powerful in term of performance, which is currently in line with the classic regression.
• The criterion used to define which algorithm is the best fit for a specific time series and thus produces the forecast was changed. The number of test points was increased, and a system of cross-validation tailormade for time series was introduced. We expect up to 44% of reduction of the variability for predictable products.
• A competition on the usage of machine learning forecast was launched. The goal is to have the highest percentage of the recommended portfolio switched to automated forecasting. This percentage over the whole portfolio tripled in only 4 months of competitions.
Manufacturing companies face risks caused by limited supply capacity, which may be the source of out of stock at customer level and subsequently lead to a loss in sales opportunities. Monitoring and tracking, pro-actively, those potential capacity issues at supplier level then appear essential in order to mitigate the risks as fast as possible. This is a necessary condition for multinational companies which aim to get the most efficient Supply Chain, while meeting customer requirements. To this end, this Master’s project develops the line of thoughts of incorporating such capacity tracking tool in one of the world’s leading tobacco companies, Philip Morris International.
Thereafter, we model the production schedule at the supplier level so that the capacity constrain is not violated. With the 6 month forecast shared by PMI to its supplier and the capacity reserved in advance, a cost-effective production schedule is implemented so that any delays in deliveries to PMI, due to a lack of production capacity, are prevented.
The supplier’s lack of production capacity appears as one of the main concerns a manufacturer can think of. Monitoring, thoroughly tracking the supplier’s capacity then comes out as a ”must do” in multinational companies. However, proactive supplier capacity management calls for exchanging relevant data and transparent information between a manufacturer and its supplier, which will without any doubt reduce the well-known bullwhip effect. The usage of a business intelligence and analytics software such as Tableau is helpful for PMI, which aims for more agility in its supply chain. As a matter of fact, this kind of tool, by its automation and clearness, and because it saves times, triggers actions that will hopefully avoid out of stocks at consumer level.
Besides, even though the optimization model designed in this paper is a simplified version of reality, it can be further developed by adding taxes, or depreciation of the machines. The aim here is to give insights about the trade-off between inventory and manufacturing costs. In the printing industry, as in all the industries in which storage does not require specific needs, the inventory cost is marginal. Hence, supplier is willing to merge current and future purchase orders. Our results indicate that the optimum tends to a merge of all possible purchase orders known in advance, as inventory costs are very low in the printing industry.
This case study was carried among the smallest privately owned US based mobile phone manufacturer. The intent was to outline the key strategical elements that allow the conversion of a technological accomplishment into a commercial success within the framework of disrupting a legacy technology seen under the prism of a bounded scope of action.
The notable strategic decision undertook by the company, that have proven to be most impactful on the blueprint towards differentiation were first, a design and engineering of product lines aligned with status quo in smartphone technology and orchestrated around feedback of specific end users, leads to productivity gains on the customer enterprise dimension and a differentiated workforce experience at the user level. Second, in a context of weak technology protection, leveraging access to complementary assets through powerful strategic partnerships with both private and public entities enables a facilitated adoption and commercialization process. Lastly, the implementation and monitoring of a platform ecosystem for managing mobile application ecosystems allowed mutual awareness generation for both hardware and software solutions, and imposed new rules for strategy at the company level, such as the harness of value creating activities over volume.
The response time is one of the most important elements for the firefighters, because the
ability to save lives and rescue people depends on it. Every fire service in the world seek
strategies to decrease this response time and several analyses of what could impact it have
been conducted in the past years. In the meantime, these organizations have been collecting
data about their interventions, yet, a few of them actually use Data Science to develop a datadriven
decision making approach. This thesis is organized along two main axes. In the first axis,
which is purely technical, I presented a complete framework to predict the response times and
discussed the possible usages for this prediction engine. In a second axis that is more
managerial, I presented the barriers for bringing Data Science into the management of fire
services and the changes required in the organization. The project was implemented during a
collaboration between the Computer Research Institute of Montréal (CRIM), the City of
Montréal and the fire department of Montréal (SIM).
The predictive tool was tested using the data of the fire department of Montréal and showed
significant improvements over the baseline models and current systems in use in Montréal.
Because the tool can use external dynamic factors such as the weather, I recommended to use
it to run advanced simulations and better plan the strategic response of the firefighters. Then, I
showed that the barriers for organizations were mainly data quality and consistency, and the
lack of trust from managers towards the Artificial Intelligence technology. To overcome this
change, organizations should establish a clear data governance strategy and provide education
and training to managers. Finally, I recommended that the city of Montréal communicate about
this successful innovation to drive further collaborations and improvements in smart
Freight forwarders ops team replies to requests for quotation all day long. This process is
fastidious, time consuming and repetitive.
