CountaBill
AI Transparency Notice
Last updated: 24 August 2026
CountaBill uses artificial intelligence for one job: reading a photo of a receipt so you do not have to type it in. This notice explains exactly what the AI is used for, what data it sees, and how you stay in control. It supports EU AI Act Article 50 and the OECD AI Principles.
Where AI is used
| Feature | Provider | Purpose | What the model sees |
|---|---|---|---|
| Receipt scanning (line-item extraction) | Google Gemini, paid tier, via Firebase Cloud Functions | Extract the merchant, date, and line items from a photo of a receipt | The receipt image and a fixed instruction prompt |
| Detecting personal details on receipts, so they can be blurred | Google Gemini, paid tier | Find names, addresses, and card numbers so we can blur them before storing a training copy — and only if you opted in to that | The receipt image and a fixed instruction prompt |
No other AI providers, chat assistants, or generative models are used. The specific Gemini model is set on our servers and may be upgraded within the Gemini family without changing what data is sent or what it is used for. A change of provider or purpose would require an update to this notice and 30 days' notice to you.
What the AI does not do
- It does not make automated decisions with legal or similarly significant effects on you. No credit scoring, no fraud declines, no eligibility judgments. GDPR Article 22 and POPIA section 71 do not apply.
- It does not learn from your data. The paid Gemini tier contractually prohibits Google from training its models on our inputs.
- It does not profile you or generate synthetic images of you.
- It does not infer sensitive information about you such as race, health, or religion.
- It is not a chatbot. There is no conversational AI in the app.
You are always the one who decides
Everything the model extracts is shown to you for confirmation before it is saved. You can edit or delete any of it. Nothing the AI reads is recorded as fact until you press Confirm.
- If scanning fails or the result looks unreliable, the app says so and lets you enter the receipt by hand.
- We do not rely on model output for anything that moves money or determines eligibility.
Training data, and your part in it
- By default, your receipts are not used to train any model.
- If you opt in (Profile → Data & Privacy → "Help improve receipt scanning"), a copy of the receipt is stored with detected personal details Gaussian-blurred, EXIF location data stripped, and your user ID replaced with a salted one-way hash.
- Those copies are kept for 180 days and then deleted.
- Deleting your account deletes them too.
Accuracy and limitations
- Accuracy depends on the photo — focus, lighting, and angle all matter.
- Unusual layouts, handwritten receipts, and faded thermal paper reduce accuracy.
- Currency and date formats can be misread.
Always check the extracted values before saving. We are not liable for errors you accept without review — which is why the confirmation step exists.
Your rights
- Avoid AI entirely by entering receipts manually.
- Turn off training-data collection at any time in Profile → Data & Privacy.
- Export your data, including any training copies linked to your account (Profile → Export my data).
- Delete everything (Profile → Delete account).
Our position under the EU AI Act
Based on our current assessment:
- The system is a general-purpose AI service consumed as a downstream deployer. Google is the provider; we are the deployer.
- It is not high-risk under Annex III — no biometrics, no employment decisions, no credit, no access to essential services.
- The Article 50 transparency obligations that could apply to us are met by this notice and by the in-app confirmation step.
Contact
For AI-related questions or complaints: support-budget.app@logicalsolutionsinnovations.com.
You may also complain to your local supervisory authority — see the Privacy Policy for who that is where you live.