Predictive analysis: How to make it a success

Is the future written in the data? Five steps for effective predictive analysis
Written by Stephen Pritchard, Contributor

Is the future written in the data? Five steps for effective predictive analysis

Predictive analysis can help organisations predict opportunities that may lie ahead, Stephen Pritchard explains.

Organisations are always looking for more sophisticated ways to predict the outcomes of the decisions they make or the impact of external events on their businesses. Predictive analysis helps companies peer into the future by crunching data to make a forecast on the likely outcome of a strategy.

Whereas other methodologies for finding out what could happen in the future centre around identifying risk, predictive analysis is used more for identifying opportunities, and so is more often used to help decision making in sales and marketing.

Using data to predict the future
The larger the volume of data the more accurate the forecast, and often predictive analysis is used to identify which segments in a group (of products, or customers for instance) will behave in a certain way.

This analysis is dependent on the construction of complex mathematical formulas, based on the variables considered most influential to the outcome: for example if a retailer wants to know where a certain brand of jeans will sell well in a particular month, variables could be population density, disposable income and the weather.

"If the past couple of years have taught us anything it is that a lot of people could have made much better decisions," says Jeanne Harris, executive research fellow at the Accenture Institute for High Performance in Chicago, and co-author of Analytics at Work: Smarter Decisions, Better Results.

"People were using data and models but they were using models the wrong way, or the models didn't identify the right things. They used it to support what they were going to do anyway. So people are now looking to use predictive analysis more thoughtfully."

Businesses have continued to build up larger and larger volumes of data, even during the economic downturn. A big factor in this trend is increasing regulation forcing companies to collect and hold more data to meet regulations or for financial controls.

predictive analysis – looking into the future of your business

Can predictive analysis help your business work out what the future holds?
(Photo credit: Shutterstock)

But companies are also gathering more data for sales and marketing purposes, in order to improve customer retention and to reduce the cost of sales by allowing more targeted campaigns.

The Royal Shakespeare Company (RSC), for example, used seven years' data, and a sample of two million theatre goers, to identify the customers who were most likely to visit its venues again in the future. This has allowed the theatre company to cut the cost of its promotions - by focusing campaigns on people more likely to visit its venues - but still increase ticket sales at its main Stratford-upon-Avon venue by 50 per cent.

Above all though, businesses are being driven by competition to be smarter in the way they predict the future, and the data...

...the business generates - whether from loyalty cards, the shop floor or from the call centre - are a vast resource that could be made to work harder.

Predictive analysis gets sophisticated
This has led to ever more sophisticated tools from companies such as Fair Isaac Corporation (FICO), SAS Institute, KXEN, and SPSS - now owned by IBM. The large enterprise software vendors such as SAP, through its BusinessObjects arm, and Oracle, are also emphasising predictive analytics rather than conventional business intelligence and reporting.

The technology, too, has become cheaper: point of sales and supply chain systems are better integrated, and the cost of the IT resources needed to process data for analytics continues to fall. The RSC, for example, was able to link box office, fundraising and marketing data in a single database.

"Reporting and analytics - business intelligence - tells you what happened yesterday. If you compare reporting and forecasts, reporting is what your sales were and analytics is [predicting] which of your customers are more likely to churn," says Mike Upchurch, COO at Fuzzy Logix, a business advisory firm that uses mathematical models to help companies predict the future.

"Business intelligence is now the baseline. Your competition is looking at analytics and the early adopters are seeing huge increases [in sales and profits]. We lowered one telecom company's churn by 10 per cent with predictive analytics."

Using analytics allows companies to save money by being much more focused in areas such as promotions, and pricing.

In the case of a phone company, for example, analytics can spot customers who make fewer calls.

If this behaviour corresponds with the end of a contract, there is a good chance the customer is thinking about signing up with someone else. The company can then target that customer with a text, letter or phone call, offering a deal to convince them to stay.

predictive analysis

Predictive analysis can help businesses tailor offers to customers
(Photo credit: Shutterstock)

Predictive analytics lets companies tailor offers to the customers the data suggests are open to persuasion: it would be too expensive to send those offers out to everyone coming towards the end of a contract.

The technique can also be used to see how customers might respond to price changes. Premierline Direct, a commercial insurance arm of Allianz, uses predictive analytics to test how a price change might affect customer renewals across market segments.

Once a price change has been implemented, the company runs the analytics again to see how effective the projections were - a technique sometimes referred to as Closed Loop Analytics. Using these methods, Premierline increased renewals by 16 per cent in six months.

