How I found $300/month I was wasting — using AI and my bank data
I thought I had a pretty good handle on my money.
I knew my rent. I knew my salary. I checked my bank balance most weeks. I had a vague sense that I "spent too much on food sometimes" — but nothing alarming. Nothing that would suggest I was quietly burning through $300+ every month on things I either didn't need or had completely forgotten about.
Then I connected my bank to Claude.
What followed was one of the more uncomfortable — and genuinely useful — hours I've spent thinking about money. This is the story of what I found, how I found it, and what I did about it.
The setup: five minutes, one connection# permalink to this section
I'd been using OpenBudget to sync my transactions to Claude via MCP. The setup took about five minutes — connect your bank via Plaid, install the MCP server for Claude, done. From that point on, Claude had live access to every transaction across my Chase checking, Amex Gold, Apple Card, and Chase Sapphire.
I didn't go in looking for $300. I went in curious. My first question was simple.
Step 1: I asked for the big picture# permalink to this section
Give me a full financial summary for April.
The answer was humbling.
April was my most expensive month of the year — $3,847 across 84 transactions. That was $312 more than March, which was already $245 more than February. My spending had been creeping up for four months straight and I genuinely hadn't noticed.
The month scorecard was the part that got me: B for overall spending. D for dining control. A for savings rate. C for hitting my budget goal.
A D for dining control. Four months of data confirmed it wasn't a bad month — it was a pattern.
Step 2: I dug into dining — and it was worse than I thought# permalink to this section
Show me a month-by-month breakdown of dining expenses for 2025.
I spent $295 on dining in January. By April it had reached $632 — more than double, in four months. Claude's framing hit harder than any number: "The jump from Feb→Mar (+$110) and Mar→Apr (+$204) suggests a lifestyle shift, not a one-off splurge."
It wasn't one expensive dinner. It was a new normal creeping in. Nobu was appearing in my top merchants every month from February onward — 3 visits averaging $137 each. At April's pace I'd spend approximately $7,500 on dining in 2025, up from an estimated $3,500 if I'd stayed at January's level.
Waste found so far: ~$200/month above my January baseline.
Step 3: The DoorDash audit I didn't want to do# permalink to this section
I knew I ordered delivery. I didn't know how much.
How many times did I order DoorDash last month?
Nine times. $243 total. Up from four orders in March.
Average order: $27 including fees and tip. Six of the nine were on Friday or Saturday. At April's pace I'd spend $2,900 on DoorDash alone this year. The kicker from Claude: "Capping at 4 orders/month saves ~$130."
That wasn't cutting delivery. That was just going back to March's level — something I'd apparently done without noticing it was reasonable.
Running total: ~$330/month in identifiable waste.
Step 4: The subscription audit# permalink to this section
I thought I paid around $80/month in subscriptions. I was off by nearly 3×.
List all my subscriptions and their monthly cost.
Claude found 11 active subscriptions totaling $231/month ($2,772/year). Three of them were flagged as unused:
- Adobe Creative Cloud — $54.99/month. No logins in 73 days.
- Audible — $14.95/month. No books redeemed in 90 days.
- YouTube Premium — $13.99/month. Flagged as overlapping with Spotify, which I also pay for.
That's $83.93/month in services I wasn't using. Adobe alone costs more than Netflix and Spotify combined, and I hadn't opened it since February.
Running total: ~$414/month in identifiable waste (including the dining overage).
Step 5: The charges I'd completely forgotten# permalink to this section
This was the one that surprised me most.
Find all transactions over $200 in the last 30 days.
Seven transactions over $200 in 30 days, totaling $3,126. Two were flagged as unusual — no similar charges in any prior month. The Delta Airlines flight ($748) I remembered. The West Elm furniture charge ($387) I had completely forgotten about. It had been on my Amex Gold and I'd scrolled past it three times without registering it.
Claude's note: "The Delta and West Elm charges have no prior history — together they add $1,135 that's outside your normal spending pattern. Everything else is expected."
That West Elm charge wasn't waste — it was a one-time purchase I'd chosen to make. But finding it sitting there unexamined was a reminder: I had no real visibility into where large amounts were going. I was flying blind.
