Our use of Artificial Intelligence (AI)

We use AI because it helps us to improve the services we provide and the productivity of our staff. We are careful in how we use AI and consider how it affects our customers and Aotearoa New Zealand.

Our priorities and guidelines

Our AI governance follows All-of-Government directives. We are committed to keeping our customers safe and have signed up to the Algorithm Charter for Aotearoa New Zealand.

Our priority is to make sure we use AI in a way that considers our obligations under the Revenue Acts, the Privacy Act, the Algorithm Charter and all other New Zealand Government authoritative guidance. This includes embedding a Te Ao Māori perspective in the development and use of algorithms consistent with the principles of the Treaty of Waitangi. 

Algorithm charter for Aotearoa New Zealand - data.govt.nz

Our staff

Our staff must use our Artificial Intelligence staff use policy and Artificial Intelligence use case guidelines. These set out our approach to using AI safely and securely in the workplace. They help us to make good decisions and deliver effective and efficient services.

We have existing governance groups who decide when we will implement AI tools. We have established 2 additional governance groups to identify and test AI tools and solutions.

Use of algorithms within Inland Revenue

Many of the products we use have integrated rules and algorithms to support productivity and efficiency. We use many business rules which enable the automation of business processes. These support our service delivery to both staff and customers.

We also have machine learning algorithms which work across large data sets to provide advanced analytics, decision support and predictive modelling. All activities undertaken as a consequence of these algorithms are subject to human oversight and human decision-making.

Uses of Artificial Intelligence within Inland Revenue

Our use of AI focuses on supporting decision-making in the management of tax/social policy compliance risk, intervention design and on increasing staff productivity.

The table below details all current uses of AI within Inland Revenue.

The table does not include any use of AI where publication could adversely affect the integrity of the tax system or prejudice the maintenance of the law.

More information

If you would like official information about how we use AI and algorithms, you can contact us online, or by phone or post.

Official information requests

Uses of AI within Inland Revenue

Name	Description	Product status	AI technology	Use
ABBYY FineReader 16	Text recognition and document conversion tool, used to convert PDFs into excel.	In production	Optical character recognition	Staff support
AI Futurist	Enables querying and summarisation of content.	Available	Generative	Staff support
Āwhina Mai	Customised guidance for customers completing forms and reviewing guides on our websites	Pilot	Generative	Supporting customers directly
Assurity Intelligence	Test scenario generation	Pilot	Generative	Staff support
Coveo	Our public websites, internal staff intranet and knowledge base use Coveo as a search platform. Coveo uses machine learning to continuously learn and improve from user searches and patterns to inform the best results to display.	In production	Machine learning	Supporting customers directly and staff support
Creative desktop applications – Adobe Acrobat Pro DC, Adobe Photoshop, Adobe Premiere Pro, Adobe InDesign, Adobe Captivate, Adobe Illustrator, Adobe Lightroom, Adobe After Effects, Adobe Audition, Adobe Media Encoder	Understanding/making use of large data sets, creation or generation of new content. Primarily for training and marketing purposes.	Available	Generative, Machine learning	Staff support
Dragon Naturally Speaking	Screen reader	In production	Machine learning	Staff support
DDoS protections – AWS Shield, Azure DDoS Protection, Cloudflare, Magictransit, F5 BigIP, Oracle DDoS	We use a range of tools to prevent our systems and services from unexpected outages due to network attacks.	In production	Machine learning	Internal use
Email-user link prediction	Predicts the similarity of email addresses and usernames so we can identify the probability that they are controlled / used by the same real-world person.	In development	Machine learning	Internal use and staff support
Figma	Prototyping software that enables us to develop mock-ups of intended changes to products and services across both e-services and our internal/external websites.	In production	Machine learning	Staff support
Financial intelligence network detection	Links, matches and identifies multi-dimensional risks of Crypto asset users via operational and strategic visualisation.	In production	Machine learning	Staff support
Genesys Agent Assist	Creates summaries of conversations with contact centre agents for post-call notes.	In production	Generative	Staff support
Genesys Agent Copilot	Group of functions to support contact centre agents, including surfacing relevant knowledge content, recommended next best actions, and predicted subject code.	Pilot	Natural language understanding	Staff support
GST integrity model	A predictive model to assess the risk of GST returns requesting refunds.	In production	Machine learning	Internal use
Graph Entity Resolution	Analyses and compares information we hold to external datasets provided by third parties to determine if records are referencing the same entity.	In production	Machine learning	Internal use
Microsoft 365 Copilot	Copilot is integrated into the M365 suite of products and is designed to enhance staff productivity.	Pilot	Generative	Staff support
Microsoft Copilot Chat (Bing/Browser)	AI-powered chat service, which generates answers based on web content.	Available	Generative	Staff support
Microsoft Defender	An enterprise-capable host protection solution that is integrated with a range of other Microsoft Apps, observes activity on devices for potential malicious behaviour.	In production	Machine learning	Internal use
Microsoft Teams Intelligent Recap	Provides recaps of Teams meetings.	Pilot	Generative	Staff support
Microsoft power BI	Dashboard/reporting software that connects to multiple Azure services.	Available	Machine learning	Staff support
Microsoft Purview	Portfolio of products that span data governance, data security, and risk and compliance solutions.	In development	Machine learning	Internal use
Nuix	Digital forensics and analysis of information.	In production	Machine learning, Natural language processing	Staff support
Overdue income tax return RIT prediction	Predicts residual income tax (RIT) on overdue returns.	In production	Machine learning	Staff support
Qualtrics	Analyses customer feedback from survey data.	In production	Machine learning, Natural language processing	Staff support
Power Automate	Low-code solution that supports automating tasks.	In production	Machine learning	Staff support
Propensity to read letter or log-in to myIR	Helps us to select and use the right channels to communicate with customers.	In production	Machine learning	Staff support
Receipt, invoice, statement and tax/employer return review	Text recognition.	In production	Optical character recognition	Internal use
Tableau Desktop	Data analytics and graphing tool used for analysing and visualising performance test results.	In production	Machine learning, Natural language processing	Internal use
Viva Engage	Provides people with personalised content feeds and recommendations of communities to join or follow.	In production	Machine learning	Internal use
Voice biometrics	Identity confirmation for contact centre calls.	In production	Biometrics	Supporting customers directly
Windows Hello for business	Authentication for Inland Revenue devices.	In production	Biometrics	Internal use
Z scaler	Detection and classification of web traffic and websites.	In production	Machine learning	Internal use
ZoomText	Screen magnification software for accessibility purposes.	In production	Optical character recognition	Staff support