Predictive Analytics has been in existence for a while; however 2014 will see it move further to the centre stage of hot topics. Why? Organisations are demanding more agility and innovation through technology to keep competitive. Their market differentiators require frequent re-definition and evolution to maintain a leading position and strengthen customer loyalty; thus requiring innovative ways to understand, engage and influence their customers. So, technology needs to step up, even more, to help deliver better foresight, proactivity and automation.
Behaviour, let’s say customer behaviour, is what many organisations wish to understand. For example; using data from contemporary business channels (mobile, social, web, etc…), companies are devising ways to predict future activity, based on a range of influential business scenarios, so as to stimulate current behaviour of existing and prospective customers. They often hold great volumes of data on their client’s activities from business and social sources where predictive analytics tools would be in their element processing all this data to detect behavioural trends and its impact on revenues.
Figure 1 – Forecasted revenue trends, using SAP Predictive Analysis
To the layman, predictive analytics can be seen as a very specialist topic and may indeed struggle to see where it could be of value to them. In response, vendors are vying to bring predictive technology to business users through user friendly interfaces that apply agile methodologies to acquire and visualise data. The objective being; users should not need to be expert statisticians to apply predictive technology to their everyday business data.
Business Insights? That’s so last year… Business Foresights is the new Business Insights.
Naturally, data is at the root of everything we do and today’s technology strives to simplify the process for businesses to acquire, blend, enrich, analyse, etc..., their social and business Intelligence as swift as possible.
Throughout 2013, AgilityWorks have been working with a variety of new tools with predictive and visualisation capabilities to discover a common thread among them in their approach to agile analytics (Figure 2).
Figure 2 – Simplified approaches to model and visualise data
We cannot state enough the value many visualisation tools can bring to organisations. Now that these tools are extending to predictive capabilities the business user has never been more enabled to innovate, conceptualise and qualify gut feel or hunches through simple, unified and structured methods; without IT.
To walk you through…
1. Data Acquisition & Preparation - Data is acquired from a variety of sources to build up a broad range of attributes on a set of cases. Specific attributes selected for modelling are otherwise known as predictors. The data will often go through a range of pre-processing models which can include the following:
- Data quality checks (eg: removing nulls, incorrect data, outliers)
- Normalisation or De-Normalisation
- Data enrichment
2. Data Model Development - Predictive models will be used to generate a range of outputs against a set of cases, for example: probability scores; Boolean data; forecasted values; and so on. Best practice determines that models should go through lots of training during their development lifecycle against data that has proven outcomes. Predictions are made and compared against the proven data; and the model is validated or revised as more data becomes available. Examples of predictive models are decision trees, neural networks, logistic regression or rule based models… But that's getting a bit too deep… As already mentioned, the aim of up and coming predictive tools is to bring predictive capabilities to the business without users needing to have a PhD in statistics.
3. Model Visualisations - After the data has been processed the next step is simpler, visualising it. Model visualisations can include; Model performance by way of its robustness and accuracy; what the influential predictors in the data are; predictive scores against each case and/or clusters of cases; etc…
Where is Predictive Analytics applied?
- Customer Churn – Understanding and capturing those customers who might leave your organisation
- Response Probability – Capturing the customers likely to be responsive to targeted campaigns.
- Fraud Pattern Analysis and Prediction – Looking at patterns found within customer purchase data to detect outliers
- Customer Product Recommendations – Marketing of specific products based on buying propensity.
- Employee Retention – Understanding and capturing the business & social attributes which affect high performing employees who leave your organisation so as to predict future likelihood.
- Data Extrapolation – Using time series models which summarise the past trajectory of the data to give forecasted trends
- Sentiment Analysis – passing textual data through models to predict sentiment scores, for example ‘1’, ‘0’, ‘-1’ for positive, neutral and negative respectively
What topics can predictive analytics shake up in the near future?
- ‘Big Data’ will resonate more with business because of its relevance to predictive. Its value is realised, not purely through volume (ie. rows and rows of data), but through the broader array of attributes and characteristics that come with it. Namely, social data, machine sensor data, web data and so on.
- Master Data Governance and Data Quality Processes – for example Customer data, Supplier Data, Finance data, HR data, Sales Data - will be even more pervasive in organisations that may not have taken these issues as seriously as they could have. Predictive Data models require data that is the best quality possible so as to help provide robust and accurate predictions.
- Enabling Prescriptive Analytics – What is it..? Predictive technology enables users to visualise a range of future probabilities. Prescriptive BI goes beyond predictive by analysing options presented and recommending the next best course of action. For now, most if not all organisations apply descriptive analytics for their BI. This is a fundamental and basic form of analytic in the journey to predictive and prescriptive analytics. (Figure 3 below describes the transition from descriptive into prescriptive analytics to bring further value to companies)
With regards to disruption, the next step (i.e. predictive) will, no doubt, ask questions of existing data mining and knowledge discovery capabilities within the business. A further move into prescriptive BI will attract a degree of disruption to get there.
Figure 3 – Analytics maturity
- Cloud Services - We’re seeing this offering from a few vendors enabling organisations to leverage predictive capabilities in the cloud.
To round up…
On the whole, Predictive analytics is technology that provides data on probability and likelihood – not absolutely factual. Its USP is the ability to give some insight from data we do not yet have, without too much complication, for the everyday business information analyst or consumer
It is not a substitute for human intuition. It is a statistical support mechanism that can provide additional assurance to complement gut feeling and other forms of analytics, research and information we would normally consume to formulate a decision.
I’m excited to see it being presented to broader base of business consumers over 2014 and beyond. “Predictive and prescriptive analytics for the layman” is where we predict a strong probability for success in those organisations that see its potential.
Please email questions and comments to email@example.com