2. “Predictive analytics is the practice of
extracting information from existing
data sets in order to determine
patterns and predict future outcomes
and trends”
webopedia.com
#ProvokeEvent
Predictive analytics is at its core the use of historical data to try and predict future outcomes and trends.
Predictive analytics is not new. At least the concept has been around for years. People have historically tried to use past data to predict the results for the future.
We have been doing this for many years. For example using previous sales data to try and predict peaks and troughs to drive sales and marketing activities.
In the past though the outcomes and trends have had to be gleamed by humans. Sometimes this was really successful, but sometimes it went spectacularly wrong.
1987 – the great storm.
Met office even with 150 years of historical weather data and some very clever meteorologists got it wrong.
Now though we can combat this inconsistency by taking advantage of machine learning.
Machine learning is……
The idea that we can take data and ask a computer to analyse it to spot trends and predict specific values.
But that is essentially the same as what a human could do….what is the difference?
Humans are slow…..computers are fast.
Over simplification and not always the case, but when we think about the vast quantities of data this is when this element comes to the fore.
Some Azure statistics?
Secondly, when we use machine learning, computers can see things that we can’t. Subtle patterns and trends that you might not even think to look for.
What about marketers?
3 types of real world scenario you could use it for.
1. Segmentation / clustering – Behavioral clustering – do they use the web site? Or do they call the call centre?
Product / category based clustering – One group of customers may only buy sweaters, other maybe active wear. Useful for deciding which offers or email content to send to which customersBrand based clustering – Identifying which brands people like and then also other links, such as customers who like Calvin Klein tend to also like Nine West
2. Prediction / Propensity – Modelling a users behavior to identify and predict things such as
Lifetime value
Likelihood to engage, convert or buy – this can be useful to target customers with different types of discounts. E.g. Customers deemed to be likely to buy won’t need huge discounts, whereas those with a low propensity to buy will need a more aggressive approach.
3. Recommendation –
Up sell, cross sell, next sell recommendations.
Worth calling out that many of you will already do modelling of these types and use it to great effect. However, the difference here is that all of your data can be mined and analysed in real time with immediate benefits.
Netflix recommendations……. http://techblog.netflix.com/2012/04/netflix-recommendations-beyond-5-stars.html
Personalised recommendations
Beyond simply ‘like this movie so recommend this movie’
Based not only on historical viewings, but also specific behavior such as they like to watch comedies on a Friday
Also uses data from around Netflix as well as from within Netflix, such as their Facebook activity
Target used predictive analytics to predict when customers were likely to be pregnant and to offer them specific discounts
This seemed to work well – in one case they managed to predict a girl was pregnant before her father knew
Starts with a need.
As Jon said - Still need marketers and the human touch.
But then what about collecting, managing and processing all of that data?
Previously we would need our own super computers……
……Big platforms now available, such as Azure. Provides many tried and true formulas from real world use on Xbox, Bing etc…..
48 hours ago….Cortana Analytics Suite. In preview in the coming weeks.
Perceptual intelligence – the ability to analyse faces, speech and text – for example, garnering sentiment from blog comments or social media
Recommendations, forecasting and churn included as pre-configured solutions which can then be tailored for individual business needs