Why your own numbers beat a calculator
Rewards calculators ask you to estimate monthly spending on dining, groceries, gas, and travel. Most people guess wrong, usually low on dining and delivery and high on travel. A card that wins on a guess can lose on your real numbers.
An agent connected to your accounts can skip the guessing. It totals what you spent, and you give it the terms to score.
Get twelve months of spending by category
Connect every card and checking account you spend from (see connecting multiple banks), then ask:
“Total my spending by category for the last twelve months across all accounts. Exclude credit card payments, transfers between my accounts, rent, and loan payments, since those can't go on a rewards card. Then list the top 15 merchants by total.”
The agent calls get_spending_summary grouped by category and again grouped by merchant. The merchant list is what you need for the next steps, because it shows where the category labels may be misleading.
The exclusions matter. A card payment from checking duplicates purchases already on the card; see stop double-counting credit card payments.
Give the agent the card terms
Open each card's page on the issuer's site and copy the reward rates, category caps, annual fee, and any rotating or choose-your-category rules. Paste them in as plain text. Include your current card so the comparison has a baseline.
Do not ask the agent to recall card offers from memory. Terms change and a model can be confidently out of date.
Score each card
“Using my twelve-month category totals, compute the annual rewards for each card above. Apply caps, subtract annual fees, and ignore sign-up bonuses. Show a table, then the same table if I keep my current card for the categories where it wins.”
The output looks like this. The cards and numbers below are a made-up illustration of the format, not offers. It assumes $20,000 of yearly card spending, $8,000 of it on dining and groceries:
| Card (illustrative) | Rewards | Annual fee | Net per year |
|---|---|---|---|
| Current flat 1.5% card | $300 | $0 | $300 |
| Card A: 4% dining and groceries, 1% else | $440 | $95 | $345 |
| Card A for dining and groceries, current card for the rest | $500 | $95 | $405 |
Ask for a sensitivity check too: “How much would my dining spending need to drop for Card A to stop beating my current card?” If the answer is a small change, the annual fee is a bet on a habit.
Category traps to check
The categories your agent sees come from the bank data provider. Rewards are paid on the merchant category code the card network assigns. They usually agree, but these often do not:
- Superstores and wholesale clubs.Groceries bought at a big-box store or warehouse club often do not earn a grocery bonus. Ask the agent to list your top grocery-labeled merchants and check each issuer's exclusions.
- Delivery apps. Food delivery may be labeled dining by your data and coded differently by the network.
- Gas at warehouse clubs and grocery stores. Frequently excluded from gas bonuses.
- Bills paid through a payment app. A utility paid through a payment service may not count as utilities.
A follow-up prompt that helps: “Recompute assuming Costco and Walmart earn the base rate.”
After you switch
Connect the new card to BankBridge so your agent sees it (each bank is $5 per month; a card from a bank you already connected adds no cost if it appears under the same login). Two checks are worth running in the first months:
- Move recurring charges to the new card, then ask your agent to list recurring charges by account so you can see which ones still bill the old card. See finding recurring charges with AI.
- Watch for unfamiliar charges on a new account. The Claude Code plugin's
/fraud-checkcommand and the fraud detective workflow cover this.
Connect your accounts and run the twelve-month total first. It is the number every other step depends on.