2.1.1 Define Business Problem or Opportunity

Although the tasks in business data analytics are iterative and not sequential, defining the business problem or opportunity is often the first step performed in any business data analytics initiative. This task is where those with strong business analysis skills can assist with the work.

Often, when analytics engagements start, the task of identifying a problem is not given enough attention or the right stakeholders are not identified. There is often an urgency to see results from the investment in analytic initiatives. This can create a tendency to jump into a solution focus rather than devote enough attention to identifying to problem and engaging the right stakeholders. This can lead to problems not being analyzed in sufficient detail and resulting in a misdiagnosis of the business problem.

Business data analytics involves analysts performing business problem discovery in parallel with the task of identifying and understanding the right stakeholders. An analyst may facilitate discussions with stakeholders to elicit, observe, and analyze through a process of continuous discovery of any and all relevant information that will help the team understand the context of the situation.

Sometimes the organization is experiencing a problem that business data analytics can help solve, such as understanding why there is a sudden decrease in internet sales. In other situations, the organization may be interested in using business data analytics to uncover opportunities—as in the case of a manufacturing company looking to collect maintenance and performance data on its machinery to determine how to predict and avoid equipment outages. In either scenario, analysts use various business analysis elicitation and problem analysis techniques to obtain the necessary information required to define the business problem or opportunity that analytics might address.

It is important to note the outcome of this Define Business Problem or Opportunity involves identifying the business problem that may lead to one or more research questions/problems. Then, further analysis is conducted to formulate the right analytics questions from the business problem or opportunity under consideration.

Business problem or opportunity is usually a higher- level description than a research question/problem.

When defining the business problem or opportunity, analysts use several elicitation techniques such as interviews, job shadowing, surveys, and workshops. Business and organizational knowledge are useful when facilitating discussions.

2.1.2 Identify and Understand the Stakeholders 

Identifying and understanding stakeholders allows analysts to actively engage and collaborate with the variety of stakeholders involved in an analytics initiative. Each stakeholder group:

• articulates different needs and objectives,
• poses different types of research questions,
• is interested in different volumes and timings of analytics results,
• holds different skillsets for interpreting those results, and
• possesses different levels of education in, and experience with, analytics.

Understanding the unique characteristics of each stakeholder group increases the analysts' effectiveness with each group. Before an initiative starts, analysts seek to answer the following questions:

• Who are the stakeholders?
• What is their level of knowledge about analytics?
• What aspect of the project is of interest to them?
• What communication methods and techniques are appropriate?
• When should stakeholders be communicated to?

It is critical to understand the custodians and consumers of data to identify the relevant stakeholders for an analytics initiative.

When identifying and understanding stakeholders, analysts use techniques such as brainstorming, interviews, or reviewing process flows and organizational charts.

In an analytics initiative, it is critical to have a data view of the business problem where analysts try to understand who:

• is creating the relevant information,
• is exposed to the data created within the organization, and
• are the decision-makers being influenced by the insights derived from the data.

Analysts use models such as stakeholder matrices and onion diagrams to depict aspects of their stakeholders. Analysts also create models to show how the organization’s strategic goals relate to the organizational goals and objectives and the stakeholders or stakeholder groups impacted. They create or review personas to gain a deeper understanding of stakeholders. Facilitation and communication skills, along with knowledge about the business or specific organizations, help analysts perform stakeholder identification.

2.1.3 Assess Current State

Business data analytics is used to enable organizations to make informed decisions. Understanding the current state of the organization or context of the proposed change is fundamental to informed decision-making.

Information obtained from a current state assessment provides important context so that the results of data analysis can be better interpreted.

Analyzing the current state involves understanding the business need and how it relates to the way the organization currently functions. The results of the current state analysis set a baseline and context for making a change.

Whether discussing the changes associated with the implementation of a new customer relationship management (CRM) system or the process changes proposed after gaining insightful information from the results of a business data analytics effort, analyzing the current state is an important task.

A current state assessment can include understanding the business value chain or how data and information flow throughout the organization. From a data analytics perspective, the analysis of the current state involves determining what the existing data is pointing towards. This can be in the form of insights gained from previous analytics engagements or through the exploration of existing data using statistical or mathematical models or intuitions.

