By Neale Swinnerton - Senior Data Engineer - infoNation
Data Science, Artificial Intelligence (AI) and Machine Learning (ML) are having a big impact across many sectors. Those leading the charge are those where data, the fuel to our algorithms, is readily available. More and more businesses are starting to realise that data can be extracted and used to drive business growth. There are practical, moral and ethical issues to be wrestled and it's important when applying AI/ML to understand that it's a tool to support the business rather than a silver bullet. There are real risks too, particularly when processing Personally Identifiable Information (PII), and regulators are starting to catch on. The EU's recent General Data Protection Regulation (GDPR) has real teeth in terms of enforcement options.
With that in mind here are some general areas where AI/ML is broadly applicable across any business.
Application 1 - Marketing
Modern marketing initiatives generate a wealth of data. Smart business use that data in feedback loops to validate and improve the efficiency of their marketing spend. By using Machine Learning marketing efforts can be more precisely targeted. With well defined metrics the success of any campaign can be quantified. In a traditional model, customers might be segmented by, say, age or geographical region. With ML clustering algorithms such as K-Means or Expectation Maximisation of Gaussian Mixture Models, allow segmentation into clusters with more subtle characteristics or higher dimensionality. This allows marketers to both experiment with complex marketing strategies and measure their success, often in real time as the campaign in rolling.
Application 2 - Sales Recommendation
As more and more sales move online, businesses are rapidly evolving how they engage with customers at the point of sale. Recommender systems have really taken off in this area, where users are presented with suggestions of what they may like to buy before, during and after a sale has completed. These systems leverage knowledge both of the customer and the product, recommending products similar to those bought or products previously bought by customers. Algorithms such as collaborative filtering where we measure the interactions of users with our systems and use those interactions to simulate 'collaboration' between the users to reach a decision and content-based filtering where we look at the history of the user or product and use that to inform decision making. Increasingly recommenders use a hybrid approach.
Application 3 - Business Intelligence
As businesses become more complex, so the challenges of managing forecasts, fraud and risk become more difficult. With its basis in statistics, ML allows development of sophisticated forecasts. With the addition of AI-based expert systems it can be possible to capture and incorporate data from more 'opinionated' or intuitive sources. Anomaly detection allows identification of 'odd' transactions or behaviours that warrant further investigation. With respect to risk analysis, the challenge is one of scale, analysing large amounts of historic data to quantify the probability of risk events in the future.
Application 4 - Operational monitoring
Businesses have many corporate assets with varying degrees of sophistication. Monitoring the performance of those assets is increasingly possible and desirable. Much of the hype in the Internet of Things space will be realised here. Applications currently in use include room/desk utilisation allowing more efficient usage of the estate. With an increasing awareness of environmental impacts, measuring the efficiency of buildings and vehicles becomes useful and optimising power usage, for example can lead to real savings. Many of the cloud compute providers support variable, demand based pricing and knowing when you need capacity and often when you don't need capacity can greatly reduce cost and environmental impact.
Application 5 - Human Resource Management
Increasingly HR departments are using AI/ML in the hiring process. Natural Language Processing algorithms allow candidate CVs to be pre-screened. Companies are realising that automated systems can, with care, reduce conscious and unconscious bias in hiring. Increasingly companies publish diversity statistics and are measuring that diversity in hiring can have a business benefit over and above the moral and ethical obligations. Similarly performance reviews can be processed using Topic Modeling to look for themes, positive and negative, occurring across the reviews. Analysis of the data may suggest new options for optimizing the composition of a teams within an organisation.
More from our blog: