Let’s look at Grice’s 4 maxims and their applications for a data scientist

Paul Grice, was a great philosopher of language and is best known for his theory of implicature and cooperative principle

A maxim is a concise expression of a fundamental moral rule or principle

1. The Maxim of Quantity: Where one tries to be as informative as one possibly can, and gives as much information as is needed, and no more

One of the most important things that you will do day in and day out as an analytics or Data Science professional is perform data analysis deep dives.

While statistical moments (mean, variance, skewness, and kurtosis) would give you a general understanding of your data distribution, it becomes critical to analyze metrics via different cuts, which could be:

• Temporal analysis: How the metric or your data moves over time
• Cross-sectional analysis: Analyzing different subsets of your population (e.g. geographies, user types, different products segments, demographics, etc.)

As they say that the devil is in the details, you’re expected to look at the data from all sides and perform various data cuts using different segments to get a 360-degree view of the data problem.

However, it is equally important to understand the constraints of time and resources, and not fall into the trap of “Analysis Paralysis”.
Yes, Data analysis also follows the law of diminishing marginal utility, the additional gain in insights by spending one extra hour in analysis gives poor returns after a certain time.

Hence, as data scientists or analytics professionals we should focus on providing as much information as needed, but no more.

2. The maxim of quality, where one tries to be truthful and does not give information that is false or that is not supported by evidence

This should be bread and butter for analysts, as supporting information backed by strong data evidence is their primary job.

Statistical significance:
The connotation of “truthfulness” for Data Scientists and Analysts changes to not just publish data-backed stories but provide statistical significance of data as well. For instance, publishing observations without confidence intervals and p-values is a cardinal sin. It is crucial to have a strong grasp of statistical concepts like hypothesis testing, performing controlled experiments, statistical inference, etc., in order to deliver truthful information.

Being cognizant of personal and data biases:
Missing information: Current big data is just “found data” and N (population) is NEVER = ALL. So always ask what numbers are and aren’t collected, what is and isn’t measured, and who is included or excluded.
Confirmation bias: Improving metrics might be in our favor for personal growth, hence leading to confirmation bias and thus we might look for data to reach a certain pre-defined conclusion. Refrain from doing this!

Understanding CCC or Causation-Correlation-Conundrum:
Correlation doesn’t imply causation, a famous example is how ice-cream sales are correlated with sales of sunglasses but one doesn’t cause the other (I bet it’s the summer! 😀). Be extremely cautious in drawing inferences just on the basis of correlations and be mindful of false conclusions.

Hence, performing statistical significance tests, avoiding biases and understanding causation are some ways to maintain truthfulness in your data narrative.

3. The maxim of relation, where one tries to be relevant and says things that are pertinent to the discussion

Though this might sound self-explanatory, there are many important concepts under this umbrella to be understood.

Problem understanding and definition:
Framing business problems and turning them into a statistical challenge is a top skill and doing it in an incorrect way will lead to the GIGO effect or Garbage In, Garbage Out. Spend enough time on explaining and building context around the problem you’re solving so that the follow-up analysis feels logically relevant to it.

Selecting relevant metrics and representing them correctly:
Building context around metrics is crucial to make them relevant.
Selection: Select relevant metrics which measure the primary objective, can be improved, and inspire actions. Explain types of metrics — Leading, lagging, guardrail, etc.
Representation: Showing percentages vs absolutes for some metrics, changing the axes in a graph when scales are different (sales in dollars & orders in units), etc.

Providing Base Rate and Scale:
Base Rate: The naturally occurring frequency of a phenomenon in a population. Explain how different the results are from the base rate to make relevant comparisons.
Giving a sense of scale: Is 0.1% too large or too small to focus on?

Faced with a statistic, simply ask yourself, “Is that a big number?”
 — The Data Detective

Answering “So what” questions:
It’s great that you have complex analysis and beautiful graphs, but each data insight has to lead to a relevant actionable outcome. And thus, always ask yourself after analyzing data - “What am I going to do with this information”, or in short “I have this info, So What?”.

Hence, providing the right context about the problem and metrics, and answering “So What” questions are important in providing relevant analysis to solve business problems.