Conducting a current state assessment involves understanding organizational capabilities, resources, and business processes, in order to fully understand the business problem and derive research questions from it. All the tools, techniques, and models used are evidence-driven. For example, a business leader simply stating that there has been a reduction of internet sales due to lack of customer engagement does not conclusively link reduced sales to low customer engagement. Analysts look for concrete evidence from data and analysis to substantiate that hypothesis.

The analyst may uncover insights such as whether the organization has an appetite for analytics, the budget, and the expertise to perform the work. They necessarily become knowledgeable in the business domain and understand trends and evolving business models.

When conducting a current state assessment, analysts use business model canvas, organizational, scope, and process modelling to elicit, analyze, and visually depict the current state of the organization. Conceptual and systems thinking, along with business acumen and solution knowledge, help analysts understand and communicate the current state.

Example of Current State Analysis to Refine Research Questions Using current state analysis to refine research questions can be demonstrated by reviewing some challenges of forecasting for a typical newsvendor problem. A family-owned company that specializes in making world-class bicycles and parts has gone through a recent leadership change. Like many organizations that conduct a yearly forecast, the new leader is also faced with the most critical aspect of the job— forecasting and committing to fulfilling the total product demand, thereby determining a significant share of the next year's success. Forecasting was a pressing issue as the product portfolios had continued to grow even as product life cycles, along with customer patience regarding delivery, were becoming shorter. As always, the company’s order commitment to its suppliers had to be made six months before the start of the season in September, without having any early indication of demand. The new leader wanted to adopt new best practices, knowledge, and processes that included new ways of forecasting the demand. In this case, the business need is to accurately forecast the demand for the products ahead of time, which corresponds to a research question on discovering the factors that influence the forecast. The accuracy of the forecast would determine the strategy for the company for the entire year and impact decisions across the value chain such as procurement, logistics, and marketing. A current state analysis, in the data context, usually involves the study of the business model, supply-chain analysis, analysis of the existing forecasting process, and analysis of the utility of the data available for forecasting. The analysis may reveal many constraints, insights, and assumptions that affect the nature of the research question. If the current state analysis was not performed, the research question would only involve assessing the current forecasting model. An accurate current state analysis may reveal that the business model could be revised. Instead of using the same supplier for procuring bicycle parts, different local suppliers could be used, and the bicycles could be assembled locally. This would change the current research question from trying to improve the forecasting model to researching the factors that would maintain the quality, pricing, and sales. Similarly, if the current state analysis of historical data revealed a clear trend of seasonal demand during longer holidays, the research question would involve a study of seasonal patterns that could be used to upgrade the forecasting model. These two examples, a change in the business model, or a change in the forecasting approach, are clear results of current state analysis that refine the original research problem to new or re-framed research problems that are more relevant.

2.1.4 Define Future State

Defining the future state creates a vision of the desired outcome of the change. Defining success for a business data analytics initiative is as important as any other change initiative. Defining the future state includes ensuring:

• the future state is clearly defined and understandable,
• that it is achievable with the resources available,
• that key stakeholders have a shared vision, developed by consensus of the outcome being sought, and
• measurable objectives are established to ensure the desired vision is met.

According to A Guide to the Business Analysis Body of Knowledge® (BABOK® Guide) version 3, purposeful change includes a definition of success.

To establish measurable objectives, analysts facilitate discussions between stakeholders to determine the types of metrics to consider. Working collaboratively, decision-makers select the most appropriate measures to assess using business data analytics. These measures may be a combination of strategic and operational key performance indicators (KPIs). Some KPIs may focus on assessing performance for a specific geography or a target audience. There may be industry-specific metrics such as average revenue per user (ARPU), which is used in telecom, or “store footfall” which is used in retail to count customers visiting the store.

Another important aspect of defining the future state is establishing the scope for the analytics effort. Establishing the scope involves understanding which areas of the organization are participating in the analytics effort and determining what stakeholders have questions to raise and information to provide.

A future state, concerning an analytics initiative, could also include setting a vision about the length and breadth of analytics capabilities. For example, tracking more KPIs, increasing the frequency of reports being generated from monthly to daily/weekly, automating reporting functionality, or having data available in real-time. Apart from descriptive objectives like tracking KPIs on the past data, predictive and prescriptive analytics may involve certain anticipated changes to business processes that drive multiple change initiatives.

