The Data Science Landscape

Data science is a constantly evolving discipline. There are a lot of buzzwords!

Here is a guide to the tricks of the trade to help you navigate your way through all the noise. It's a work in progress, so come back soon to check out new entries...

artificial intelligence

Artificial Intelligence

AI is machines making decisions. Whether learned from data (machine learning) or by rules imputed by humans, computers are able to take in new information, process it and make a decision that maximises the chance of success or minimises the chance of loss.

Sceptics say the goalposts for 'what is AI?' are moving constantly; some say 'AI is whatever hasn't been done yet'!

machine learning

Machine Learning

Machine learning is a particular application of AI. There are two kinds: supervised and unsupervised.

Supervised machine learning is where an algorithm is trained using historic data to recognise how inputs ('features') relate to outputs. For example, using millions of images that are labeled 'dog' and 'not dog' an algorithm can learn to recognise attributes of photos containing dogs vs. those that don't.

Unsupervised learning, sometimes called data mining, is where algorithms try to identify patterns / splits in the data. For example, it might be able to identify customer segments by behavioural patterns.

r training


Analysing historic data over time, identifying a pattern for example trends or seasonality, and projecting out into the future. For example using past data on sales or store footfall to forecast demand over the summer.

marketing mix modelling

Marketing Mix Modelling

How many sales did print drive? Did TV and Display support each other? Did messaging on product X cannibalise sales from product Y? How should I design our marketing strategy to maximise sales?

A common way to address these questions is to build a marketing mix model. These are models that attempt to disentangle the effects of marketing and other drivers on some KPI e.g. sales. The models can then be used to create scenario planners and optimised media plans.

data science training

Customer Churn prediction

Knowing when a customer is likely to become an ex-customer is hugely valuable. We can predict the likelihood a customer cancels a subscription or switches to a competitor using a statistical model. We can also use this model to understand the levers we need to pull to avert this.

For example, a churn model could be used to build an alert system that generates an email with a discount code for customers looking likely to move elsewhere.

customer segmentation

Segmentation / Cluster Analysis

Which customer groups should we target with promotions? Should we use different messaging for young, high frequency, mid-value, metropolitan female customers vs, low frequency, high value, rural male customers?

Cluster or segmentation analysis is the process of splitting your customers into groups based on their demographic profiles and past behaviour.

Customer segmentations are a very common analytical task. There are many (often very expensive!) pieces of software out there. The key is good data and specialist fit-for-purpose statistical analysis.

R python training

Decision Theory

Understanding the trade-offs of different business decisions is crucial. Decision theory is a framework for allowing you to understand the impact of different decisions. What are the possible outcomes of each decision? What are the opportunity costs?

Decisions scenarios can be understood and optimised using Monte Carlo Simulation (see below).

data consulting services


After how long are customers expected to re-order your products? When should you be giving them a nudge? After how long can you expect employees to be searching for new roles?

Predicting the time until some event is the purpose of survival analysis. Survival analysis models allow you to proactively respond to threats to your organisation with optimised interventions.


Bayesian Analysis

Bayesian methods are becoming a hugely popular in business analytics. There are two key benefits to adopting a Bayesian approach to business analysis:

  1. Ability to supplement data with other sources of information e.g. internal business knowledge. This helps improve results

  2. Results are in the form of probabilities. This means we can report uncertainty in our recommendations and provide give best case, likely case and worst case scenario predictions.

a/b testing

A/B & Multivariate Testing

A/B / multivariate testing allows you to compare alternative decisions in a statistically valid way. This could be two versions of your website or different email subject lines.

It allows you to translate findings from a test into recommendations with confidence.

data strategy consulting

Natural language processing / Text analysis

Computers are getting better and better at reading text. NLP / text analysis allows you to understand the sentiment of tweets, or the topics discussed across articles. NLP can be used in conjunction with machine learning to cluster text documents or to predict the emotion of the language.

data modelling big data

Deep Learning

Deep Learning is a sub-branch of machine learning commonly used to classify images. The typical example is getting a computer to recognise objects like chairs or people in photos or videos.

With deep learning the algorithm discovers what features or characteristics to use to predict the outcome (e.g. chair or not chair). It uses large volumes of data and requires very powerful computer processors.

propensity modelling

Propensity Modelling

Propensity: "An inclination or natural tendency to behave in a particular way."

Propensity modelling is used to predict and anticipate customer behaviour. Customer journeys can be tailored to optimise conversion; product recommendations can be targeting at particular customers.

ai companies

Customer Lifetime value

New data and new modelling techniques mean brands are able to dynamically learn and predict future customer value. Customer LTV calculations allow brands to prioritise and tailor customer experiences, maximise retention and boost purchase frequency.

There are a lot of expensive, black-box one-size-fits all CRM platforms out there. Every organisation is different. Generally it is more cost effective in the longer run to develop fit-for-purpose custom tools and resources.

elasticity modelling

Price Elasticity

Price elasticity is how responsive customers are to changes in price. If responsiveness is high, price decreases might lead to many more sales and therefore greater sales revenues. If responsiveness is low, price increases might boost revenue.

Optimised pricing can massively increase profitability. With the right data and the right data scientists you can predict elasticity to a high degree of granularity. Sometimes prices can be dynamically optimised; by person, by hour, by place.

monte carlo simulation

Monte Carlo Simulation

Monte Carlo simulations allow organisations to understand and predict the impact of their decisions. It's like a computer game that simulates your business. You can pull different levers and watch the effects on metrics you're interested in.

For example, a Monte Carlo simulation could be built to understand how alternative marketing strategies will affect sales. Simulations allow you to understand the uncertainty involved in decision making. What the best case, likely case and worst case scenarios look like.


DS Analytics are a data consultancy. We help our clients get value from their data. Get in touch to find out more or email us at