“And the weather yesterday was a sunny 18oC with warm spells in the south and showers in the north. This is similar to the pattern we saw last Thursday.”
Imagine if the weather forecast only restated what happened in the past -- would we bother waiting until the end of the news each night to watch? Compare this to organisations -- how much of an organisation’s business intelligence is a historic view of what happened? And when organisations do use forecasts, they’re often a rehash of previous numbers with an aspirational target added to improve sales performance.
In the UK, weather forecasting benefits industry to the tune of over £1bn per annum, for an annual spend of £120m (see Public Weather Service Value for Money Review, March 2015 for details). It’s the predictive nature of the weather forecast that makes it so valuable to so many sectors of industry, from aviation to civil planners to retailers.
Predictive modelling delivers the same insight and value to businesses, and the ROI seen by the Public Weather Service is not uncommon.
In his book Competing on Analytics, Tom Davenport presents numerous cases showing how analytically-driven organisations outperform their less analytical competitors.
Where to start
For many organisations, it can be too much of a leap to hire a whole bunch of data scientists and hope to get value out of it. The first place to start is your organisation’s data; the old adage “garbage in = garbage out” is definitely true on the analytics journey.
Try some basic predictive models. Decision trees are a good place to start, and SAS includes them in our Visual Analytics suite. Decision trees produce rules like “if customer visits store > 3 times AND age > 30 AND item <$30” THEN “purchase = yes”. The system works out the best variables to choose and the best place to split the rule (for example, those aged above 30 in the case above).
These kind of models can show considerable benefit versus a set of manual rules where in-house experts use skill and experience to create business rules. The rules are easy to update quickly so that you can incorporate the latest customer behaviours.
Where to apply predictive models
Predictive models are being used in every area of business -- the most common include:
- Marketing – predicting which customers are most likely to purchase, churn and cross-sell.
- Fraud – which claims / transactions are most likely to be fraudulent.
- Credit risk – which loans are most likely to default.
- Human Resources – which employees are most likely to leave.
- Manufacturing – which products are most likely to break down.
- Health – which patients are most likely to respond to a particular treatment.
Internet of Things
The Internet of Things (IoT) is generating so much data that organisations are struggling to process and store the data, let alone analyse it. I have a smart meter which measures millisecond changes in electricity usage, yet within a week most of the data has been thrown away, with only hourly data remaining. Predictive analytics could identify key anomalies in this data, differentiating between the children coming home versus someone robbing the house.
Next steps
Most organisations on their analytical journey use a variety of techniques across the organisation and gain true competitive advantage. Let me know how you get on. In the meantime, read the SAS white paper: Drive Your Business with Predictive Analytics.
Connect via Twitter @sukcag