The future state of an analytics initiative evolves throughout the life cycle of the engagement. Analysts manage and record the changes to future state.

Given the potential evolution of the vision, analysts are challenged in describing the changes reflected in the current understanding of the future state. Like most other activities in an analytics initiative, defining the future state is continuous and iterative.

The desired output from defining the future state is a clear understanding of the business objectives and the value the business is seeking to obtain from the analytics effort.

Analysts use metrics, KPIs, and different models to visually communicate the future state. This includes scope models to understand boundaries and stakeholder maps to identify those who might be impacted by this work.

Conceptual thinking skills help analysts understand the big picture and provide the context for the analytics work. Interaction skills, communication skills, analytical thinking, and problem-solving skills are useful when leading discussions to identify metrics and establish objectives.

Real-World Problem in Defining the Future State for a Predictive Classification Problem Detecting fraud is a perennial problem in multiple industries such as banking, finance, insurance, and telecom. It is a typical use case in analytics called binary classification. That is simply saying a particular transaction based on the analytics model is classified as fraud or not. For such a problem, the measure of success is governed by the business context and the identified business problem which the analysts formulate while defining the future state. There are some standard measures such as precision, recall, specificity, or accuracy that are commonly used for such types of problems described by the following formulas: Precision = TP/(TP+FP) Recall = TP/(TP+FN) Specificity=TN/(FP+TN) Accuracy=(TP+TN)/(TP+FP+TN+FN) Where: TP = True Positive. The number of transactions predicted as fraud which are actually fraudulent. FP = False Positive. The number of transactions predicted as fraud but are not fraudulent. TN= True Negative. The number of transactions predicted as not fraud and are not actually fraudulent. FN = False Negative. The number of transactions predicted as not fraud but are actually fraudulent. Consider a scenario where a business wants to detect fraudulent transactions for credit cards. There are many factors (transaction time, location, and amount) which influence a transaction to be classified as fraudulent. When this type of fraud is detected algorithmically, there is a possibility that many transactions will be misclassified. A transaction may be predicted as fraud but in reality, it may be a valid transaction (a false positive). Similarly, a transaction can also be misclassified as a false negative. Depending on what the business wants to achieve, the criteria of success for the fraud detection analytics model may change. If the business wants to detect as much fraud as possible, the analytics model is adjusted so that the maximum number of true positives are detected. But, this also increases the chances of false positives. If the business stakeholders take a conscious decision that false positives are not a concern then the analytics model may only focus on precision as the most appropriate metric to maximize. On the other hand, the business may want to define success as a measure of the actual cost to the company. The actual cost would be a trade-off between cost saved by predicting fraudulent transactions versus cost incurred for incorrectly predicting a fraud (cost of false positives and false negatives). In this case, precision will not be the right metric to pursue. The key takeaway for analysts from this discussion is depending upon the business context the success criteria of an analytics initiative changes. The analyst must be able to articulate business context to the analytics team and similarly, explain to the business stakeholders any mathematically complex measure in simple business terms.

2.1.5 Formulate Research Questions

Before any of the detailed analytics work is performed, the stakeholders formulate the question that the analytics will answer. The research inquiries are derived from the business needs. The business need is problem or opportunity of strategic or tactical importance to be addressed.

For example, if the business need is to improve the customer experience of a retail store, the questions will be:

1. What are the factors that influence customer experience? (Descriptive analytics)
2. What are the measures for evaluating customer experience? (Descriptive analytics)
3. How do you classify individual transactions on the retail side as a positive or negative experience? (Predictive analytics)
4. Will customer experience improve by adding a new feature such as a pay wallet? (Prescriptive analytics)

Business needs can lead to different solutions and approaches which may or may not involve analytics initiatives. One or more of the analytics problems or opportunities may lead to one or more analytics initiatives where the research questions are further refined until they can be identified using a measurable success standard.

Formulating the research question involves facilitating discussions to identify the different problems to be explored, specifying the questions in an easily understood language, and bringing the team to a consensus as to the best set of analytics questions to answer.

Analysts require the skills to identify the right problem or opportunity and to focus the team on the right question to ensure the analytics work is guided properly. Discussions move beyond brainstorming a list of ideas to producing a concrete list of specific analytics questions the team believes are worth pursuing. On occasion, the team may need to identify what data are available before determining which ideas are achievable with analytics. The question, once formed, guides the scope and drives the activities of the analytics team.

The results of the analysis obtained when defining the business problem or opportunity, analyzing the current state, and defining the future state provides context when formulating the analytics questions. The analytics team, including business stakeholders, may start with a long list of questions and require ongoing collaboration to reduce the list identifying the highest valued questions to use. Technical resources or the analyst, based on their understanding of the data and the business problem or opportunity, may suggest an analytics problem that could be explored.

Good analytics questions are clearly stated and do not use technical language. The questions are reviewed with all stakeholders to ensure consensus that clearly articulates what the organization is looking to answer through analytics. In the Perform Data Analysis task (for more information, see 2.3.4 Perform Data Analysis), the data scientist/analytics experts restate the analytics questions using more mathematical language.

There are situations where it is more efficient for an analytics team to address a group of questions for multiple initiatives, rather than individual initiatives asking one question at a time.

When formulating research questions, analysts utilize a variety of elicitation techniques to facilitate discussions with stakeholders, decision models to help the team reach consensus, and templates to guide the development of the question. Strong facilitation, leadership and negotiation skills are useful when facilitating consensus among stakeholders.

2.1.6 Plan Business Data Analytics Approach

Planning the Business Data Analytics Approach defines how analytics work will be performed. When planning a business data analytics approach, analysts:

• determine the capabilities and capacity of the organization to perform analytics so the team understands what is realistically possible,
• identify “quick wins” versus longer-term efforts,
• determine the type of analytics questions being asked for (descriptive, diagnostic, predictive, or prescriptive), and
• maintain traceability of business needs, objectives, research questions, and their sources (for example, stakeholders who asked the research questions or analysis that pointed to specific research questions).

Planning is an iterative process and changes to the approach are made as new knowledge is gained. Each of the six business data analytics domains includes an element of planning which may influence the overall approach to analytics.

There is no right or wrong answer as to the degree of formality of the business data analytics approach. Some organizations may choose to formally document the decisions made when defining their approach by using a business data analytics planning template. Other teams may choose to build more visual models to capture the decisions and include the information on shared wikis and within the team's workspace.

When planning the business data analytics approach, analysts use techniques such as brainstorming to quickly identify a list of activities needed to be performed, functional decomposition to break down high-level concepts into lower-level tasks, and estimation to assess how long it may take to complete various activities. Analysts planning the business data analytics approach use facilitation, leadership skills, and negotiation skills to obtain stakeholder consensus.

Planning Business Data Analytics Approach at Various Stages When research questions or the solution approaches involve more complexity than anticipated, analytics initiatives are typically implemented in multiple stages of maturity: A proof of concept stage which focuses primarily on feasibility of the analytics approach. A pilot stage which focuses on limited scale solution to discover integration and quality issues. A production stage which focuses on business value for customers or internal stakeholders. Deployment Stage Level of Formality Assessment of Organizational Capabilities and Resources Planning Outlook Data Characteristics Proof of Concept Low Qualitative assessment of capabilities Independent and agile teams Low governance Only key stakeholder engagement Flexible solution and data architecture Near-term Static and limited data without data pipelines from different sources Noisy but consciously curated Pilot Medium Both qualitative and measurable assessment Larger teams with cross- functional capabilities Some governance structure defined Most stakeholders are identified and engaged Solution and data architecture are defined Mid- to long-term Dynamic data integrated to most of the known data sources Usable data with defined transformation procedures Production High Both qualitative and measurable assessment Larger teams with cross- functional capabilities Governance structure deployed Most stakeholders are identified and engaged Solution and data architecture are implemented and integrated to enterprise architecture and data strategy and governance Long-term Dynamic data integrated to most of the known data sources Usable data with defined transformation procedures

2.1.7 Select Techniques for Identifying the Research Questions

The following is a selection of some commonly used analysis and analytics techniques applicable to the Identify the Research Questions domain. The list of techniques does not represent a comprehensive set of techniques used by an analyst in the Identify the Research Questions domain but presents a small, but useful, set of techniques that can be used.