Step 6: The hidden fees I didn't know I was paying# permalink to this section
Show all late fees and interest charges this year.
This one hurt.
$743 in fees and interest charges from January through May. Three late fees totaling $105 — all missed by just 2–4 days, all completely avoidable with autopay. Plus $638 in interest charges across Amex Gold (APR 29.99%) and Chase Sapphire (APR 24.99%).
Claude's projection: "You're on pace to pay ~$1,780 in fees and interest in 2025. Two fixes eliminate most of it: turn on autopay for minimums (stops late fees immediately), and pay more than the minimum on Amex each month to reduce the interest base."
The late fees were pure waste. $105 that I paid for the privilege of being 3 days late on a payment I had the money for. That's $21/month, annualized to $252/year — for nothing.
Running total: ~$435/month including late fees.
Step 7: The reality check# permalink to this section
By this point I had a clear picture. I asked Claude to put it all together.
Am I on track to stay under $3,000 this month?
Not on track. Projected to overshoot by $364. At my current daily pace of $124/day, I'd need to spend just $38.60/day for the remaining 10 days to hit $3,000 — a 69% drop in daily spending that wasn't realistic.
But the specific insight was useful: "Dining and travel together account for $373 of the projected $364 overage. Curb those two categories and you're back within budget."
Not a complete overhaul. Two categories. That was actually doable.
What I actually changed — and what it saved# permalink to this section
Here's the honest summary. I didn't become a monk. I made four specific changes:
Plus the dining reduction — I didn't put a hard cap on it, but being aware that I'd gone from $295 to $632 in four months was enough to change behavior. I dropped back to roughly $420/month — about $200/month less than April's peak, and a level I was genuinely comfortable with.
Combined: approximately $420/month saved. More than I expected when I started.
The thing nobody tells you about personal finance# permalink to this section
The problem was never that I didn't care about money. It was that I had no visibility.
I wasn't making bad decisions consciously. I was making them invisibly — one $27 delivery order at a time, one subscription auto-renewal at a time, one 3-day-late payment at a time. None of it felt significant in the moment. None of it was visible in aggregate.
The moment I could actually see the pattern — dining nearly doubled in 4 months, DoorDash orders more than doubled in one month, $83/month in subscriptions I'd stopped using — the decisions made themselves.
That's what AI with your real data does. It doesn't tell you what to do. It makes the invisible visible.
How to do this yourself# permalink to this section
The whole process — connection to first insight — took about 5 minutes of setup and an hour of uncomfortable questions.
Step 1. Go to openbudget.sh and create an account.
Step 2. Connect your bank accounts via Plaid. Takes about 2 minutes. Supports 10,000+ institutions.
Step 3. Install the OpenBudget MCP server for Claude or ChatGPT. Copy-paste commands — no technical knowledge needed.
Step 4. Start with these five questions, in this order:
- "Give me a full financial summary for last month"
- "Show me a month-by-month breakdown of my top spending category"
- "List all my subscriptions and their monthly cost"
- "Find recurring charges I might have forgotten about"
- "Show all late fees and interest charges this year"
If you're like most people, you'll find something in step 3 or 4 that surprises you. The subscriptions question alone saves people an average of $80–120/month in services they've stopped using.
The questions that found the most money# permalink to this section
Not every question was equally useful. Here's what worked best:
The pattern: the questions that revealed trends were more valuable than the questions that revealed totals. It wasn't that I spent $632 on dining in April. It was that I'd spent $295 in January and the number had quietly doubled.
One month later# permalink to this section
I haven't become someone who tracks every dollar. I still eat out. I still order delivery sometimes.
But the three subscriptions are canceled. Autopay is on. DoorDash is capped. And I check in with Claude maybe once a week — usually just one question: "Am I on track this month?"
The answer is usually yes now. Not because I spend less on things I actually enjoy — but because I stopped spending $300+ on things I'd either forgotten about or never really wanted in the first place.
That's what visibility does.
OpenBudget connects your bank to Claude and ChatGPT — so you can find the patterns in your spending that are invisible in any single bank statement. Get started at openbudget.sh →
OpenBudget
