By Dan Somers, CEO of Warwick Analytics (www.warwickanalytics.com)

Processes within the financial services industry have become more automated in recent years. Processes such as customer services, spotting fraud and error, credit scoring and processing insurance claims are becoming more automated thanks to predictive analytics and machine learning (sometimes referred collectively as “AI”).And there is plenty of evidence that automation is becoming more widely known and accepted. Indeed The Financial Times has reported that the CFA Institute is updating its Certified Financial Analyst (CFA) exam so that starting in 2019, it will include questions about artificial intelligence (AI), big data and robo-advice, reflecting the growing impact of machine learning on the finance industry.

But whilst it all sounds encouraging, the roll-out isn’t as fast as many commentators expect or hope. According to a PwC Report 2017, only 30% of large financial institutions have invested in AI. This could be because much of the core technology hasn’t changed for decades, be it decision trees, neural networks, Bayesian statistics etc. The practical application is also limited because of the amount of work that data scientists need to spend building and then curating machine learning models. Most importantly however, the complexity of the financial services industry is increasing, driven by ever-changing consumer behaviour and expectations, new disruptive businesses, the sophistication of fraudsters, the explosion in data (including a lot of unstructured data) and indeed more regulation.

This complexity is slowing the progress of AI, as it requires more data scientists to train and deploy the algorithms and cleanse and handle the data. This is particularly true when the processes and datasets involve unstructured data such as text:In a recent survey carried out by Warwick Analytics on AI in text, surveys, social media and queries/complaints were identified as the most common datasets being used. However, most analysts (53%) using text analytics wanted more insight from it and nearly half of those (23% out of the 53%) were not satisfied with the output of the analysis that they were getting.

Customer expectations are changing, with more interactions across more channels, and larger, richer datasets and increased need for personalisation and segmentation.Criminals are also getting more sophisticated with their own technology and organisation and financial institutions needs to move more swiftly to stay ahead.In the larger institutions there are still many humans interacting with customers and making operational decisions in front-, mid- and back- office which could be done more effectively by, or with the aid of AI (sometimes called “Robotic Process Automation” or “RPI”).

These new challenges within the industry require more sophisticated technologies and solutions and long-established business models are being disrupted by fintech newcomers that are creating new services, disintermediating the traditional value chain, and driving down costs. 88% of incumbents are increasingly concerned they are losing revenue to rival innovators (PwC Global Fintech Report 2017).

The good news is there are AI solutions appearing which help to automate data science itself, and minimise the input (both the time and skill-levels) necessary from humans to adopt and deploy.

One such example of this is Optimized Learning (OL) which is a technology developed by a company called Warwick Analytics (originally a spinout from The University of Warwick). OL is a type of machine learning that automatically analyses and structures heterogeneous and unstructured data (i.e.text) to provide automation and actionable insight. This text data can be anything, for example Voice of Customer (“VoC”) data such as CRM notes, voice transcripts, queries, complaints, reviews, surveys and social media. It can also be data specific to financial services such as investigations, bank statements to automatically generate credit reports as well as news and financial reports to generate insight and automation for investment and trading algorithms. The key point to Optimized Learning is that it ‘learns how to learn’ and asks for human intervention in a minimal way in order to limit the time and skill-level required to train and deploy. This greatly lowers the time and cost to deploying AI and rapidly unlocks the many identified use cases where making sense of heterogeneous data are involved, without hiring an army of data scientists.

Some use cases where this type of technology is being adopted today are (i) automating contact centres by analysing customer interactions (voice, CRM notes, chat, emails etc.) to assist agents, automate responses and improve chatbots and FAQs to prevent enquiries in the first place as well as (ii) gleaning what people are talking about from the VoC data and which issues cause positive or negative sentiment (e.g. driving customer loyalty or churn) to improve the customer journey and customer experience as well as (iii) personalisation, customer segmentation and targeted marketing. Similarly, (iv) complex automation on business processes such as claims handling, fraud and error investigation, and (v) enriching credit scoring by analysing e.g. bank statements and even social media profiles of customers. It has even been used to (vi) help with recruiting and staff monitoring by analysing CVs and monitoring emails (where appropriate) to predict employee happiness and identify skill gaps and training requirements.Within the wholesale financial industry it can help (vii) with predicting micro and macroeconomic events that drive stock price movements, classifying relevant signals from the text in reports, news and social media and in particular the concepts not just the keywords or named entities. Many of these use cases are already in play, but their current accuracies and the manpower required to develop and curate them reach a certain limitation. With technologies such as OL, those boundaries can be dramatically pushed further by minimising the effort and skill levels required to optimise the models whilst building it into existing processes without the need for separate teams.

And there are of course other novel AI technologies and applications appearing, that help not only to innovate but to allow other disruptive innovations to happen by facilitating better insight and automation. Disruptive predictive analytics technologies are here to shake up the world of financial services, and with AI automation technologies the disruption is set to get faster.