Techniques	Usage Context for Business Data Analytics	BABOK® Guide v3.0 Reference
Business Model Canvas	Used to understand how the enterprise produces value by analyzing the key component and business eco-system. It can be modified to include a data view on how data is used by the enterprise at a high level.	Chapter 10.8
Concept Modelling	Used to organize business vocabulary and their interrelationships. This can be used as a starting point to identify and validate data requirements as well as to correlate a business need for research questions.	Chapter 10.11
Decision Modelling and Analysis	Used to establish a decision hierarchy by capturing the flow of decisions that are commonly undertaken in an organization using various tools such as decision trees, tables, decision networks.	Chapters 10.16 and 10.17
Document Analysis	Used to understand the context through minutes of business meetings, internal reports, reports from other organizations, academic literature, previous analytics project reports, the data and methodology employed, the statistical results, and the subsequent business decisions.	Chapter 10.18
Interviews and Workshops	Used to understand the business problem or opportunity from the perspective of different stakeholders, or the organization as a whole, and to elicit more concrete research questions that contribute to the business problem or opportunity.	Chapters 10.25 and 10.50
Metrics and Key Performance Indicators (KPIs)	Used to measure the performance of the organization in different functional area or business goals. Many analytics engagements are undertaken to optimize or explain the KPIs and metrics.	Chapter 10.28
Organizational Modelling	Used to understand the organization’s capabilities, resources, and group structures to discover levels of data insights and research questions posed by different groups in the context of their roles and capabilities.	Chapter 10.32
Prioritization	Used throughout the business data analytics effort to focus attention on the most urgent items. For example, when determining what is important to resolve, formulating research questions, sharing insights, and recommending actions.	Chapter 10.33
Process Modelling and Analysis	Used to understand the organization’s processes where data is generated, and consumed, and to discover ways to identify improvement opportunities where analytics solutions may improve the business value.	Chapters 10.34 and 10.35
Risk Analysis and Management	Used to identify and manage any risks originating from a specific course of action, decision, or assumption that affect analytics engagement.	Chapter 10.38
Root Cause Analysis	Used to understand the business problem by systematically examining the probable causes to develop an intuition for the nature of research questions.	Chapter 10.40
Scope Modelling	Used to understand the scope and boundaries of the analytics engagement and the external context within which a future analytical solution may operate.	Chapter 10.41
Stakeholder List, Map, or Personas	Used to understand stakeholders and their characteristics with additional focus on how they generate and consume data and insights.	Chapter 10.43
Exploratory Data Analysis	Used to quickly understand the readily available data and insights to build intuition about the analytics problem and any contributing factors.	N/A
Hypotheses Formulation and Testing	Used to develop a premise for a particular result based on business stakeholder or SME’s opinion which is statistically justified through the data to formulate research questions accurately.	N/A
Problem Reframing and Shaping	Used to facilitate deliberate thinking about a specific business or research problem from multiple perspectives such as stakeholder, market, or customer. Problems are restated for an analytics initiative. For example, if the analytics problem relates to identifying high-value customers of an organization for a custom marketing campaign, understanding what high-value means and reframing the problem. In this case, it may mean lifetime value, highest spending, or customers interested in high- value purchases.	N/A

2.1.8 A Case Study for Identify the Research Questions
 
.1    The Challenge

Marsha has worked as a business technology consultant at a large investment bank for the last four years. In that time, she has worked on several high- profile engagements and has forged strong working relationships with several key bank employees. John, one of the trading floor managers in the Chicago office, recently attended a big data conference and learned about some new approaches for leveraging predictive analytics tools on the trading floor.

John was convinced that his team would benefit from more sophisticated application of technology and analytics tools. He also worked with Marsha previously to establish and automate various types of trading floor transactions, and he knew Marsha was the ideal person to help with this work. John connected with Marsha and they discussed opportunities to help improve his team's efficiency using some of the newer approaches. After a conference call with Marsha and her consulting manager, John realized there were a number of outstanding issues that needed to be addressed before implementing any new tools. He also realized careful analysis of data would help identify the best way to move forward. John assured Marsha that any information, trading data, resources, and support from his team would be made available as she needed.

.2    Approach for Analyzing the Business Problem

As a seasoned professional, Marsha knows that seemingly simple business problems or opportunities often hide layers of complexity. She has seen the results of poorly conceived data analytics initiatives and knows that various key stakeholders have differing views of potential solutions. She also understands the need for developing a shared understanding of the business problem that needs to be solved as a crucial first step.