The goal is to optimize the reply process by analysing the behaviour of employees taking care of
this. This includes understanding the patterns, the bottlenecks and repetitive actions.
To achieve this, we used customer-oriented tools and methods (funnel conversion, heatmaps,
surveys, etc.) applied on the back-office system, then proposed and implement solutions in
order to reduce as well as the waiting time for the customer and the time spent by the ops
team replying to the requests for quotation.
We set up a process to automatically see the pain points in the process. We then implemented
various ameliorations based on those results.
We reduced by 25% the effective time spent by an operations agent to answer a request for
quotation (1.5 hours, +/- 10 minutes)
We also also reduced the time spent tracking the ships by half (1 hour per shipment became 30
minutes), which is a huge improvement.
Lean Manufacturing is widely recognized and accepted in the industrial environment. The philosophy of achieving more by doing less is central to current ways of working in order to focus on value-adding activities. However, a new paradigm called Industry 4.0, or the Fourth Industrial Revolution has recently emerged in the industrial sector bringing with it an integration of machines, products, individuals and new technologies into a seamlessly connected network over the entire value chain. Now, a question arises if, and how, Industry 4.0 could revolutionize or simply lead to an evolution of the current Lean practices. Through a practical example, this thesis aims to illustrate the outcome of incorporating a novel technology into a standard Lean protocol for waste (defect) management. Ultimately, the question of to what extent novel technologies brought by the Industry 4.0 mega-trend can enable Lean Manufacturing will be answered. This is the result of 6 months of work at Philip Morris International, in the Engineering department of Neuchatel, Switzerland.
The project delivered a new metric in the form a KPI in order to track tobacco waste generated during production over time. Full transparency and visibility of performance on the shop floor is thereby ensured at a sufficient level of granularity to easily identify the source of waste and to eliminate it in shorter periods of time. Through standardized guidelines, the results drawn are leverageable and meaningful as affiliates can now compare their results and track changes. The pilot for this Lean protocol was successful at providing the data sufficient to prove the added value of the newly introduced KPI. Nonetheless, it also helped recognized important limitations to extend the KPI full scale. Such limitations were identified as logistic flow complexification, unavailable infrastructure and additional human resources required.
As a result, the project was complemented by an innovative proposal called the SmartBin. This field sensor device captures the fill-level of the bin on the shop floor and communicates the information in real time onto an IIoT (Industrial Internet of Things) platform. Through automation, the additional activities initially required to collect and weigh the bin and to record the data were successfully eliminated. Furthermore, greater data accuracy was ensured and human error prevented. The entire protocol could be made more flexible and streamlined.
One major finding of this thesis work is that in order to achieve more, often more resources, time and efforts are required. With the adoption of enabler technologies, the same target of achieving more can be met, but with less additional activities. It can therefore be argued that such extra resources mobilized are in fact wasted as they could easily be prevented.
The traditional commodity trading industry is at a strategic crossroad. On one hand, external factors increase the pressure on margins and transparency. On the other hand, new technology is entering this industry with a disruptive character. The goal of this project is to assess the performance of operations with data driven metrics in a commodity trading company and displaying areas of investments or inefficiencies for the implementation of such technologies.
The project consists of two different parts. First, the approach to derivate operational metrics by analysing the internal and external environment. Secondly, showing the approach of implementing such metrics and giving recommendations.
Dior wanted to rationalize and to improve its non-commercial articles supply chain management.
Non-commercial articles (internally called TOPS1) are contributing to convey a consistent brand image.
These articles (VM, Packaging, Product environment, CRM gift, Store supplies…) are fully part of the
customer experience. Main objectives are to facilitate TOPS management at the store level, to gain
visibility and transparency across actors involved, to realign the whole supply chain and to offer a
homogeneous experience to the customer.
