Liz Goli, Commissioner of Queensland’s Office of State Revenue (OSR), sat back in her chair: “…in that way, the machine can actually improve our empathy with our customers”, she reflected . Now that’s interesting – the idea that an unfeeling machine could help human beings be more empathetic towards other human beings! Late last year, OSR implemented a successful machine learning prototype with SAP, and they’re now moving forward with a production pilot of this exciting emerging technology. “We don’t want a system where the machine is making decisions. But we do want the machine to offer up next best action recommendations to our staff, that they have the option to follow – or not – based on their experience and knowledge of how the legislation should be applied… We’d also like a system that can ingest big data and take action within certain parameters. For example, in case of a natural disaster, the machine might be able to find out which customers are impacted and replace debt collection notices with proactive letters giving additional time to pay”.
From action-reaction to proactive and personalized
The Office of State Revenue (OSR) is responsible for collecting taxes and royalties, and administering the First Home Owners’ Grant for Australians residing in the State of Queensland. The revenue collected by the Office provides about A$17 billion (€11 billion) annual income for the State, which is reinvested in roads, schools, hospitals and community services. With less than 500 staff servicing over two million taxpayers, the Office needs to deliver highly-efficient and automated services, while minimizing costly and time-consuming manual processes. Moreover, OSR needs to unlock the information contained in the Office’s vast data holdings to deliver the kind of customer-centric digitally-enabled services desired by government, businesses and the community.
Timely collection of taxation revenue is key to the government’s ability to fund essential services. But each year up to 5% of revenues are uncollected by the due date, amounting to an A$882 million (€555 million) liability in FY16. Default on Land Tax is particularly high, with over 15% of revenues uncollected by the due date, amounting to a A$112 million (€70 million) liability in FY16. For most of us it’s difficult to fathom how such sums of money could be recovered, but as described by Ms. Goli, what happens at an individual level is actually quite simple: “…we expect the debtor to pay, and if they don’t then we start to remind them, and after each reminder we sit back and wait. Our process can be described as action-reaction, action-reaction – every action we do is supposed to prompt a reaction from the customer. But because we haven’t historically done a lot of analysis about what reactions our actions provoked, we haven’t always understood our customers’ motivations”.
Therefore, to achieve their strategic objective of reducing liabilities, Ms. Goli and her team knew that they first needed to understand what factors lead some customers to pay on time, whilst others do not. The challenge was how to uncover the insights buried within the Office’s big data holdings – this is where machine learning came in. The Office’s machine learning prototype analyzed 187 million records to provide a prediction of risk by taxpayer, and identify the events and influences that ultimately led to payment default. These may be things that OSR has control over (e.g. processes and interactions), things that the government dictates (e.g. policy and legislation), or external impacts (e.g. natural disasters). The machine makes the links between cause and effect, enabling the Office to be proactive in their responses and personalized in their treatment.
Understanding customer motivations
But at the individual level, how do you begin to understand the motivations of someone who you’ve never even met? The answer lies in visualization of their journey. Ms. Goli explains: “…traditionally we’ve worked with data in spreadsheets, but we’ve discovered that data visualization is really important. People are visual, and we’re better able to identify patterns with a visual representation of data than with data in a spreadsheet”. So, it’s not only the surfacing of key events and influences, but also how these are presented on a timeline that enables staff to truly understand customer motivations.
In the example of one high-value taxpayer, OSR discovered that his behavior over five years has been to ignore the Office’s debt collection notices until he receives a final legal notice, at which point he promptly settles his debt. Visualization of this particular customer’s journey caused OSR to conclude that his behavior is not motivated by an inability to pay on time, but by a deliberate tactic of delayed settlement. Now the Office has the insight required to design a debt collection strategy for this cohort of one. “We can write to him explaining that we’ve noticed that he only ever pays on the final notice, so we’re not going to bother him anymore with multiple reminders – from now on he’ll get one reminder, then the next letter will be a final legal notice. Equally, for taxpayers who typically do the right thing but are non-compliant in a particular instance, we can design a strategy for them”.