4. The maxim of manner, when one tries to be as clear, as brief, and as orderly as one can in what one says, and where one avoids obscurity and ambiguity

While number crunching and statistical analysis are the means, but the end objective is to solve problems and that should always take precedence.

More analysis doesn’t always imply better analysis:
Re-evaluate your hypothesis and supporting analysis, and choose the appropriate graphical representations to build a clear and concise story. Review your work with peers, it helps a lot!

Cover all bases and drop clear explanations
Including clear answers to all possible questions about your data narrative avoids side-tracking discussions. Some questions to include would be:
What were the assumptions used, how did you treat null values or missing information, what biases were present in the data, why did you select these metrics, how did you pick a certain model, etc.

Tailor narratives based on audience:
Another great skill to have is to understand your audience while you build your data story.
Technical presentation: Focus more on your approach to data analysis, model building, feature selection, data biases, statistical tests performed, etc.
Product presentation: Focus more on user stories, product challenges, how this informs product strategy, how much business impact is created, etc.
Leadership presentation: Focus on high-level context, important metrics, data science input to the project, impact-generated, next steps, etc.

Concise analysis through giving clear explanations to possible questions, tailoring narratives based on audience, and peer-reviews are some ways to be clear, brief, and orderly, and avoid any ambiguity or obscurity in your data story.

Overall, it would be extremely useful for any data science or analytics professional to incorporate Grice’s maxims while they’re building their data-backed narratives to generate powerful data stories.

Author: Utsav Shah
Follow me on LinkedIn

For more such informational articles, videos, and mentorship sessions to advance your career, subscribe to Career Nub’s YouTube Channel & Linkedin page!

In this article, you’ll learn about:

• What is a Product?
• What is Product Data Science & Analytics?
• What are the skills required to become a Product Analyst / Data Scientist?
• What are the top tools used by a Product Data Scientist / Analyst

Youtube Video Link | Slide Deck here

What is a Product?

Most businesses in the current world now have an online presence, and hence they will have an online product which is a mobile app or a website.

Product Verticals

However, this is too broad a definition and hence companies have different product verticals.
So Facebook could have so many different product verticals like NewsFeed, Marketplace, Groups, Jobs, etc.

Product Features

Product Verticals are further broken down into small chunks called Product Features. Because the world of business is always changing, products are continuously optimized by:

• Introducing new product features, or
• Tweaking existing product features

Generally, every product feature has its own work-stream.

Feature vs Benefit

Every product feature will correspond → to a benefit or set of benefits for users. For example:
Product feature: Show stories on Facebook
Benefit for users: See the most relevant and important stories of their liking

Product Questions

There can be multiple product questions attached to each product feature based on the maturity of the product:

• Should we launch Stories?
• Which markets to launch first?
• How to design Stories?
• What stories should we show to which user?
• What algorithm would power the ranking of these stories?

Leveraging Data to solve product questions:

More generally, these are some questions that data can help answer:

• What product features to introduce?
• Which users to target?
• What to optimize?
• How to measure success?
• How to prioritize different features?

Companies leverage data science skills to shape & optimize the product

Stakeholder Duties:

• Product Managers: Strategy, gather requirements (PRD) & drive execution
• Engineers: Build these features and drive technical execution
• Product designers: Build design copies, defining UX, conduct user research
• Product Data Scientist/Analyst:
  Build more science around the whole process by leveraging data

What is Product Data Science & Analytics?

While solving problems through data remains the broad theme, this role demands working closely on product related challenges and optimizations

Three most imp things: Deriving Insights, Developing Hypothesis, and Drawing Inference

What are the skills required to become a Product Analyst / Data Scientist?

Data Science Skills:

• Data Extraction [Query Building & Algorithms] (Advanced)
• Data Manipulation (Advanced)
• Experimentation (Advanced)
• Data Visualization (Moderate)
• Data Pipelines (Moderate)
• Data Modelling (Basic)

Statistics Skills:

• Descriptive Statistics helping Exploratory Data Analysis
  Quantitative & Graphical Summary Statistics
  Univariate & Multivariate Analysis
• Inferential Statistics focused on Randomized Experiments
  Hypothesis Building & Experiment Designing
  Statistical Analysis of Experiment Results

Product Skills:

• Product Sense — understanding what makes a product great
• Understanding & Defining Key Metrics
• Product Strategy — driving decisions through data
• Understanding data architecture for the product

Communication Skills:

• Strong documentation
  DS Request For Comments, DS Requirement Doc, Experiment Plans etc.
  Technical & Product Preso, Leadership Reviews
• Stakeholder Management
  Being a thought partner with product managers
  Work seamlessly with Engineering

What are the top tools used by a Product Data Scientist / Analyst

• SQL (Advanced) :
  Check out our Intro to SQL Course
• Python / R (Moderate)
• Internal / External XP Platform (Advanced)
• Excel, Docs & Slides (Advanced)
• Data Visualization: Internal or BI tools like Tableau (Moderate)
• Data Monitoring: Grafana, Datadog etc. (Basic)

Overall it is an extremely interesting role!
You’ll have ownership and accountability of business metrics, creatively solve business problems, get to perform online controlled experiments, and shape the future of the product.

Author: Utsav Shah
Follow me on LinkedIn

For more such informational articles, videos, and mentorship sessions to advance your career, subscribe to Career Nub’s YouTube Channel & Linkedin page!

With the growth of consumer internet and fintech companies in recent years, the application of Data Science to Risk Management has increased rapidly.

In this article, you’ll learn about:

• What is Risk Management?
• What are some of the applications of risk analytics across industries?
• What does a Risk Analyst / Data Scientist do?
• What are the skills required to become a Risk Analyst / Data Scientist?
• What are the challenges in this role?

What is Risk Management?

Risk management is the process of identifying, assessing, and controlling threats to an organization’s capital and earnings. Risk Analytics is the application of data science knowledge to manage risk in a way that helps in revenue growth with minimal losses associated with fraudulent activities by bad actors.

Risk Analytics Applications

Risk is pervasive across all industries but the problems could vary

A few typical examples of the problems that can be solved in risk management in the fintech sector include:

• Calculate the probability that a user will default
• Calculate the credit line that can be extended to a user
• Conduct in-depth credit reviews of high-risk users
• Identify any out of pattern transactions automatically
• Predict if a transaction will stay unsettled

In consumer internet companies dealing with payments, Risk Management problems could include:

• Determine if a stolen card is used for a transaction
• Identify if the refund process is being exploited
• Block fake account signups
• Prevent account takeovers
• Prevent promo abuse
• Ensure that risk friction only impacts bad users

What are the key responsibilities of a Risk Analyst/ Data Scientist?

• Define, track, & maintain KPIs
• Loss forecasting
• Fraud pattern recognition using big data mining
• Build fraud detection features and models
• Optimize profit-based risk management decisions
• Classification of fraudulent users

What are the skills required to become a Risk Analyst / Data Scientist?

• Data Science Skills
• Industry & Domain Knowledge
• Stakeholder Management
• Effective Communication
• Stay updated with evolving threats!

These are some of the essential Data Science Skills for the role:

• Data Extraction
• Data Cleaning
• Data Pipelines
• Data Visualization
• Data Analysis
• Experimentation
• Data Modelling & Evaluation
• Statistics
• Tools- SQL, Python, Excel, Hadoop

Here’s what you should know about the industry:

• Source of Revenue
• Source of Loss
• Payments ecosystem
• Card value chain- issuer, network, acquirer

Some key Financial Terms that every Risk professional should know:

• Underwriting
• Write-off
• Delinquency
• Exposure
• Debt Capacity
• Credit Score
• Chargeback

This role requires great stakeholder management and effective communication skills. You need to use data to explain the rationale of your actions to multiple stakeholders. A Risk Data Scientist needs to collaborate with engineers to develop tech capabilities and product managers to create products less prone to risk.

Fraud trends are rapidly evolving. It is very important to stay updated with the current trends to perform well in this role.

The most challenging part of this role is to balance your actions so they only create friction for bad actors. The precision of your actions should be high and the False Positive Rate should remain low to minimize the friction for good users.

Overall it is a very interesting role that lets you solve mysteries through data and save losses to companies. Each day comes with a new challenge, which makes the role very exciting.

Author: Shivi Sharma
Follow me on Linkedin

For more such informational articles, videos and mentorship sessions to advance your career, subscribe to Career Nub’s You Tube Channel & Linkedin page!