The aim of this thesis is by no means to be exhaustive but to present the comprehensive approach
that I took to consider the NCAs2 supply chain problematic. In a first part we will outline the project
framework and the initial situation. To have a better insight we will analyze different logistics flows as
well as different Dior processes all along TOPS lifecycle. After having conducted an
important number of interviews, turns out that most of the internal processes are nor efficient neither
controlled. In addition to a wide variety of NCAs, management practices are very heterogeneous
(depending on division, zone, store, items). To end with initial situation, we will highlight root causes of
inefficiency (adding up some tangible example) besides an investigation around the bullwhip effect
(causes and how to avert it). This bullwhip effect interest was kindled due to an important variability
observed in the demand. Then, in order to realign Dior’s supply chain, it’s necessary to switch from a
flexible and responsive supply chain to an efficient supply chain. For this purpose, we want to change the
transportation method from air-freight to ocean-freight implying the need for logistics facilities (local
warehouse) in each strategic zone for Dior. At this point, Dior hasn’t the logistics means in each zone,
thus it was decided to outsource the supply chain with a third-party logistic (3PL) which is actually a
partner. With that said, the information flow and the ordering process should be completely redesigned.
An efficient, dynamic and flexible Web Order Tool is required for information flow management across
the supply chain. The design brief containing all the different requirements was crucial to give both Dior
and the partner a clear vision of IT developments requisite. Once the new supply chain is correctly shape
with suppliers, 3PL, Zone and stores, the last part of the thesis is developing the different supply chain
best practices always good to set up in a context of supply chain rationalization and optimization. Hence,
we will review partners alignment and incentives, SKU rationalization, Forecasting methods and
1. The overall project is called On Top to represent what comes on top of the product (a TOP is synonym of Non-
2. NCA stands for Non-Commercial Articles
Philip Morris International (PMI) had set out new rules and introduced a Quality Management System
(QMS) due to its new strategy: the development of Reduced Risk Products (RRPs) that will eventually
replace its conventional cigarettes. Like all companies that need to follow internal and external
regulations, Philip Morris has a Quality Management System (QMS): set of documents that summarize
how employees must perform critical activities in order to ensure compliance with regulations and
standards. The Information Services (IS) Information Protection Governance (IPG) department decided
to carry out this project to understand the amount and cause of the reluctance to change to the new
QMS, and accordingly, developed the solution. The project lasted for a period of six months. Interviews
were carried out and the results were analyzed: first by a chi-square analysis, and second by using a
design thinking approach.
The results of the chi-square analysis showed that usage frequency depended on the location of the
employees. Other demographic factors did not specifically relate to usage frequency or pain points. The
design thinking approach allowed to further understand the employee pains and resulted in the
following persona and Point-of-View (POV): “Andi needs a better way to navigate through the
documents in order to feel more knowledgeable about where each document origins and how it is
related to his work”. Ideas were brainstormed and a solution for the POV above was developed using
Tableau® Software. The solution consisted of a hierarchy of the QMS documents and linkages from the
software to the original documents. This allowed users to have a visual understanding of the origin of
each document and how it is related to their jobs. It also helped users to easily navigate through the
documents and search for the document they are looking for. The solution is now under testing in the
company and will be used to guide the employees through the different documents in the QMS.
Viasat is a global communications company based in Carlsbad (CA), which provides residential
internet via satellite among other services. Viasat is developing a geomarketing tool intended to
optimize marketing expenditure by targeting campaigns in specific regions where the
subscriber acquisition opportunity is largest.
This tool uses datasets from public and private sources, at different geographical scales (from
provinces to rooftop-level), ranging from availability of competing internet services or
purchasing power to Airbnb prices.
Although Viasat has been an internet provider for over a decade in the United States, it is just
entering the European market now through a joint venture (JV), Euro Broadband Retail Sarl,
with French satellite operator Eutelsat. For this reason, even though the US market is pretty
well understood and the tool is fine tuned to identify potential subscribers in the US, a lot of
work needs to be done to understand what are the drivers of satellite broadband take-up in
Europe. With support from Viasat’s data analytics team based in Denver, I am leading this effort
out of the JV’s headquarters in Lausanne.
Additionally, as this tool has been developed by Viasat, for IP reasons only Viasat employees
working for the JV have access to the source code. In my new role I would be also be the point
of contact between the JV and all the distributors and marketing agencies across Europe for all
Using 880 sales in Norway as an example, a statistical analysis of the sales data by location
revealed a negative effect of fixed broadband coverage of estimated sales, but a positive effect
for fixed wireless broadband coverage, suggesting that Viasat’s residential satellite broadband
product can thrive in rural areas despite the availability of other competing services based on
alternative technologies. The analysis did not support the hypothesis that affordability is a main
driver of take-up of this service, unlike in some other markets.
Due to the limited sample size, perhaps the most valuable learnings from this analysis are the
identification of additions and modifications of the current process. These include adding
measures of customer satisfaction, more details about sales (plan mix, date), marketing
campaigns or population density. Once the monthly sales volumes pick up and new plans and
promotions are introduced, panel data could be used to analyse performance. Finally, a need
for automation in the database update process and geographical nomenclature harmonization
among business units was identified.
