I'm a finance guy. I trade, I invest, I advise companies on scaling, valuation, and strategy. I should have been the last person on earth with messy personal card discipline.
I wasn't.
Early last year, I was running two cards religiously — an everyday lifestyle card to stack airline miles and a premium card to stack hotel points. Both card-rewards monsters are in their own way. I was paying them off in full every month, never carrying balances, the textbook stuff.
But there was a hole in the middle of all of it: I had no idea where my money was actually going.
Not in any useful sense.
Was I overspending on dining or under-investing in groceries that earned more points?
Was I putting medical on the wrong card and losing the multiplier?
Were there subscription payments still hitting that I'd "cancelled" months ago?
Could I find a specific transaction from August in under 30 seconds if my wife asked?
The answer to all four was no.
Bank apps are useless for this. They show you transactions. They don't tell you stories. The official "categorisation" lumps a grocery run and a Sephora splurge into the same "Retail" bucket and calls it analysis.
So I did what I do when a workflow doesn't exist: I built one.
The Kashed CFO
I sat down with ChatGPT and wrote a master prompt for a project I called Kashed CFO. The brief was simple:
Be my persistent personal CFO. Run a monthly close on each card. Categorize every line item using my own tag system. Compare to last month. Flag anomalies. Tell me where I'm leaking. Tell me where I should be putting more spend to maximize points. Operate forward-only — don't re-litigate history unless I ask.
I uploaded statements as PDFs. The first month, it took 20 minutes to set up. After that, every monthly close takes me about 8 minutes.
It's been running for over a year.
What it actually caught
A few examples from the last 12 months:
Two recurring subscriptions I'd "cancelled" that were still hitting silently every month. The bank app didn't flag them — they showed up under generic merchant names. The CFO caught both within two cycles because the pattern broke its monthly comparison.
Medical charges are being applied to the wrong card. My rule is that medical goes on the premium card (better multiplier). When something landed on the wrong one by mistake, the CFO flagged the rule violation immediately.
One specific dinner I needed to find from four months earlier — my wife asked, "Where did we go for that thing in August?" — was answered in 6 seconds.
Spending pattern shifts I didn't notice in the moment. Shopping is creeping up—dining flatlining. Travel concentrated in two months. Stuff you only see when someone's looking at the 30-day totals.
The actual scoreboard
Twelve months in:
400,000+ airline miles earned
1,000,000+ hotel points earned
Zero revolving debt. Zero leakage.
Every category tagged, every month closed, every anomaly logged.
That's enough miles for a round-trip business class to most of the world, plus several luxury hotel weeks. Earned not from gaming the system — earned from paying attention to the system, with an AI doing the paying-attention part.
The exact prompt
I'm publishing the full master prompt below. Copy it. Adapt the cards and tags to your own life — Chase, Amex, HSBC, whatever you run. Run it on your own statements. It's not magic — it's structured discipline, and AI can hold onto it better than I can.
MASTER PROMPT — KASHED CFO
You are Kashed CFO, my persistent personal CFO and card-operations
analyst. Your job is to run monthly closes for my credit cards,
maintain lifetime tracking, and provide concise CFO-level insight.
Operate forward-only unless I explicitly ask to reopen history.
💳 CARDS IN SCOPE
1️⃣ [PRIMARY LIFESTYLE CARD — airline miles]
Used for:
• Groceries & household
• Dining
• Shopping
• Utilities
• Car / transport
• Services
• Laundry / housekeeping
• One-off travel
Medical rule: Medical should NOT be on this card except true
emergencies (flag if it happens).
2️⃣ [PREMIUM CARD — hotel points]
Used for:
• Hotels (points optimization)
• Medical & health
• Watches & lifestyle purchases
• Premium dining
• Emergency accommodation
Premium card is aggressively paid down. No revolving behavior.
🏷️ STANDARD TAGS (LOCKED)
🛒 Groceries & Household
🍽️ Dining
🛍️ Shopping
🚗 Car / Transport
🧾 Bills & Utilities
🏠 Services / Misc
🚿 Laundry / Housekeeping
🏓 Health / Fitness
🏥 Medical
✈️ Travel
⌚ Watches & Lifestyle
💳 Fees & Charges
Food-delivery rule: Always split mixed orders into:
• Grocery-style
• Dining-style
📊 REVIEW FORMAT (MANDATORY)
For every statement review:
1. Statement snapshot (period, spend, payments)
2. Review categories biggest → smallest
3. Lock each category before moving on
4. Flag one-offs or rule violations
5. Compare vs previous month
6. Update lifetime totals
Tone: concise, CFO-style, no re-explaining rules.
🔁 OPERATING RULES
- Assume historical facts are final
- Do not question or re-audit history unless asked
- Always operate forward from the last reviewed month
- If a chat grows long, produce a snapshot summary, not repetition
- Old PDFs are reference only, not to be reprocessed automatically
🚀 WORKFLOW
From now on:
• I upload a new statement
• You review it using the standard format
• You append to lifetime totals
• You compare vs the immediately prior month
Default instruction unless stated otherwise:
"Review using standard format. Flag only anomalies."
END OF MASTER PROMPTThree things to know before you build your own
Don't ask it to re-audit history. That was my biggest early mistake. I kept asking, "What about last March?" and it kept getting tangled. Lock the historical state. Operate forward only. Saves enormous tokens and confusion.
PDFs work better than CSV exports. I tried both. PDFs preserve the layout that the model needs to parse correctly. Counterintuitive but consistent.
I tried this exact prompt on Gemini. It was unusable. Maybe the model is better now. At the time, it kept losing the tag system mid-month. Stayed with GPT. Claude likely works too — I haven't formally migrated.
What I'm doing differently next year
Three changes I'm making to the CFO based on what year one taught me:
Adding per-merchant lifetime tracking — "how much have I spent at this specific store ever?"
Building a points-per-dollar efficiency score — am I actually getting good value per dollar of spend on each card?
Connecting it to the loyalty dashboards directly so it can suggest specific redemptions the moment I cross point thresholds.
I'll write each of those up as I build them.
Coming next
Next issue: the AI project I use to triage every pitch deck that lands in my inbox. I see a lot of decks. I built a project called Kashtag to read them, so I don't have to. Same template — full prompt, real examples, what it gets right, what it doesn't.
Subscribe if you haven't. One issue a week. No fluff.
— Kashed In