Ethan Bloch on Using ChatGPT Finances to Review Spending and Model Decisions

Ethan Bloch, product lead for ChatGPT Finances, joins Peter Yang to demonstrate ways to examine spending and model financial decisions. The walkthrough uses a fictional character, Jim, and some synthetic data. Wispr Flow sponsors the episode; Riverside and Linear are also listed as sponsors.

Bloch describes a long interest in personal finance, including losing his early trading gains during the internet downturn. He founded Digit to help people save and says the limitations of earlier technology shaped his interest in using current models with financial data.

His first prompt asks the assistant to review 90 days of transactions for subscriptions, price increases, duplicate charges, fees and unusually high bills. The fictional demonstration flags several candidates. A repeated salon charge turns out to have an explanation, illustrating why a flagged transaction still needs verification.

Bloch describes drafting a message to resolve a suspected duplicate storage charge and recounts a separate personal hotel reimbursement. Yang shares a similar customer-service example. These are anecdotes, not guaranteed savings; communications and financial actions depend on the user’s permissions and review.

Bloch demonstrates the account connection and dashboard. He says the bank connection uses Plaid and describes the connection in the video as read-only. The dashboard groups spending, shows balances and investment information, and lets users correct categories; the speakers acknowledge that automated categorization can be wrong.

On privacy, Bloch says raw balances and transactions are not used for training, while information brought into a chat may be treated differently depending on settings. This is his explanation at the time of the interview; users should check current product disclosures and settings before connecting sensitive accounts.

Bloch describes an automatically created weekly update and optional notifications for spending thresholds. In his fictional DoorDash example, the assistant tracks monthly spending against a chosen amount. He says users can turn off the weekly update and create other scheduled checks, subject to plan limits.

For taxes, he asks for possible savings based on the available context, then requests an estimate, a visual summary and a to-do list. He describes using the assistant to gather documents and prepare responses for his accountant. Suggested tax savings are estimates, and eligibility, current law and individual circumstances need checking with a qualified professional.

For investment accounts, Bloch suggests asking for a comparison of fund expenses and possible alternatives. A lower expense ratio alone does not establish equivalence or make a switch appropriate: holdings, taxes, trading costs and account rules also matter. His example is a starting point for research.

For credit cards, he asks the assistant to compare spending categories and card benefits. A visualization shows a hypothetical reward cap and possible changes. Rewards, caps, annual fees and eligibility depend on the actual card terms, and the illustrative estimate is not a promised gain.

For time off, Bloch uses a fictional family scenario and asks for an interactive model with several paths. Sliders change the length of the break and spending assumptions so the couple can discuss their cash reserve. The display is only as reliable as its inputs and calculations.

The speakers also discuss retirement and a rent-versus-buy model. Bloch says a person can supply figures manually rather than connecting every account. The purchase model compares outcomes under adjustable assumptions; its numbers are illustrative, not a recommendation to buy or rent.

Bloch describes recurring checks for subscriptions, family spending and idle cash. He says bank connections are read-only and browser-based actions at financial institutions may be blocked or vary by provider. He treats these routines as prompts for review, not a guarantee of savings or execution.

On the product team, Bloch says the group ships frequent small changes while applying testing and care to sensitive data. He describes a workflow in which engineers and designers use agents to build prototypes, then review those prototypes together rather than relying on lengthy specifications for every feature.

Asked about the future, Bloch says he has no specific announcement and wants to make the existing product easier to discover and use. The discussion closes with the potential for financial tools and voice interactions, framed as a product direction rather than a promised feature.