Marsha asked that they pause and take some time to develop a shared understanding of the problems to be addressed through application of new technology. Although John was eager to move quickly, he agreed to follow Marsha's recommendation of conducting a discovery workshop to fully understand the problems and prioritize the ones to be solved by use of analytics. Marsha proposed the following approach:

Workshop Stages	Workshop Planning and Activities
Pre-Workshop prep	Identify relevant stakeholders using her experience with the bank and include John's suggestions for additional stakeholders. Conduct initial research to develop relevant knowledge on pricing, market volatility assessment, and how trades are conducted on the trading floors. Coordinate workshop logistics with John and develop the workshop agenda.
Workshop plan	Highlight some of the key areas in investment banking where deep and non-traditional analytics approaches are being successfully used. John expects that it may be a big change for many of the stakeholders to consider new ways to conduct trades. He wants to provide a short explanation on the use of deep analytics. Present some of her high-level research on the topic such as an industry point of view, benchmarking results, and competitive analysis. Use analysis techniques such as root cause analysis, 5 Whys, and business model canvas to identify problems to be addressed. Concretely define the business problems framed as research questions that data can help solve and outline the next steps.
Post-Workshop wrap-up	Share the list of research questions and other workshop results with attendees. Highlight next steps and follow up on the action items that may set the direction of future scoping and elicitation activities.

Pre-Workshop Activities

Based on her experience with the bank, Marsha assembled her initial list of stakeholders which included stakeholders from the Quantitative Research and Analysis team (Quants team) and the office of the Risk Management team (Risk team). John added additional stakeholders from his trading teams, operations, IT, and corporate governance. After agreeing on the workshop participants, agenda, and logistics, Marsha started her research work in preparation for the workshop.

Marsha conducted one-on-one interviews with John and some of the traders to learn about the business of trading desks. She learned how trade call sheets are communicated and how trades are executed on the trading floor, and created process models to capture this information.

She learned trade calls are passed to the traders by the Quants team who have modelled a complex process of predicting option price paths with a combination strategy using Black Scholes and RiskMetrics™ models. She saw that trades are divided into various groups such as speculations, hedges, and arbitrage teams. Marsha also learned how the Risk team sets daily Value at Risk (VaRs) based on traders' book, experience, and prior positions. She followed the process through, noting these parameters and trade plans are fed into the trading terminals. Although aware of the basics of these practices, Marsha realized she would need to understand additional process complexity to help the team discuss alternate sources of information for traders to consume. She decided to leave those discussions for the workshop.

The Discovery Workshop

The workshop started well as Marsha discussed how analytics models can be used in investment banking use cases. However, she felt uneasy with some of the comments. After asking some additional questions, it became apparent that both the Quants team and the Risk team were reluctant to provide their support. The Quants team members felt their predictive models were already cutting-edge and comparable to industry benchmarks. The Risk team were concerned the VaR computation would be impacted by the introduction of new models and significantly impact today’s widely used processes. Marsha was able to redirect the participants to the benefits, including potential bottom line improvements that could occur. She was also able to leverage her research to demonstrate how competitors had addressed similar challenges. Both teams agreed it was important to proceed even if it meant changes to current practices.

With everyone on the same page, Marsha tackled the primary workshop objective which was to identify problems which could be answered through use of data analytics.

Post WorkShop Wrap-up

Marsha was able to consolidate the findings from the workshop and shared these with the group of stakeholders for agreement. By re-framing the business problem successfully, Marsha was able to demonstrate aspects that can be solved through business data analytics, as well as opportunities for both process and application development improvements. Marsha was able to share this outcome with the team and scheduled a follow-up discussion to prioritize these goals with John and other relevant stakeholders.

.3    Outcomes Achieved

Marsha leveraged her facilitation skills to drive the attendees towards a shared understanding and agreement about next steps. To get there, Marsha used specific business data analytics techniques to achieve incremental agreement, as follows:

• Validating business needs: Marsha decided to use the 5 Whys technique (as referenced in 3.15 Problem Shaping and Reframing) to both validate and develop a shared understanding of the business needs and goals John wanted to achieve. John was instrumental in helping everyone understand how even a small improvement in trading floor efficiencies could result in significant return on investment (ROI) and lead to a competitive advantage for their firm.
  By applying the 5 Whys technique iteratively, Marsha was able to determine that the root cause of some of the recent loses stemmed from specific models currently being used for option pricing. It established that there was a clear business need to improve trading decisions through better analytical approach for predicting price movements. Marsha determined that the next step should involve investigating how existing models work.
  
