Statistics for Business

Can Businesses Measure their Risks? Can these Risks be Managed?

risk

 

All businesses have some level of risk. Whether it be customers who do not pay their credit or customers who stop buying from them.

Can risk be measured? Risk of course, has a level of uncertainty attached to it, and may seem difficult to measure.  Many people find it difficult to believe that Risk can be measured.

Risk can be measured. We are fortunate, in today’s times, with advanced technology and historical data, we can use data to recognize historical cases that put businesses at risk. Exploratory Data Analysis and techniques such as Machine Learning can help businesses to identify patterns in the data that are associated with Risk. Using statistical significance testing we are able to identify which factors scientifically contribute to the risk under study.

Of course, not all cases will be identified as ‘Risk’ but some cases will have a higher probability of risk than other cases. This means that a business would need to have a strategy in place to manage the different levels of risk. Some “Risks” are “critical” and need urgent response and action, while other “Risk” are important but less urgent.

In the case where your business provides credit to customers, it is important to identify which customers are highly not likely to pay their loan and the size of their loan. As some customers will have a loan size of $1000 with a probability of 90% chance of not paying back the loan. Another customer may have a loan size of $10 Million with a probability of 60% chance of not paying back the loan. Businesses often forget to look at the loan size and only focus on the probability of someone being a credit risk, when in fact, the higher loan amount is most times more critical that it be paid back than the smaller loan sizes. Businesses need to have different risk minimization strategies based on “Size of Loan” and “Probability of Credit Risk”.

When Businesses have a Strategy for Risk Minimization, they can integrate their credit data analysis results with their risk minimization strategy and automate it into a complete decision support system for their credit officers.

Fortunately, with the large volumes of data, businesses can use machine learning techniques to crunch customer transactional, demographic, social media, and lifestyle data, etc to score customers faster and smarter giving businesses the enhanced customer experience and competitive edge in decision making.

Should you be Learning more about Machine Learning Algorithms?

pexels-photo-186461.jpeg

More and more businesses are switching from reacting to situations to anticipating them. How do they do this?

In order to anticipate or predict a situation, an organization must have a business question in mind. Questions such as:

Is this a fraudulent transaction or not?

Will the stock price in the next week go up or not?

Is this particular person a credit risk or not?

Will this customer buy item ‘A’ or not?

How many units will be sold next week?

What will the stock price be tomorrow? etc

The best way to make business decisions is to analyse the relevant historical data and to identify the patterns in the data. To identify the relevant data, it is important that the analyst speaks with the domain experts to help them identify the relevant data. Further, it is not only the analysis of the historical data (being reactive always) that can help answer the above business questions. It is also important that businesses use predictive data (being proactive and sometimes preventing negative activity) to help answer the above questions.

Machine Learning is a method for data analysis, in particular, Machine Learning Algorithms focus on using the relevant data inputs to make Predictions for the business or Classify the business products, services or customers.

Machine Learning is an algorithm that explores the data for patterns and identifies the key data (factors) that will help the business to predict/classify its target outcome of interest (like the above business questions). Machine Learning uses statistical techniques (such as the logistic regression, neural networks, decision tress, etc) and computer programming languages (such as Python, R, SAS, Matlab, etc) to automate the predictive/classification models using algorithms that iteratively learn from the data and find the hidden insights that humans would not usually have known about. These hidden insights are like ‘gold for the business’ as they usually improve the business predictions/classifications without being explicitly programmed where to look. The Machine Learning Algorithms further improve when they are regularly updated with new data and knowledge.

For Machine Learning Algorithms to be effective, it is important for businesses to be able to capture the relevant, good quality data in real time or almost real time and to allow the right people in the business to apply the predictions/classifications timely and thereby gain advantage over their competitors.

Are you interested in learning more about machine learning algorithms and how your business can improve its sales/predictions and become ore efficient? Contact me at carol.hargreaves@dataanalyticsexperts.com

What makes a Good Analyst?

_techniques

 

To be a great analyst you need to be able to think clearly and ensure that you understand the problem at hand well. A great analyst will ask many intelligent questions with regards to the problem and would be able to determine through speaking with a domain expert, which variables will be relevant for solving the problem. A great analyst is a very detailed person who would ask lots of questions and ensure that key information has been considered and collected.

In addition, a great analyst would select the right visual display to demonstrate the results of his/her analysis. Good communication skills are also a key trait as often the statistical results are too complex for the client to understand. So, the analyst needs to present the results in an easy to understand manner. Analysts who present the results in a story like manner with great visuals and good use of technology will be well liked for their presentation skills.

Further, accuracy is very important in the analytical world. So a great analyst will build predictive models as part of their solutions, but will always compare and evaluate the accuracy of their model with a few other models and also ensure that the accuracy is of a reasonable and acceptable level. Further, a great analyst will test their model results by validating the model against unseen data and against the actual data when it becomes available.

In a nut shell, a great analyst is passionate about solving challenging problems and works hard to ensure that the analytical solution while being highly accurate is at the same time making business sense and offers business value for the client. A great analyst will always exceed expectations and deliver more insightful information than the client expects and is paying for.

If you have any questions, please do not hesitate to ask me. You are also most welcome to share your thoughts here, on what makes a good analyst. If you would like to learn more about ‘Analytics’ or how to start ‘Analytics’ in your company, learn more at http://www.datanalyticsexperts.com or contact me.