How Avis Europe uses predictive analysis
Investing in advanced analytics need not be expensive, and it is quite possible to start a predictive analytics project for just a few users. Some businesses, such as Avis Europe started with spreadsheets.

The car rental company uses predictive analysis to target its email marketing and to fine-tune its campaigns. The company has a relatively small deployment of SPSS' analytics tool alongside its main business intelligence environment.

The company started with a five-user licence, to test the software and see whether predictive analytics would...

...provide both a return on investment and competitive advantage for the business, according to Chris Parker, analytics specialist on the company's marketing team. Avis Europe was also looking for a relatively simple set up.

Avis Europe, Parker says, tended to organise its business around transactions, rather than customers. Using analytics gives a better understanding of how customers, segmented into groups will react to promotions or other marketing programmes.

"Avis Europe sends 18 million customer emails every year, so a simple deployment was essential to avoid adding more complexity to an already complex email marketing strategy," says Parker. "It was simple in the sense that the software had fairly simple tasks to conduct but over high volumes of data."

"In the 2008-2009 financial year, Avis Europe cut the cost of email marketing as a percentage of revenue almost by half," says Parker. "This has a significant impact on the business when viewed across the millions of emails we send every year."

But increasingly, the company is using analytics in areas such as customer retention - predicting customers that might be about to move to rivals - as well as to spot corporate customers who might also want to rent cars for leisure.

The process does have disadvantages, he concedes, including the large amount of data produced by the process, and the risk of applying overly complex models to solve simple problems - either making extra work, or producing campaigns that might target too narrow a group.

As a result, Avis Europe plans to reduce the number of variables in its SPSS models - some criteria that seemed important in the initial deployment are rarely used in actual campaigns, says Parker. "Sometimes it is the simple solutions that work best," he says.

Costs of predictive analysis
Given that most companies have a spreadsheet application already, the main cost of predictive analytics will be in setting up the software and collecting the data. This could cost anything up from a few thousand pounds.

For example a basic, single user licence for SPSS costs £1,300, but server licences for analytics software can cost from £25,000 upwards, depending on the users and feature sets.

At the top of the range, a full-scale analytics project based on software from SAS or one of the large business software vendors, such as SAP or Oracle, can cost more than £100,000.

Whatever you spend though, the technology is only as good as the methodology around your research. Much will depend on the type of questions businesses are trying to answer.

"We've seen some clients start with a proof of concept that can be done with manual number crunching or a spreadsheet," says George Johnston, of the professional services firm Deloitte. "But if you want to go to the far end of sophistication with high volumes of disparate data and high degrees of automation, you are talking about more money."

Moving to predictive analytics, though, is about more than simply buying new software. Companies need...

...historic data - most will have this - but they also need clean data.

The precision of forecasts will depend on the accuracy of data, as well as data quantity and how broadly the datasets are based.

In its simplest form, clean data means checking that information is up to date and looking for duplicate records, but even less than perfect data can be used as long as the analysts understand the shortcomings.

The model - the formulae used to analyse the raw information - needs to take into account both data quality, and how the business wants to use the predictions. A highly targeted (and expensive) marketing campaign will need a different model to analysis looking at broad demographic trends.

And, although automated predictive analysis systems come into their own when handling large volumes of data - analysis a sales executive would struggle to carry out with a spreadsheet - any model is only as good as the business knowledge of the person designing it. The deeper the understanding of the workings of the business, the better any predictive analytics model will be.

"It is also about how thoroughly you can understand the variables that drive the decision," warns Accenture's Harris. "If you don't understand something, you have no business automating it."

As a result analytics still needs someone with a knowledge of the business in charge: data mining, for example, will always need a marketing person. Using predictive analysis effectively also means understanding that it is a methodology, not a static technique for producing definitive answers. As the banks found during the sub-prime crisis, changing events can leave even the best models behind.

"As things change you have to keep your models up to date," says Colin Shearer, senior vice president of strategic analytics at SPSS.

"If your customer behaviour and preferences change, you have to keep up with that. In the credit crisis the pattern of default changed rapidly. Banks should have been updating their credit models more quickly."

For businesses willing to put the effort in, predictive analytics is one of the most cost-effective ways to make more of an asset they already own - their data.

Five steps for effective predictive analytics

  • Understand that data quality is important but that it is also important to understand how comprehensive the data is - usually the more data, the better the results.
  • Ensure people who really know their part of the business help design the analytics, even if non-specialists will run the models.
  • Start with a small project and prove that it pays its way before committing to large investments.
  • Compare any new analysis against the results of previous campaigns, to measure accuracy.
  • Keep reviewing the analysis set up: what works today might not be accurate tomorrow.

Editorial standards