Using analytics to predict the future and do better

Anton Antich from Veeam Software outlines the benefits of applying data science and machine-learning techniques to historical data

Guest
By Guest on 30 November 2015
Using analytics to predict the future and do better

This article first appeared in the Autumn 2015 issue of OnWindows magazine.

Predictive analytics systems are now among the emerging software areas that drive business efficiency. A wave of new startups removes the hurdle of employing teams of scientists, computer engineers and mathematicians for big data analysis and makes analytics platforms and services available for companies of all sizes. While updating their top industry trends for 2015, Gartner analysts included ‘advanced, pervasive and invisible analytics’ as one of the top ten strategic technologies. IDC, meanwhile, forecasts that the advanced and predictive analytics software market will attain a 9.9% compound annual growth rate (CAGR) by 2018. With this in mind, today’s businesses can make even more efficient data-driven decisions based on predictive analytics.

Veeam Software looked extensively at the emerging landscape of predictive analytics, specifically predictive lead scoring platforms, when searching for a way to improve its sales ­efficiency and optimise marketing. Over the last six months, Veeam has engaged with 19 predictive lead scoring vendors. While many claimed fantastic results from their existing customers, we soon realised that we couldn’t just rely on customer references to select our preferred solution partner. We needed to take a rigorous testing and benchmarking approach to compare the quality of prediction that each solution provides.

Behind the scenes of any predictive lead scoring system lies a machine learning mechanism. It detects the patterns and dependencies in historical data volumes and builds a predictive model via a learning algorithm. To evaluate the accuracy of predictions we sent each potential vendor two historical sets of data from our CRM database. The first dataset contained information about opportunity closing and was used for machine learning. The second dataset was used directly for predictive lead scoring testing and went without deal closure information. Comparing the vendor’s results with our real data, we could easily evaluate the quality of prediction. Our proof of concept really helped in shortlisting — while some vendors showed close to 100% accuracy in predictions, other ones hardly reached 30%.

While the blind test on prediction accuracy helps to choose a platform with a reliable underlying algorithm, two other factors are important to consider.

First, predictive analysis based on a higher amount of external account and lead signals usually brings better results. It’s important for a predictive model to combine a good machine learning algorithm with an ability to enrich existing data. An ideal vendor can mine data on companies’ websites, job boards, social networks, government filing and data providers’ databases to create a full profile for each lead in a database. Enriching the scope of analysed data with the variety of additional buying signals and marketing information allow a predictive lead scoring system to provide more accurate and rich outcomes. We found that some predictive lead scoring solutions lack additional data, while others limit it to extracting persons’ interests and job functions from a single social network.

Second, it’s important to understand the value each signal brings to scoring because it will help to focus marketing efforts related to future demand generation. For example, if one of the signals shows that in a specific geographic area the conversion rate is higher, you might want to adjust the demand generation accordingly. In relation to sales teams, consider the following overlooked point: the ability to explain to sales teams why certain leads rank higher. This will drive for more confident sales conversations. Unfortunately, many predictive vendors see the signal scoring as their proprietary knowledge and offer little to no insight.

Predictive platforms can help a company to drive their business with real-time decision-making based on hundreds of data points. The implementation of smart lead scoring is a first step for Veeam on the way to further sales process improvements. Through sales team action (meetings, content pieces presented, call scripts used, etc.) analysis, we aim to create a self-learning sales assistance system. This will help by suggesting the best next step to our sales reps based on the customer profile and historical win/loss data to maximise closure rates and shorten sales cycles.

While automated marketing is a hot area right now with a lot of vendors and buzz, using machine learning for more complicated sales processes, from marketing lead generation and nurturing to direct sales touch to channel management, is a relatively new area, which should not be overlooked in order to continue beating the competition.

Anton Antich is senior vice president of strategic operations at Veeam Software

 


Number of views (3681)/Comments (-)

Tags:
Comments are only visible to subscribers.

Theme picker