My First Job in Analytics – A Walk Down Memory Lane….

 

_techniques

My first industry job, I worked as a Biostatistician at the National Health & Medical Research Council at the Clinical Trials Centre in Australia. The work was very challenging as I was working with real data, and even more stressful, I was working with patient data.

As a Biostatistician, I had to work closely with the oncologists, surgeons and doctors, and determine the Statistical Analysis Plan for the Clinical Trial. Will the Trial be a Double Blinded or Triple Arm Study, Will a Placebo be used, How many patients do we need on the trial, What are the inclusion and exclusion criteria, When will we stop the trial, etc. All these questions were answered using rigorous statistical techniques based on clinical knowledge and historical study results.

I would often share my solution with my manager, Val Gebski, who too often challenged my solution by throwing a lot of “what if” scenarios and getting me to think on several possible outcomes. We often spoke and debated for hours, sometimes the whole day, about different approaches that could be taken.

I remember some nights, in the middle of my sleep, I would get up to write out a method to use the next day to justify my solution. And, first thing in the morning I would run my method on some data and be very excited when the results agree with the typical method and in this way validates our new innovative approach.

I have many examples that I can share but I would say that problem solving is what I enjoy as it challenges me to think of new methods on how to approach a given problem.

I strongly believe and have seen that statistical/analytical techniques are powerful in producing repeatable and consistent results that are accurate when validated and, in a nut shell, Analytics is all about understanding a problem, understanding the relevant data that can help you solve the problem, asking lots of questions and then applying statistical/analytical techniques to make comparisons, predictions and finally optimizing your solution.

But How do we Segment our Customers?

How should you segment your customers? The first question I would ask, what is your business objective?

Do you want to understand when to promote to a customer? If your answer is yes, then I would recommend studying your customer behaviour.

What do I mean by customer behaviour? We have to acknowledge that not all customers are the same. Some customers buy more often than others, some buy more consistently than others, some need a ‘push’ to buy, some will not buy no matter what you do…

So we need to understand our customers in terms of…
1. Which customers will buy
2. Which customers need a ‘push’ to buy
3. Which customers, no matter what you do, will not buy

There are various customer metrics that can help you to understand your customer behaviour. Two top of mind metrics for better understanding customer behaviour is:
1. RFM (Recency, Frequency, Monetary Value)
2. Latency (The Average Time between Purchases)

These two metrics are extremely good as they are actionable metrics…they indicate to you WHEN TO ACT. Very important!
Never compute metrics that do not indicate to you when to ACT or DO NOT assist you in making proactive decisions.

Today, I will speak a little on RFM. RFM allows you to score every customer based on:
1. When last they bought from you (Recency)
2. How often they bought from you (Frequency)
3. How much they have spent from you (Monetary)

Each customer, is scored on a rating from 1-5 for each metric Recency, Frequency, Monetary Value, where ‘5’ is most favourable and ‘1’ is least favourable.

The customers with the highest scores, for example, ‘555’ are your most valuable/profitable customers (They have bought most recent, most frequent and spent the most with you). These customers are likely to buy again on their own and do not need a ‘push’. They are probably your loyal customers. Customers who are already in love with your products. Pampers these customers.

The customers who were scored previously, for example, ‘111’ and now are scored ‘333’ are probably fairly new customers, you need to give them a ‘push’, promote to them through a ‘Cross Sell’ or ‘Up Sell’ Campaign. These customers may not be familiar with all your products as they are fairly new customers so by using ‘Cross-sell’ or ‘Up-Sell’ Campaigns, you are likely to get them to spend more with you & increase their likelihood of becoming loyal customers.

On the other hand, customers who have been previously scored, for example, ‘333’ and are now being scored a ‘111’, it is likely that these customers are about to leave you. Promote to these customers as this is where you are able to get high impact ROI, help these customers to spend more with you before they leave you. Send these customers a campaign, probably an anti-defection campaign.

Customers who have previously been scored ‘111’ whom you have promoted to since, and they still have not bought, these customers are likely to have left you already. Do not promote to these customers.

Does the above make sense? Let me know….If you have any questions, please feel free to ask them

Business Analytics – What is Business Analytics?

Business Analytics – What is Business Analytics Carol Hargreaves

Interest in Business Analytics is growing each day. Many still do not understand the difference between Business Analytics & Business Intelligence. There is quite a big difference between the two.

Business Intelligence is your dashboards, reporting tools, etc that allow you to see how the company is doing in meeting its KPIs or to see how the company is doing now compared with last week or last year this time. Business Intelligence is a summary of how your company is doing usually presented visually using pie charts, bar charts or trend charts.

Say for example your reporting system presents your sales this week as an upward trend. Business Intelligence tells you your sales is increasing but, does not tell you why it is increasing. Secondly, when seeing your Business Intelligent information, your action is likely to be reactive.

On the other hand, Business Analytics allows you to model your sales and better understand why your sales is trending up. Using models such as the logistic regression or decision tree, you are better able to understand what is driving your sales. And, in turn, you may act proactively and do more of “what is driving your sales”. Business Analytics allows you to be proactive and hence gain competitive advantage over your competitors.