Kadmy - Fotolia
AA Ireland turns to predictive analytics in fraud crackdown
Roadside assistance company invests in analytics software to fight fraud and introduce dynamic pricing after its split with the AA
Dublin-based roadside assistance organisation AA Ireland is turning to data analytics technology to combat fraud.
The company, which offers home, motor and travel insurance, said fraudulent applications can add up to €50 to the cost of a typical motor insurance premium.
The initiative will enable the organisation to analyse historic insurance quotes for discrepancies that could indicate fraud.
The same technology will allow AA Ireland – which has 350,000 customers – to adjust its pricing to changes in the market, optimise its prices for each client, analyse risks more effectively, and provide better management information, said Colm Carey, head of pricing and management information.
The project is expected to pay for itself within 18 months.
“Fraud is a huge issue in the industry – it probably adds €50 to every €700 policy,” said Carey in an interview with Computer Weekly. “It is one of the big call-outs and we need to get a lot better at using data.”
AA Ireland’s quantitative analytics team is building predictive models that will analyse customers’ quote history to identify attempts to manipulate quotes with false information to secure a better price.
“What we have found is that people who have manipulated a quote four times for specific factors had a 70% higher chance of getting into an injury claim,” said Carey.
The company plans to use a technique known as “fuzzy matching” to analyse a range of quotes to identify people who might have applied for quotes using slight variations in their data, such as different spellings of their name or a different address.
“Fraud is a huge issue in the industry – it probably adds €50 to every €700 policy”
Colm Carey, AA Ireland
The history revealed will allow AA Ireland to offer quotes that are more closely tied to the potential risks of each applicant.
“You might be looking back at five or six million records, fuzzy matching in a live environment, with no lag, to find all the quotes for a particular individual, see how they have manipulated quotes, and price effectively off the back of that,” said Carey.
AA Ireland expects to have its anti-fraud system up and running within 18 months.
Off the radar
The company had previously relied on software widely used by the insurance industry, Emblem, and Radar insurance pricing optimisation software, supplied by Towers Watson, to manage its product prices.
But the technology meant that each business unit operated independently,with little integration of the different stages of the insurance process.
The AA parent company sold AA Ireland to Carlyle Group & Carlyle Cardinal Ireland for €160m in July 2016, which necessitated splitting the firm’s IT system.
Carey said AA Ireland decided to take a risk by moving away from the industry standard software to software with more analytical capabilities.
It looked at four or five alternative software platforms and rated them against a 14-point checklist, which included the ability for the software to optimise pricing based on an analysis of customer data, and the ability to perform real-time analytics, with a maximum lag of 200 milliseconds.
The IT team quickly discarded two of software platforms because they could not operate quickly enough on the company’s existing hardware.
It selected software from Palo Alto supplier Tibco, choosing the company’s Insight Platform, a suite of tools that would allow AA Ireland’s quantitative analytics team to design and build predictive models with the ability to learn and improve their performance.
The AA had previously used Tibco’s software in 2015 to create an “integration layer” to allow its software platforms to exchange data more effectively.
The project will allow AA Ireland to take a more coordinated approach by linking data from its call centre with its customer relationship management (CRM) software and pricing software.
“We were all going off and doing different jobs and not talking to each other – so we wanted to take a step back,” said Carey.
For example, under the current system, a sales director could decide to push through a large number of quotes to meet the sales quota in the same week that a large number of motor insurance customers are due to renew their policies – putting a strain on the call centre.
AA Ireland has gone live with dynamic pricing for new membership business, and is due to go live with membership and mid-term adjustments soon.
The organisation is able to calculate, on the basis of probabilities, what type of customer each person approaching the AA is, to advise how quotes should be priced and what type of discounts should be offered.
The software will compare the outcome predicted with what actually happened, allowing AA Ireland to optimise its pricing, for example to maximise profits or to maximise growth in the market.
Previously, the company had to update its pricing model manually every three to six months. In future, it will receive an automated report every week that will allow it to update its pricing in response to significant changes in the market. The software will also be able to learn and make adjustments to smaller changes in the market each day.
“Let’s say there was a specific insurer that stopped writing a particular age group or occupation,” said Carey. “Maybe we want to reduce our discounts because we don’t need as much discount to write that particular customer, and we can spin the discount better to bring in a different type of customer.”
Learning from the past
The project has enabled AA staff to understand the history of each customer, based on the type of vehicle breakdowns they have had, and previous conversations with the company.
Agents will know when customers have already told the AA they are not interested in particular products, and will not annoy the customer by offering the same products repeatedly.
“So you are trying to price the customer for what they actually really want,” said Carey.
The IT team had to meet a tight deadline to develop a dynamic pricing model on the new business platform before the company turned off its existing software in April 2017.
AA Ireland’s data analytics technology
Tibco Insight platform:
Allows the AA team to build applications to analyse streaming data.
Allows AA Ireland to build models that can respond to real-time events and generate insights that can be acted upon.
Visual analytics software
Tibco Enterprise Runtime for R – a statistical engine that allows the AA to build data analysis programmes using the R programming language.
At the same time, the organisation was moving its servers and data warehouse from its former parent company.
“More time would have been good, but we really did not have the choice,” said Carey.
The project has meant that business units, rather than the IT department, will now be responsible for running pricing models.
“In the new world, the pricing team has become more a semi-pricing, semi-development, semi-user acceptance testing and semi-support team,” said Carey. “If something goes wrong, it’s on me, so it’s a lot more responsibility.”
That has led to some early discussions about the changing shape of the company, and what sort of people it should recruit in future.
“It needs someone who can take an automated report and understand it and then work off the back of that,” he said. “We have had conversations about what the business is going to look like in a year’s time – will we need the same sort of people?”
Carey said that the quantitative analysis team would increasingly take responsibility from other business teams for analysing data.
The work would free up the marketing team from having to manipulate and analyse data, to focus on generating more business, for example.
Learning IT speak
For AA Ireland’s quantitative analysis team, the biggest challenge was learning “IT speak” – the language of IT developers.
“It was a big challenge for us to learn how to use the software, how to develop in Streambase,” said Nina Jensen, senior quantitative analyst.
Tibco consultants came in to train the analysts, but after a few months the team could develop independently. “It was a very steep learning curve,” said Carey.
The team rebuild a version of Radar, to run in Streambase, in just three weeks.
AA Ireland is still in the process of separating from its parent company, but expects to complete the move to its own servers over the coming weeeks.
It has begun work on a project to feed call centre data from its Unify telecommunications software into Tibco’s data analytics tool, Streambase.
Due to be completed in the next 18 months, the move will allow the company to predict the capacity needs of the call centre every hour, every half-day, every week and every month, said Carey.