Sources

    A Chief, Not a Calculator

    AI PRODUCT · CASE STORY

    A chief, not a calculator

    What a money app has to get right before it can help you make a real decision.

    4 min

    A bookkeeper and a chief of staff both know your numbers.

    Bookkeeper

    "Here's what you spent last month."

    Chief of staff

    "Here's what that means for the decision you're facing."

    Most money apps are bookkeepers.

    They count last month's coffee. They don't help with the big question.

    Meet Leela. She has tracked every dollar for years.

    Her real question isn't about coffee. It's this:

    "Can we move cities and still be fine in fifteen years?"

    Her budgeting app can't answer it. So every time, she pastes her whole life into an AI chat and starts over.

    Leela

    I've explained my life to a chatbot twenty times.

    Rule 1

    Start from the questions, not the categories.

    What the app is for

    What must we spend every month?

    What's optional?

    What could we cut?

    What feels like too much?

    Four questions. Anything that doesn't help answer them doesn't get built.

    PRINCIPLE

    Design for the job

    People don't want a product, they want progress on a job in their life.1 Here the job is "spend wisely," not "categorize forever."

    Rule 2

    Fix the data before you give advice.

    Paid the credit card

    $1,200

    →

    App counted it as

    Spending

    But it's really

    A transfer

    →

    The purchases were

    already counted

    On a wrong base, smart insights aren't just useless. They're confident, specific, and wrong.

    Rule 3

    Invent nothing.

    INVENTED

    "$200 a month on travel"

    One trip, spread across the year.

    REAL

    "One $2,400 trip in March"

    Not a monthly habit.

    And every number says whose it is: yours, the ledger's, or the app's own estimate.

    Yours

    The ledger's

    Mine

    Rule 4

    Don't claim a pattern the data hasn't earned.

    ✗

    "Phone bill up 40%!" She switched carriers. One ended, another began.

    ✗

    "Power bill keeps rising!" It's winter. Compare January to January.

    ✗

    "Restaurants: a subscription" Frequent isn't recurring. Only bills on a schedule are.

    PRINCIPLE

    Earn the claim

    A real pattern needs the data to support its shape, not just its numbers. When the app isn't sure, it asks instead of guessing.2

    Rule 5

    Plan with a floor and a ceiling.

    Ceiling: if things go well

    Floor: what you can count on

    Money that might not come, like rent from a house that might not find a tenant, only counts in the ceiling. Never in the floor.

    Rule 6

    Compare life paths, not just months.

    Three paths, ranked

    ~ projections

    1

    Big-city job, then move home

    2

    Stay put, then move home

    3

    Move now, let the career follow

    Instead of a dashboard to interpret, Leela gets a ranked answer to her real question. Every projected number carries a ~, because it's a projection, not a promise.

    Rule 7

    Learn what you already told it.

    Same merchant?

    AMZN MKTP US

    Amazon.com

    AMAZON PRIME

    Yes, merge them

    No

    If a person can notice it, the app should learn to notice it too. And it should never ask the same question twice.

    Rule 8

    Keep what came out of the conversation, not the conversation.

    Ask

    "What if we move in two years instead of now?"

    DRAFT PATH

    Move in 2 years

    Based on your current path, with the move date changed.

    Add as a path

    Discard

    PRINCIPLE

    Nothing changes until you say so

    The AI drafts and the person decides. Results are saved as things you can use, not as chat logs to reread.3

    Leela

    I stopped re-explaining my life. It already knows, and it tells me where it's guessing.

    BEHIND THE STORY

    Leela's problem was mine.

    I've tracked my own money for nine years, and no app ever fit. They look backward and can't run the futures. So I built Finance Chief, my personal CFO.

    Every rule in this story came from a correction I made while testing it on my real data. The goal I set for it: "Not live stingily, but not overspend either."

    The real app runs only on my own devices. The public demo uses a fictional family.

    CHEAT SHEET

    Building an AI that helps with money decisions

    Start from the questions, not the categories.

    Fix the data first. Wrong data makes confident, wrong advice.

    Invent nothing. Label where every number came from.

    Earn every pattern before you claim it.

    Floor and ceiling. Uncertain money only in the ceiling.

    Learn once, never re-ask. Keep outcomes, not chats.

    TRY IT

    Explore Finance Chief.

    The public demo runs on a fictional family. No real financial data.

    EXPLORE THE DEMO →

    You finished case story 02: "A chief, not a calculator." Finance Chief is a personal tool by Anuja Nimmagadda, built by directing Claude Code. Leela and the examples in this story are fictional.

    Sources

    1. Christensen, Hall, Dillon & Duncan (2016). Know Your Customers' "Jobs to Be Done". Harvard Business Review.
    2. Amershi et al. (2019). Guidelines for Human-AI Interaction. CHI 2019 (Microsoft HAX).
    3. Nielsen Norman Group. 10 Usability Heuristics for User Interface Design.