AI system
- System
-
ATO myTax real-time analytics prompts using nearest-neighbor comparison models
- Model
-
Not assessed
- Release
-
Not assessed
- Application developer
-
Australian Taxation Office (in-house data and analytics) - Role
-
Not assessed
- Configuration
-
Not assessed
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
Large volumes of pre-filled third-party data (over 85 million pieces in 2020) giving a comparison base. Nearest-neighbor comparison of entered amounts against taxpayers in similar circumstances. A real-time prompt inside myTax before lodgment. Taxpayers' own decision to adjust a flagged label.
- Human oversight
-
Not assessed
- Relation to AI
-
The ATO page states that nearest-neighbor methods generate the prompts and that the ATO uses machine learning. It attributes the adjustments to the prompts, though taxpayers themselves make the changes. Only the deployer's account is among the sources for the outcome. The ATO built and runs the system in-house, so its page is both the developer's and the deployer's account. The ANAO audit reported by The Mandarin found AI governance “partly effective” and offers no outcome figures in the cited excerpt.
- AI contribution
-
The ATO page states that nearest-neighbor methods generate the prompts and that the ATO uses machine learning. It attributes the adjustments to the prompts, though taxpayers themselves make the changes. Only the deployer's account is among the sources for the outcome. The ATO built and runs the system in-house, so its page is both the developer's and the deployer's account. The ANAO audit reported by The Mandarin found AI governance “partly effective” and offers no outcome figures in the cited excerpt.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
adjustments estimated to have a revenue impact of around $37 million
- Category
-
Reduced Risk
- Status
-
Not assessed
- Beneficiary groups
-
People. Organizations. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
adjustments estimated to have a revenue impact of around $37 million
- Comparison
-
Not assessed
- Results
-
adjustments estimated to have a revenue impact of around $37 million
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
revenue impact of the adjustments taxpayers made after myTax prompts at tax time 2020 (ATO's own estimate)
- Amount
-
Estimate: $37 million
- Currency and prices
-
AUD, identified from the source context. Nominal source amount; price year not separately stated.
- Period and unit
-
Australian tax time 2020. Aggregate adjustments following myTax prompts.
- Realization
-
Retrospective estimate.
- Recipients and payers
-
Recipient: Australian public revenue. Payer: Taxpayers adjusting their returns.
- Gross or net
-
Estimated tax-revenue impact, not confirmed cash collection, net program return, or aggregate social benefit.
- Included costs
-
Not assessed
- Comparison
-
Amounts entered before versus after prompts; an untreated causal comparison is not supplied.
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer:
Australian Taxation Office (in-house data and analytics). Operator: Australian Taxation Office. - Jurisdictions
-
Benefit:
Australia (national, online myTax lodgment). Developer: Australia. Operator: Australia. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Comparison groups and thresholds that keep prompts credible to taxpayers. Continued use of myTax self-preparation at scale. AI governance that the ANAO found only partly effective in 2025 being maintained or improved. The ATO's method of estimating revenue impact remaining the only measure, since no independent evaluation is among the sources.
- Replication attempts
-
Not assessed
- Later findings
-
The Mandarin reported on 25 February 2025 that the ANAO had judged the ATO's AI governance only partly effective. That report gives no later outcome figure for the myTax prompts.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0337-source-0001–0006
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Developer, Independent
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
People. Organizations. Society.
- Depth
-
People: Not assessed. Organizations: 1. Society: 1.
- Reasons
-
People: Taxpayers who corrected claims are not individually documented; the page gives no individual outcome beyond the aggregate prompt count. Organizations: The ATO's own estimate of about $37 million revenue impact from one tax time, with no independent verification and no comparison with total collections given. Society: Protects public revenue at modest scale: prompts reached 7.5% of myTax users and the stated impact is an estimate.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently reported
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Public administration: tax compliance. National revenue agency.
- Purpose
-
Built to prompt myTax users in real time to check claims that differ from those of similar taxpayers before lodgment. The recorded adjustments are that purpose.
- Dates
-
Date basis unassessed: 2021 (precision unassessed). Updated: 21 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: d2f5abf5d321…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
8 revisions. Latest: 21 October 2026.
Sources
-
Report Developer
Australian Taxation Office (n.d.). How we use data and analytics. ato.gov.au (undated page).
During tax time in 2020, nearly 340,000 taxpayers (around 7.5% of myTax users), received a pop-up message through myTax suggesting they review a specific label.
These prompts caused taxpayers to make adjustments estimated to have a revenue impact of around $37 million.
Nearest neighbour methods are used to compare amounts being entered into myTax, with those of other people in similar circumstances, along with other checks.
We use a common form of artificial intelligence known as machine learning.
-
Report Developer
These prompts caused taxpayers to make adjustments estimated to have a revenue impact of around $37 million.
-
News report Independent author
Bajkowski, J. (2025). Auditor whacks ATO over artificial intelligence. The Mandarin, 25 Feb 2025 (ANAO finding).
partly effective