101 Guide to Predictive Analytics

Introduction

For many (most?) organisations, forecasting means asking their finance team to collate historical data in a spreadsheet and apply some sort of multiplier. This might then be refined using individuals' judgement and intuition, followed by a debate and/or argument with the finance team.

However well-intentioned, such output is very likely to reflect the unconscious biases of those who created the forecast - and perhaps the very conscious desire to underplay a forecast so that they subsequently look good... And it's neither a quick nor a simple process.

What is "predictive analytics"?

The alternative is predictive analytics: using data science to provide a rigorous, data-led forecast. It will never fully replace expertise and experience but is the necessary foundation for an accurate, trustworthy forecast.

Predictive analytics is the use of data, statistical algorithms and machine learning to forecast the future based on the past. The aim is to extract as much information from the historical data as possible, and to use that information to improve the accuracy of the forecast. You may also hear it referred to as sales forecasting, demand forecasting or algorithmic foreasting.

Why might it be useful?

Clearly the uses for predictive analytics vary by organisation, but we've tried to categorise them so that it might prompt some ideas for you:

Pricing

• Maximise revenue and occupancy in dynamically-priced sectors (e.g. hotels, airlines) by using demand forecasts to inform your pricing strategy

Staffing

• Schedule staff to precisely match demand, be that customer footfall or A&E admissions. The result is some combination of lower costs and better service.

Supply chain

• Reduce over-stocks and out-of-stocks by having an accurate demand forecast at the time the buying decision is made...
• ... and similarly when it comes to allocating stock to individual stores or warehouses

Marketing

• Calculate the future value of different types of customer, which then translates into an 'allowable cost per acquisition' - the amount of money you can spend today to gain that customer whilst still remaining profitable
• Measure the return on investment from marketing / advertising activity without spending five or six-figure sums on econometrics. Comparing the actual sales following a campaign to an (accurate) forecast made without prior knowledge of that campaign gives you its true impact

Budgeting / planning

• Set accurate targets so you can understand genuine over-/under-performance (is a 'good' week really such a good week?)
• Have advance warning of any issues, so that you can be proactive in dealing with them rather than responding after the fact

Risk

• Quantify the likelihood that a given event or transaction is fraudulent

Strategic decision making

• Use a granular, trustworthy forecast to make better decisions ahead of time. This could be anything from deciding how to allocate advertising spend to choosing where to rota on your best team

It sounds too good to be true

Well yes, no forecast is ever 100% accurate. And no matter how advanced your analysis there is always a role for expertise and experience to refine it - predictive analytics should act as a complement, not a replacement.

And as we said at the start, it's not easy. Algorithms need data to work on, and smart people to build and maintain them (and as soon as you put the words 'data science' in your job description, you've probably also added 50% onto the starting salary).

But the benefits of fast, trusted forecasts can be huge and have benefits right across your organisation.

If you'd like to find out, have a read of this: Predictive Analytics in (a Bit) More Detail.

Thanks for reading.