What every CEO needs to know about AI. Part two: returns

Looking at how motor insurance claims are processed gives a powerful illustration of how applying the whole suite of AI technologies creates impact:

• Insurance companies have been using optical character recognition (OCR) to convert handwritten claims into machine-readable data. But OCR has not always been accurate. Now, California-based Captricity, using its Shreddr technology, reckons to achieve accuracy of more than 99.9%.
• Robotics can take over manual tasks such as creating the first notice of loss by verifying an applicant’s identity, searching policy details, entering loss damage and repair estimate information, checking policy coverage, uploading photographs and putting them into the right assessment queue.
• Computer vision can help verify claims. Ageas, Britain’s third-largest motor insurer uses computer vision technology from London-based Tractable to speed up the assessment of repair costs. The Tractable system stores millions of collision repair images and the corresponding repair estimates. Hence Ageas can check thousands of estimates every minute and flag up unnecessary repairs to its experts.
• Natural Language Generation can be used in writing narrative reports, identifying information from claims and repairs and then detailing trends, explanations and potential next actions.

Over the past two decades, companies have adopted lean and automation to reduce waste and boost efficiency. As further improvements were becoming tougher to achieve, AI has turned up as the white knight allowing them to move to the next level of productivity.

• Combining advances in the Internet of Things (IoT) , which collects data from industrial machines, with AI have allowed companies to reduce asset downtime. AI-based algorithms monitor sound or thermal signatures at the equipment level plus other data such as maintenance logs and weather patterns. If the system then spots something is unusual, it can flag up the need for maintenance and thus reduce downtime. Pitney Bowes has introduced General Electric’s Predix system and has seen a 20% increase in machine yield, a 15% saving on parts replacement and has cut the cost of tech support by 10%.
• Worker productivity can be boosted by using “collaborative” robots that operate with human workers on jobs that cannot be fully automated. Computer vision allows a robot to be aware of what is around it. A human worker can ‘programme’ the robot by merely taking the arm of the robot and guiding through the desired movements. The human is still required, but his or her productivity is vastly increased.
• AI can make a big impact on energy consumption. An outstanding example is Google’s DeepMind, which has cut the energy used by Google for cooling its data centres by 40%. Algorithms looked at historical data for things such as temperature, power and pump speeds to predict future temperatures and then decide optimum settings for minimising power use. A similar approach could be adopted in energy-intensive industries such as aluminium, cement and paper.
• Quality control can be speeded up by combining computer vision with machine learning to streamline inspection, detect anomalies more accurately and give consistent results. Armed with images of products that are good and those that are defective, an AI system could also identify previously unknown defects.

Machine learning and robotics already allow companies in retail, packaged consumer goods and high-tech sectors to transform their supply chains into a source of competitive advantage. Companies are using AI to accurately forecast customer demand for vast numbers of different items. They then complement the improved forecast with flexible and efficient supply. They have automated warehouse processes using robots for picking and packing; and made transport for supplies and delivery responsive to real-time information such as weather forecasts and traffic flow.

German e-commerce merchant Otto can now predict with 90% accuracy what will be sold over the next 30 days, reducing its surplus stock by a fifth and product returns by more than two million items a year. Algorithms look not only at historical sales, but also take account of advertising campaigns, store opening times, local weather and holidays to predict demand for individual lines from each outlet.

Amazon uses Kiva robots which bring packages to human workers standing on platforms. Amazon has increased inventory capacity by 50% and reduced operational costs by 20%.

AI can help companies meet customers’ ever-rising expectations for personalised service and for getting what they want straight away.

