Could you do your taxes with AI?

I thought about this when watching a video by Enrico Tartarotti. He discussed how current AI, or LLMs (Large Language Models), are like simple chatbots answering random queries — basically a validation product. To truly benefit the public, this technology must evolve into a more integrated feature that solves real-world problems.

This got me scratching my head, I wanted to illustrate how LLMs can be applied practically to solve real problems. So, I picked Nubank, a Brazilian digital bank and one of the largest fintechs in the world, taking over LATAM, and a problem that I’m sure is felt not only in Brazil but all over the world: the headache of tax season.

Disclaimer: This is a fictional feature and it plays fast and loose with some assumptions about internal tooling and the influence of the fintech. Let’s have fun!

Introducing the AI-Powered Tax Assistant

The Problem

Tax season is a nightmare. In Brazil, you have to declare your earnings to the Receita Federal (like the IRS). This means gathering earnings reports from different banks, employer statements, and investment broker trading histories. Then, you carefully enter everything into the Receita Federal’s online portal. It’s tedious, time-consuming, and often requires hiring an accountant to get it right.

The Solution: Nubank’s AI-Powered Tax Feature

Imagine getting a notification from the Nubank app during tax season. You click a button, follow a simple flow, and voilà — your taxes are done. Here’s how it works:

1. Notification: You get a ping from Nubank during tax season. Time to tackle those taxes!

2. Home Screen CTA: The app's home screen will have a prominent call to action that takes users to the feature flow.

3. Eligibility Check: The app guides you through a quick eligibility check. Initially, it’s for users with only Nubank accounts and no dependents.

4. Document Upload: You upload your employer statement. Easy peasy.

5. Data Compilation: The system gathers info from your Nubank savings, checking, and investment accounts. This is where the magic happens, with an LLM like Llama running on a private server and some request orchestration.

6. Submission: The data is formatted and submitted to the Receita Federal via an API, signed with Nubank’s “customer accounting” credentials.

All this for a fee of R$59 to R$99, depending on your tax bracket — cheaper than hiring an accountant! What if you could you do your taxes with AI?

Visualizar nos resultados da busca

Taxes with AI from Nubank (Fictional Feature)

Alignment

I do believe that this fits with the company's mission.

"To combat complexity and empowering people with responsible credit and transparency."

Why This Works

Nubank’s own investment platform is integrated into its app. Plus, with a median income of R$ 3.123, most Brazillians are in the the three first tax brackets: up to R$ 2.259,20 = 0%; R$ 2.259,21 to R$ 2.828,65 = 7,5%; R$ 2.828,66 to R$ 3.751,05 = 15%. Since I don’t have insight into the company’s internal user data, I’ll also make some assumptions (don't sue me): most of Nubank’s users are younger, likely to manage all their finances in one place, and not keen on spending hours doing taxes.

Potential Challenges

1. Regulatory Approval: This would require a partnership with the Receita Federal and possibly the Central Bank.

2. Costly Internal Team: The company would need to create an "customer accounting" team

3. Tax Fraud Risk: We must manage the risk of tax fraud carefully, weighing it against the potential revenue and user adoption benefits.

Success Metrics

1. User Adoption Rate: Percentage of eligible users who use the feature.

2. Customer Satisfaction: User feedback and satisfaction ratings post-usage.

3. Financial migration: Users would prefer to keep their money and investments inside Nubank, so, by how much has inidividual accounts migrated their money to the platform?

4. Revenue: Total revenue generated from the feature.

Counter Metrics

1. Error Rate: Incidence of errors in tax submission.

2. Support Tickets: Number of support tickets related to tax feature usage.

3. Churn Rate: Impact on user retention, particularly if errors occur.

4. Regulatory Complaints: Number of complaints or issues raised with regulatory bodies.
   

Future Iterations

If this feature is successful, we can expand it to include:

• Complex Declarations: Handling users with dependents.

• Multiple Accounts: Integrating with other banks and platforms via Open Banking. (this might not be necessary if we decide to make it a retention feature)

• Personalized Services: Offering enhanced services to Ultraviolet customers.

Final considerations

Hopefully that was an entertaining read, again, bear in mind that I wrote this for fun and with a lot of assumptions and "creative math".