Brex application practice

Begin by separating business approval from transaction state. Approval expresses authority under a policy; it is not proof that a payment has settled or that the final amount matches the estimate. Define the states your exercise needs and who is allowed to change each one.

Model the entity and policy explicitly. Which organization owns the expense? Which policy applied when approval occurred? What happens if the employee changes teams or the policy is updated before the final event? Do not rely on the user's current profile alone to explain a historical decision.

Create a small event history. Include an initial request, approval, authorization-like event, final posted amount, correction, and a duplicated notification. Use plain labels and explain that the example is a simplified model, not a complete description of card-network behavior. The exercise should make the source and timing of each fact visible.

Define the exception that needs a human decision. A final amount outside the approved limit may require review, but do not invent a universal financial policy. Use a clearly hypothetical rule and show how an authorized owner would set or change it. Preserve the reason for the decision rather than silently overwriting the original approval.

Test duplicate and late-arriving events. Can the system distinguish a repeated notification from a new transaction? What does it show when a correction arrives after someone has reviewed the record? Avoid presenting the latest event as the whole truth when the sequence matters.

Design a reconciliation view that answers practical questions: what is known, what remains unmatched, what policy applied, and what action is permitted next? A large table of raw events may be technically complete but difficult to use. Give the operator a clear summary with access to the underlying evidence.

For engineering, implement the event and state logic with tests. For product or design, show the employee, approver, and finance views. For operations, write the escalation and correction procedure. For customer success, create discovery questions that reveal the customer's policies and ownership before implementation.

Measure correct identification of exceptions, time to understand a record, and the rate of erroneous actions in your test set. Do not claim a broad reduction in financial losses from a small simulation. State the limits of the model and the reviews required before real deployment.

Conclude with the smallest useful improvement and a plan for maintaining it. The best answer may be a clearer state model and exception queue rather than a more ambitious automation. Demonstrating that restraint is useful evidence of judgment.

Second-stage test: keep approval, authorization and settlement distinct

For the spending-and-reconciliation case, introduce a transaction whose amount or timing differs from the earlier approval. Identify the states that must be reconciled before calling the workflow complete.

Add a request for an exception. Explain who needs to decide it and what evidence they need; do not make assumptions about company policy or infer employment terms solely from corporate ownership.

Produce an exception table and an accountable decision record. The exercise should demonstrate precise financial-product reasoning with fictional records, not provide accounting advice or describe Brex's private operating controls.