A right from the start approach
Ultimately, the Office’s debt collection strategy is all about proactive compliance. In this respect, OSR has borrowed the mantra right from the start from the OECD’s Forum on Tax Administration (FTA). The FTA aims to influence the environment within which tax systems operate, to move from a confrontational dialogue to more constructive engagement with taxpayers. Their proactive compliance approach recognizes that taxpayers are motivated by perceptions of: deterrence (the risk of detection and the severity of punishment), norms (both personal and social), opportunities for non-compliance, fairness (distributive, procedural and retributive), economic factors, and interactions between the taxpayer and the revenue office .
Right from the start emphasizes the need to create an environment that encourages compliant behavior by: acting in real time and up-front; focusing on end-to-end processes; making it easy to comply (and difficult not to); and actively involving and engaging taxpayers to achieve a better understanding of their perspective . For OSR, this translates into four policy and practice strategies: designing risk-based revenue management interactions; fostering meaningful relationships with customers and partners; developing enhanced services through digital methods; and building a capable, change-responsive workforce. Perhaps most importantly, the Office is leveraging the insights gained through machine learning to redesign their business processes with the customer at the center.
Customer-centricity is about efficiency and confidence
It’s not uncommon to for retail expectations of customer self-service to be transposed onto digital government initiatives. But as Ms. Goli explains: “…in a government context, customer-centricity isn’t about providing a retail-like online shopping experience. It’s about providing a highly-efficient service where people have confidence that they’re receiving the right information at the right time”. To achieve this, OSR needs to leverage their big data assets and apply them in a transparent way. “We want to get to the point that what we see is what they see. We’ll show them what we know about them, and they can correct it with us. This will create a mature relationship built on mutual obligations, where we trust them and they trust us”.
But customer-centricity isn’t just about delivering a great customer experience – it also has a role to play in delivering the Office’s proactive compliance objectives. Where traditionally revenue offices tend to look at compliance tax-by-tax, a customer-centric approach checks whether the taxpayer is fully compliant across all their tax affairs. For example, some businesses might always ensure that they’re compliant for one type of tax where the consequences for non-payment are greater, but at the same time they’re consistently non-compliant for other types of taxes. In this way, a customer-centric view gives insight into the taxpayer’s true compliance behavior. This might cause the Office to reassess whether they should continue to offer payment arrangements for one tax type to someone who is a serial late-payer of other taxes, or whether they should be taking the standard approach with someone who is generally compliant across all their payment obligations.
Further, since debt is often a leading indicator of hardship, a customer-centric approach can highlight instances where a taxpayer might be struggling or a business might be failing. This could prompt the government to proactively reach out to the customer with an offer of assistance. When asked to summarize how she expects machine learning to change the way the Office engages with taxpayers in the future, Ms. Goli replied: “…in the midst of all the digital, people want a human connection more than ever before. A connection that is proactive and personalized. Machine learning will provide OSR with a capability to deliver this to our customers, completely transforming our engagement in the future”.
Enabling data-driven policy and practice
By enabling evidence-based decision-making, machine learning is fundamentally changing the ways of working at OSR. Ms. Goli and her team see the potential for:
- Manual decision-making based on only a small percentage of the data that is available, to be replaced by machine-generated proposals based on all available data;
- Revenue agents to be able to leverage the insights garnered from machine learning whilst on calls with customers to better understand their situation and provide enhanced levels of service;
- Collection agents to be freed up from actions that drive little value, to focus on interventions that will make a real difference to both the Office’s customers and revenue outcomes; and
- Risk profiling and segmentation to be used to drive more proactive campaigns and compliance activities, aligning with the Office’s risk-based revenue management approach.
The insights gained through machine learning also have the potential to be used as input into future policy development. “We now have the evidence to support our advice that if you design it this way, this is what the likely reaction will be”. Thereby, data-driven insights can help strengthen the voice of the administrative arm of government to policy-makers, influencing legislative change based on service delivery experience.
 Interview conducted 22 February 2018.
 Organisation for Economic Co-operation and Development (https://www.oecd.org/tax/administration/46274793.pdf).
 Organisation for Economic Co-operation and Development (https://www.oecd.org/tax/forum-on-tax-administration/publications-and-products/admin/right-from-the-start-influencing-the-compliance-environment-for-smes.pdf).