• Reduce incoming failure demand: When something fails to meet customers’ demands, they are likely to contact a call centre. Identifying what has gone wrong both helps avoid repeating shortcomings and reduces demand for call centres. Transcribing and analysing customer calls can identify the root causes of complaints. Accurate transcription of calls is becoming increasingly possible. Google, IBM and Microsoft have systems that can now transcribe phone conversations with accuracy of 95% or thereabouts. Using Natural Language Processing and text mining, calls can be analysed to derive trends of demand and patterns of failure; then, appropriate action can be taken.
• Reduce the number of calls reaching call centre agents: Online chatbots that mimic how humans speak are already widely used. In the banking sector, chatbots are answering simple queries 24x7 and helping customers through everyday transactions such as checking balances and making payments. Chatbots act as gatekeepers, dealing with simple tasks, then transferring calls they cannot handle to call centre agents. The next evolution would see chatbots paired with robots that could carry out end-to-end transactions such as cancelling the stolen credit card and issuing a new one.
• Eliminate the time spent on authentication: Call centres spend on average 30 to 45 seconds simply on verifying the identity of a customer. This is equivalent to around 100,000 hours for 10 million calls per year. Banks such as HSBC and Citi have deployed voice biometrics from companies such as Nuance and Nice to eliminate this wasted time. The system compares a caller’s voice with a stored voice signature. Facial recognition is also emerging as a viable technology – particularly in China. Many Chinese banks allow users to identify themselves using only their face for online access and at ATMs.
• Speed up the handling of calls: Call centre agents typically spend up to a quarter of their time accessing different systems to find and update data during and after a call. A simplified robotic desktop console that allows data to be automatically filled or verified can drastically free up time for an agent to focus on other aspects of their job. HMRC has introduced a dashboard that cuts the number of times an agent has to click their mouse during a call from 66 to 10. The result: the average time taken to handle a call has been cut by 40%.
• Coach the agents: Call transcription can help call centre agents to learn from the way that their best-performing peers operate. As chatbots increasingly cover simple queries, the agents will be left to handle complex queries. Transcription based coaching will help them make the transition smoother. Improve capacity utilisation: Call centres can use machine learning models such as Random Forest and Facebook’s Prophet to accurately forecast call centre demand. Accurate forecasting six months in advance will help call centres take better recruitment and skill-mix decisions. Accurate forecasting a week in advance would help call centres in better scheduling of agents.

Purchased goods and services typically make up 60-80% of a product’s cost. AI can arm purchasing managers with information and insights to secure better deals. But procurement data in many companies is uncleaned, unclassified by categories of spend and spread across different systems. However, Coupa software uses machine learning to automate the clean-up and classification of information. It claims to have analysed more than $1.3 trillion of spending. Businesses can further reduce procurement costs by plugging leakages arising from non-compliance and inefficiency by automating order management, especially for long-tail spend. Typical leakage is 3-4% which is equivalent to £180-£240 million on a £6 billion spend.

There is vast scope for using AI to reimagine administrative processes. Imagine a system that takes a picture of a receipt and uses optical character recognition to read the text, with machine learning then identifying the sum, date, currency and type of expense: put it together and it could create an expense entry.

As human resource departments step up from their traditional service role to strategic business partner responsibilities, they can take a cue from Michael Lewis’s book Moneyball to derive data-driven insights about talent.

• Find potential high performers: Forward-thinking HR teams are using data science to identify the traits that make someone a high-performer and to define the profiles of employees who are likely to do well in a particular role. Also, AI can highlight where high-performing employees come from and identify promising sources of talent.
• Reduce cost and time to hire: Early adopters have embraced algorithms to speed up the screening of CVs of applicants. Algorithms can identify the most suitable candidates who are most likely to be hired and hence allow experts to focus on the prioritised pool. Companies can use AI to automatically update millions of resumes buried in their candidate databases with the latest work history pulled from the web. Beamery, a customer relationship management company, allows employers to engage with potential candidates long before they even apply for a job. Its algorithms drip feed content, news and information about jobs to candidates based on their interests. It claims that its clients have seen a trebling in the number of qualified candidates and that recruitment costs have been more than halved.
• Reduce attrition: Employees leave a trail of their job-seeking on social media. Social media metadata and historical HR data can help companies pinpoint which of their employees are most likely to quit. An employer can then work out ways of holding on to its best performers.

Abhijit Akerkar (MBA2008) is Head of Applied Sciences, Business Integration, Lloyds Banking Group

What every CEO needs to know about AI. Part three: managing risk

What every CEO needs to know about AI. Part one: growth