According to the United Nations, more and more people are living in urban areas with more than 75% of the population in developed countries. It implies the need of a new strategy for public transports. In the progressive definition of this new urban mobility, the subject of the information to travelers must be placed at the heart of the debates in the same way as security or production. Indeed, traveler information is one of the key elements to improve customer experience and to develop public transports. This report tackles this issue by analyzing some case studies of different European cities and defines key elements for an effective traveler information strategy for public transports in dense area.
Through the analysis of the different traveler information systems in European cities, we have been able to understand the impact of centralized management and new technologies on the system’s performance in dense area. This observation was reinforced by the study of the internal dysfunctions of the production organization in Paris public transports. This quantitative and qualitative study revealed many weaknesses and improvements, which are essential for a good management of Traveler Information.
Three pillars are therefore fundamental for a Traveler Information strategy for public transports in dense area. First of all, it is essential to set up a centralized management for Traveler Information in order to gain reactivity and consistency in the production and distribution of information. Secondly, disturbed situations must be scripted to increase anticipation and have ready-made and well-adapted strategies. Finally, using new technologies is crucial for the system’s performance because it improves the knowledge of the customer and facilitates the personalization of information.
This master project has been carried out in a private equity fund specialised in brownfield remediation. The fund acquires environmentally impaired sites, remediates them using the most environmentally sound techniques and sells the repositioned property to third parties at a premium. Such funds, seeking to have an environmental and social impact alongside a financial return, are more commonly called impact investing funds. However, impact funds often report less favourable risk/return ratios in comparison to traditional funds not seeking impact. Under those circumstances, it is important to develop a tailored impact measurement methodology to show limited partners that the less favourable risk/return ratio is compensated by a positive and quantifiable environmental and social impact. The main objective of this work is to construct an impact measurement model demonstrating the environmental and social return of the fund.
The construction of an impact measurement model, tailored to the fund’s characteristics, was made possible by an active involvement in the fund’s activities, site visits, in-‐depth literature review of the topic, and numerous meetings with the employees. As suggested by the literature review, a theory of change framework was applied to select the most pertinent indicators. Subsequently, these indicators were categorized in six dimensions reflecting the fund’s core activities. Finally, the corresponding data was gathered and the extra-‐financial impact of every project was evaluated. Recommendations include to use this model to research which environmental and social aspects make a project more successful and apply the model to other impact funds, taking into account their business model features.
Marketing has traditionally aimed at delivering the right message to the right consumer at the right time. In the age of digital advertising and ubiquitous Internet, consumers are reachable anywhere 24hours a day and 7 days a week and are submerged by advertising messages. In such an environment, to win the right to reach a consumer and be noticed by them, brands need to be relevant to consumers. They need to deliver tailored, one-to-one messages based on consumer’s context. The difficulty behind delivering contextual marketing messages is twofold: 1) marketers need to upgrade their technical tools to put in place a programmatic engine capable of delivering personalized communications at scale and 2) marketers need to understand what constitutes the context and when their customers are most receptive to engage with brands. In this work, we explore the difficulties and opportunities a consumer goods company – Procter & Gamble – faces while developing its digital advertising model to include contextual targeting capabilities.
The practical contributions of this work in IT Data Science department to P&G can be summarized as follows. 1) Assessment of how programmatic media fits into the digital media strategy of P&G and its key challenges, followed by a development and deployment of a new automated service tool “Automated Reporting of Programmatic Campaign Alerts” as a response to some of the key challenges. The tool aims at automating quality checks of programmatic campaigns and thus facilitating the work of media agencies and regional media managers. Campaign quality KPIs have been defined in-line with the digital media teams and tool requirements have been validated with end-users. The design of an automated solution that consists of gathering KPI information via big data platforms, generation of PDF reports and emailing them to end-users is described in the report. 2) Generation of insights about the impact of contextual factors (incl. time of the day, day of the week, geolocation in a rural or urban environment, temperature or demographics) on consumer’s engagement with P&G advertisements based on the study of click-through rates (CTR) of past campaigns using statistical inference testing. Elements of machine learning have been applied to gather hints about the relative importance of contextual factors. For the first time, data gathered through programmatic campaigns has been combined with additional data sources to study campaign impressions’ context and its impact on CTR. Findings from this study constitute a solid basis for the launch of a contextual marketing pilot to expand P&G digital marketing capabilities.