  
• Validating assumptions: Marsha approached the Quants team and the Risk team to understand the existing prediction models in depth. The Quants team and Risk team stated that those isolated losses were expected and as part of market risk, simply a cost of doing business. Marsha decided to pursue this further and suggested they test the assumptions associated with the current models. The Quants team outlined that their models are based on two primary assumptions. The Black Scholes model assumes that option prices follow a log-normal distribution and prices can be essentially predicted based on current spot price (current market price) and the historical volatility (standard deviation). The RiskMetricsTM model is an auto-regressive time series model. In simple terms it says yesterday's price has more weight than the price that was the day before. Hence, future prices can be predicted by creating a weighted average model. Similarly, the Risk team stated that the daily limits (VaR) are set by determining the probability of loss at a certain confidence interval (for example, 95%) and it assumes either Black-Scholes model or RiskMetrics in determining underlying distribution. These model characteristics indicate that the assumption around losses are valid, and there may be possibilities to better predict these losses using more variety in the data used.
  
  
• Utilizing business knowledge: Marsha noted that these model assumptions are based on very limited historical trading data, and other equally important data was being ignored. For example, the effect of the news cycle was not taken into consideration, type of industry of the underlying stock was not considered, and other macro parameters like GDP, interest rate spreads, and underlying asset fundamentals could also be part of the predictive approach. When Marsha outlined these, the operations reps and the traders stated that they always follow foreign markets, daily news, and sometimes Twitter before triggering the trades.
  
  
• Demonstrating outcome with plausible solutions: Based on previous experience, Marsha realized that without describing a potentially feasible solution option, it would be easy for attendees to get “bogged down in the details” of today's challenges, or worse yet, jump to one of the technology solutions that John had seen at the conference. She started illustrating the business opportunity in more concrete terms and outlined some recommendations to improve trading floor efficiencies:
  • Marsha described a future vision of a new predictive model for trading floor operations by describing benefits that could be derived. She focused on key benefits that would appeal to various stakeholders at the workshop:
    1. Improve confidence of trades by providing more information to traders (improve the availability of relevant information on trading terminals and presentation of heterogenous data).
    2. Limit undue risks to overall book of business (parallel pilot with existing models and/or pilot with arbitrage trades that involve low risk opportunities).
    3. Track the performance of the new model (measure the opportunity cost when trades are not executed based on the new model).
    4. Retain oversight on trades and risk levels (VaR) to maintain regulatory compliance (explore how risk levels can be employed for the new predictive model).
    5. Train traders to utilize the new predictive model results.
  • Marsha also mentioned that the exercise is not only developing a new predictive model for option trading; it would also impact various other operations. For example, it would include application development (1, 2), process optimization (1, 2, 3, 5), integrating enterprise data sources for big data implementation (1), and business intelligence (4) components. Marsha concluded the workshop on a high note by describing the business opportunity in detail and attendees enthusiastically agreed that developing a plan for executing this work was the required next step.

.4    Key Takeaways

• It can be challenging to achieve the desired future state without ensuring stakeholder alignment and understanding. The starting point is not necessarily implementation of new technology but instead developing a shared understanding of the problem that needs to be addressed.
• Clearly articulating the need and business problem requires a significant amount of upfront analysis. For example, in this scenario, the business problem could lead to five considerations, which could include initiatives beyond analytics. Without considering all aspects of a business problem, an analytics solution may be misaligned to the overall business objective.
• When reviewing existing analytics processes or models, verify the principles on which the model was built. This often leads to discovery of limitations in the existing models and leads to better solutions. For example, by examining the model assumptions such as distribution of option prices being log-normal (Black-Scholes model) or higher weight is given to more recent option prices, while predicting future options prices (RiskMetrics) shows that no other parameters are considered in the analytics model.
• Correctly applying effective business analysis techniques such as a workshop leads to collective problem analysis by different stakeholder groups. Through shared understanding and hearing different perspectives, root causes can be uncovered and problems can be effectively prioritized for solving.