What it means
Data science brings together three skill sets: mathematics and statistics, computer programming, and an understanding of the business problem. A data scientist might begin with a question such as "which customers are most likely to pay late?" and then gather the relevant data.
Much of the work is preparing the data, because real-world records are messy. Once the data is ready, the analyst explores it for patterns and tests ideas with statistical techniques.
Machine learning, where algorithms learn patterns from examples instead of being given explicit rules, is one of the main toolkits. The resulting model is checked on data it has not seen before to make sure it works beyond the sample used to build it.
The final step is turning the model into a decision or a process. That might be a dashboard for managers, an automated credit check or a forecast that feeds the budget.
A brilliant model that no one uses creates no value, so communication and change management are as important as the maths. For finance teams, data science changes the questions that can be answered.
Forecasts can draw on hundreds of variables instead of a handful, fraud systems can learn new patterns as they appear, and collections teams can prioritise the accounts most likely to pay. Teams also need controls over models, including documentation, testing and monitoring for drift as conditions change.
The nuance is that data science is a means, not an end. The value comes from decisions that are better or faster, and the cost includes people, tools and data quality work.
Start with a business question, a measurable benefit and clean data, and the principle "Gold in, Gold out" applies: weak inputs give weak outputs.
In practice
Real-world examples.
Example
A bank builds a model to predict which small business loans are likely to default using payment history, industry and cash flow data. The model scores each application, and loan officers review those with high scores. Defaults fall while approval rates for good borrowers stay steady.
Example
A subscription software company analyses usage data and finds that customers who do not log in during the first two weeks are far more likely to cancel. The customer success team contacts those customers early. Monthly cancellations decrease and recurring revenue grows.
Example
A retailer uses a demand forecasting model that combines past sales, weather, holidays and promotions to set stock levels. The finance team compares forecast and actual sales each month using MAPE. Stock-outs fall and the amount of cash tied up in inventory is reduced.
Formula
Calculation
Mean absolute percentage error (MAPE) = average of (|actual - forecast| / actual) x 100%
Suppose a forecasting model predicts monthly sales of $210,000, $240,000 and $330,000, while the actual figures are $200,000, $250,000 and $300,000. The percentage errors are 10,000 / 200,000 = 5%, 10,000 / 250,000 = 4% and 30,000 / 300,000 = 10%. MAPE = (5 + 4 + 10) / 3 = 6.33%, which means the forecast is on average about 6.3% away from actual sales.Case study
Seen in the real world.
Harlan Wholesale is an illustrative, fictional distributor with 5,000 product lines. Its finance director was frustrated that the quarterly forecast was usually wrong by more than 15%, which led to either excess stock or lost sales.
She hired a small data science team to build a forecasting model using three years of sales data, supplier lead times, seasonality and customer order patterns. The first version performed no better than the old method, because the data contained duplicated product codes and missing dates, so the team spent six weeks cleaning it.
The second version cut forecast error to about 8%, and the company reduced its stock by $1,100,000 without hurting service levels. In this illustrative story, the director credited the cleaning work, not the clever algorithm, for most of the improvement.
Watch out
Common mistakes.
- Starting with the tools instead of the business question, so the project produces interesting results that nobody needs.
- Skipping data cleaning and trusting the model, when poor quality inputs produce unreliable outputs.
- Building a model once and never checking it again, when changes in customer behaviour or the economy can make it less accurate over time.
Questions
People also ask.
What is the difference between data science and data analytics?
Data analytics usually describes examining data to answer questions about the past and present, while data science adds predictive modelling, machine learning and building data products.
Do finance professionals need to learn data science?
They do not all need to build models, but understanding the concepts helps them ask good questions, challenge results and manage risk.
How is data science different from data mining?
Data mining is a set of techniques for finding patterns, whereas data science is the broader discipline that includes data preparation, modelling, deployment and communication.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%Related