top of page
Elfworks - Pitch Deck v5.jpg

Elfworks Advice – ATO Private Ruling Validation

Updated: Sep 19

Part of the challenge when developing an AI advice product for the Australian accounting industry is building user trust. Using AI models that do not store data and do not use data for training helps to build trust. Checking the research materials of the AI models for hallucinations by validating these research lists against our private databases of legislation, ATO releases and Case Law also helps to build trust.


Another key step in the trust building process involves checking the final output against ATO material on the same subject. We all know the ATO is not the font of all knowledge on every subject, yet the ATO Private Ruing database serves as a valuable sanity check, serving to anchor the end AI advice back to the ATO view of the world.


ATO Private Ruling Database

We first created a new database just to house ATO Private Binding Rulings and then downloaded every Private Ruling released since 2008 and stored these in the database. Every month we then add newly released Private Rulings. At last count, there are just over 29,000 Private Rulings stored in this database.

We then built a search feature for this Private Ruling Database that not only returns relevant rulings on the particular topic, but the output of the search function produces a Relevance Score and the year the Private Ruling was published. This allows the user to select (via a live link) the Private Ruling they would like to review further.


Here's how we built-in the Private Ruling validation check.

 

Private Ruling Validation at the end of the Advice process

Once you run a client scenario through Elfworks advice, the platform:

  1. Checks the AI model research material list against private databases of legislation, ATO releases and Case Law (this is our Hallucination Catching Step); and

  2. Has each AI model answer the question, read each other’s output and arrive at consensus and alternative positions (this is our Multi-model Validation Step). The platform then goes back to the original scenario, facts and question(s) and finds the 5 most relevant Private Rulings from our 29,000 strong database and then compares these five most relevant private rulings to the AI model consensus position.

  3. At the end, you have a Summary statement that either confirms or challenges the Consensus position, such as - “In conclusion, the tax advisor's advice is well-founded and consistent with the ATO's interpretation as demonstrated in the private rulings. Remember that private rulings are legally binding only on the taxpayer who requested them and apply only to the specific facts outlined in the ruling. While they provide insight into the ATO's interpretation, they cannot be directly relied upon as precedent for other taxpayers.”


All of this analysis is then presented in a neat summary, ready for Exporting into a format of your choosing (Email, Work Document, PDF etc).


ree


 

A reminder on our new pricing structure – Light, Professional and Premium usage (from mid-Sept 2025)


We are moving to a three-tiered pricing structure, as set out below:

Tier

Cost (ex GST) per user /month

Token limit*

Number of Advice pieces/month**

Light

$19

3 million

~3

Professional

$49

15 million

~15-20

Premium

$89

30 million

~40

 

*Token limits are soft limits i.e., if you exceed your limit for the month, you can still use the platform however

** The above table explains token usage based on Advice pieces run as this if the Elfworks feature that burns the most tokens. Other platform features have a much lower token burn e.g., Client Query, Chat, Private Rulings and the Toolbox features can be used consistently throughout the month on the Low Usage tier without exhausting the 3,000,000 token limit.


The idea is to allow firms to find the right mix for them; firms that have a centralised advisory function may purchase several High-usage licences with other staff on the lower tiers. Firms with a more collaborate, de-centralised approach may have everyone on the Medium-usage tier.  

 

Data and Privacy

We have opted-out of storage and data-training with our enterprise subscriptions to Ai models, meaning client information is not stored or used for training the model post analysis; the data is deleted with only the analysis remaining.

 

Contact us for free trial

The next step is to experience the benefits of an AI enhanced application created to help Australian accountants in public practice with research, advice drafting, client email responses and a wide array of productivity tools.  If you would like a free trial of the Elfworks platform, please contact us on info@elfworks.ai.

 
 
 

Comments


bottom of page