The master thesis comprised of two main projects, both falling under the umbrella of process improvement and optimization.
The first sub-project comprised of a rigorous digitization level analysis on 9 machines owned by PMI. The first objective of the project was to obtain figures and indices reflecting the level of digitization each machine has, and based on that, perform an interchangeable comparison which allows to identify any benchmarking points between the machines. The second objective was to propose pilot projects on adding digital control points to areas where digital monitoring is missing, in order to be carried out in the future for the purpose of process improvement.
The second sub-project, also falling under process optimization, comprised of developing a simulation methodology to be applied on several simulation-based tasks. The methodology is based on three pillars namely: data extraction, data manipulation, and model simulation. The main application on which the methodology was applied was to predict the line efficiency change upon the integration of a product buffer machine in it. The main challenge was to properly simulate and mimic each machine’s run/stop time (i.e. MTBF and MTTR profiles). The application was finally followed by a detailed financial analysis, shedding light on the cost savings and avoidance balance of investing in the product buffer equipment.
Digitization Project: The results for this project were tremendous and truly interesting for the company. Despite the fact that it was evident that the machines contain a substantial part of digitally controlled process points and steps, there still exists a lot of room for improvement. This was reflected by the pilot projects lists that was created for each platform. A sum of 60 projects were proposed, and it was clear that the packaging machines needed the least improvement in terms of digital integration.
Simulation Project: After carrying out the financial analysis, which was heavily based on the results obtained from the simulation methodology, it was concluded that:
– Integrating the buffer on 31 out of the 36 GIMA packing lines, specifically 12 500-ppm lines and 19 400-ppm lines leads to an increased production capacity equivalent to that of one whole Link-up.
– By this, and based on the consolidation of the costs and savings incurred by the investment, around €12 million will be saved by year 5, and € 35 million will be saved as of the year 15.
Vasant Narasimhan, CEO of Novartis since February 2018, announced his plans to “lead the digital revolution in pharma and reimagine Novartis as a medicines and data science company”. This is a very ambitious and fundamental change to the company’s DNA and questions about the best strategies to conduct the transformation arise.
Indeed, on the one hand, Data Science is an emerging field with highly uncertain applications and potentials, but on the other, developing new drugs is complicated, lengthy and costly, and accelerating clinical trials could save the industry billions. How best, then, can a pharmaceutical company such as Novartis approach this topic to maximize value?
This master thesis approaches the question through 52 semi-structured interviews with actors in the pharmaceutical industry.
To articulate an answer, the first part of this work studies the exploration versus exploitation trade-off, and known results of the Multi-Armed Bandit Problem are discussed. The value of experimentation and staged financing in these times of uncertainty are then considered, and the model proposed by Nanda2016 is studied. Finally, we introduce the make or buy dilemma.
The second part of this paper first studies the processes required to market new commercial drugs, focusing on the clinical trials phase.
Afterward, we present the findings obtained from conducting 52 semi-structured interviews with actors in the pharma industry. We found that while adoption was timid, a majority of the respondents were familiar with the technology and that having a workforce educated on the topic was positively correlated with adoption. We provide the example of the Electronic Data Capture (EDC) system to highlight that the industry is slow-moving.
Using the output of the interviews, we identified barriers responsible for the slow adoption of AI: Regulations, Data/Business units operating in silos, a strong risk-averse culture, a lack of infrastructure, weak management, a lack of talents and general technophobia, unclear use cases, unknown existing solutions, unproven technology, and high costs. We also unearthed concrete opportunities for AI and digital in the clinical development process: estimate the feasibility of a protocol, improve patient recruitment by easing the way patients find trials and trial managers find patients, virtual trials, increase staff productivity and identify bad sites
While a significant amount of value lies in citations from the interviews, a key takeaway is that the technology is there, and rather than a technical issue, it is more a change management challenge, as breaking free from the old silos and risk-averse culture may require a different culture and refreshing the workforce.
Traditional A/B testing consists of splitting leads randomly in two groups, send a different
message to each group and check the conversion rate of each message. My thesis consisted of
gathering publically available information to build a lead profile. The obtained profile is used to
split the two groups in a pseudo-random way. The idea of the splitting is to have two groups
with similar conversion rate (if sent the same message). This method gives a more robust way
to pick the better message from the A/B testing.
Analyzing the lead as an individual helped us reduce the error by 20% in comparison to
randomly sampling the groups. Analyzing the social interaction between leads helped reduce
the error by 30%.
The obtained results cannot be generalized as we had only 415 data point.