Can AI help you save money? We tested it for 30 days

The question sounded almost too simple: if you connect your bank account to Claude and actually use it every day for a month, does it change how much you spend?

We decided to find out. One person. One bank connection. Thirty days. No manual tracking, no spreadsheets, no financial advisor — just Claude with live access to real transaction data through OpenBudget.

Here's everything that happened.

The setup#

The participant: A 29-year-old living in New York City. Income: $4,800/month. Spending: somewhere around $3,500–4,000/month, depending on who you ask (turns out the answer was closer to $4,200, but we'll get to that).

The tools: OpenBudget to connect bank accounts via Plaid. Claude as the AI. The MCP integration to give Claude live access to every transaction.

The accounts connected: Chase checking, Amex Gold, Chase Sapphire Reserve, Apple Card.

The rule: Ask Claude at least one financial question per day. Act on what it finds. Track what actually changes.

The starting assumption: "I have a decent handle on my money."

What actually happened: That assumption was wrong by about $1,000/month.

Day 1: The uncomfortable reality check#

The first question was supposed to be easy.

Give me a full financial summary for April.

April was the most expensive month of the year — $4,218 across 91 transactions. That's $683 more than March. Savings rate: 12%. The scorecard was brutal: C+ overall, D for dining control, D+ vs prior months.

The number that hurt most: "At this pace you'll save $1,800 less in 2025 than you started the year on track for."

Day 1 lesson: the vague sense that "spending has crept up lately" was hiding something specific. Dining hit $748 — the highest ever. Three dinners over $100 drove most of that jump.

Day 2: The income vs expenses gap nobody talks about#

Seeing April's number raised an obvious follow-up.

What's my average monthly income vs expenses?

Income: $4,800/month, stable. Average expenses: $3,967/month, trending up. Average surplus: $833/month — which sounds fine, until you see the trend. In January the surplus was $1,584. By April it had collapsed to $582. That's a $1,002 drop in monthly savings in four months, with income unchanged.

Claude's note: "If the trend continues, you'll be saving under $200/month by July."

That sentence ended any remaining sense that things were basically fine.

Day 4: The spending trend that made everything make sense#

Is my spending increasing or decreasing over the last 6 months?

Spending had grown $1,002/month since December — a 31% rise in 6 months. The pattern was clear: gradual increases December through March (+$319 total), then an explosion in April alone (+$683). Dining, delivery, and shopping together grew $691/month since December.

The projection that nobody wants to read: "At this rate you'll be spending $5,000+/month by August."

This was the moment the experiment stopped feeling academic.

Week 1 results: three finds that together add up fast#

By day 7, three specific problems had been identified — none of which were visible from a normal bank statement review.

Find #1: Subscriptions ($96/month hiding in plain sight)#

List all my subscriptions and their monthly cost.

Fourteen active subscriptions. $257/month. $3,084/year. Four of them flagged as unused or overlapping — Adobe Creative Cloud ($54.99, no logins in 84 days), Audible ($14.95, unused 90 days), Peloton App ($12.99, opened twice in 60 days — redundant with Equinox), and Dropbox Plus ($9.99 — duplicate of iCloud 2TB already being paid).

Canceling all four: $96/month saved. $1,152/year.

The kicker: "Adobe alone costs more than Netflix and Spotify combined — and hasn't been opened since February."

Find #2: Forgotten charges ($113/month, $339 paid for nothing)#

Find recurring charges I might have forgotten about.

Five recurring charges flagged as forgotten — $113/month, $339 already paid in the past 3 months with almost nothing to show for it. Some overlap with the subscription audit, but the framing here is different: these weren't just unused, they were forgotten. YouTube Premium ($13.99) overlaps with Spotify. Dropbox Plus duplicates iCloud storage.

Total if cancelled immediately: $113/month, $1,356/year.

Find #3: Late fees and interest ($681 in avoidable charges)#

Show all late fees and interest charges this year.

Sixteen charges in 2025 totaling $681 — $135 in late fees (4 charges, every one missed by just 2–3 days), plus $546 in interest on Amex Gold (APR 29.99%) and Chase Sapphire (APR 24.99%). On pace for $1,634 in fees and interest for the full year.

Claude's verdict: "All 4 late fees were completely avoidable. Autopay on minimums takes 2 minutes and eliminates them permanently."

Autopay was turned on that evening. $135/year in late fees: gone.

Week 1 summary#

Week 2: The behavioral patterns nobody wants to see#

Week 1 was about finding money being silently drained. Week 2 was about understanding behavior — the patterns that don't show up in any single month but are obvious when you look across months.

The dining trend#

Show me a month-by-month breakdown of dining expenses for 2025.

Dining grew every single month for four months straight: $458 → $497 → $556 → $748. That's +63% since January. Nobu appears in the top merchants every month from February onward, averaging $180/visit. Claude's framing: "This isn't a one-off spike, it's a pattern. At April's pace you'll spend ~$9,000 on dining in 2025."

Compare that to January's pace: $458 × 12 = $5,496/year. The lifestyle shift was costing approximately $3,500/year extra — and it had happened gradually enough to feel normal.

The DoorDash habit#

How many times did I order DoorDash last month?

Eleven orders. $312. Up from 4 orders in January. Nearly 3× per week, 8 of 11 on Friday or Saturday. At April's pace: $3,700 on DoorDash alone in 2025.

The specific insight that changed behavior: "Cutting to 4 orders/month saves $196/month — $2,352/year." That's not cutting delivery. That's going back to January's behavior — something that had been sustainable for months before the habit escalated.

The weekend premium#

Do I spend more on weekends?

Weekend days cost $168 vs $97 on weekdays — 73% higher. Saturday alone averages $196/day. Weekends are just 2 of 7 days but account for 37% of total monthly spend. The gap by category: dining +$196, shopping +$143, entertainment +$88 on weekends vs weekdays.

The saving: "Bringing weekend spending to 1.3× weekday rate instead of 1.73× saves ~$290/month."

This didn't require giving up weekends. It required being aware that Friday night DoorDash + Saturday dinner + Sunday brunch was a $500 weekend, not a normal one.

Week 2 lesson: patterns matter more than totals#

The most useful thing about Week 2 wasn't any single number — it was seeing that every problem had a direction. Dining wasn't just high in April; it had been climbing for four months. DoorDash orders hadn't always been 11/month; they'd been 4 in January. Weekend spending hadn't always been 73% higher; the gap had been widening.

When you can see the trajectory, you can interrupt it before it compounds further.

Week 3: Building the plan#

With the problems clearly mapped, Week 3 was about making concrete decisions.

The debt situation#

How long will it take to pay off my Amex balance at my current rate?

The minimum payment answer nobody wants to hear: at $83/month, the $2,890 Amex balance would take 3 years and 10 months to clear, with $891 paid in interest on top of the principal. Paying $400/month instead clears it in 8 months and saves $728 in interest.

Claude's specific connection: "Your current surplus of $582/month covers $400 to Amex — but only if you cut dining back. Cancel Adobe + halve DoorDash = $139/month freed, more than enough to fund the aggressive payoff."

The savings from Week 1 were enough to fund the debt payoff. The two things were directly connected.

The specific cut plan#

What could I cut to save an extra $300/month?

Five cuts. $334/month total. $4,008/year. The important detail: cuts 2 and 4 require zero lifestyle change — just canceling two apps not being used. That's $68/month for 2 minutes of work. The big lever is dining: dropping from 12 restaurant dinners to 5/month (the January baseline) saves $140/month. Not eliminating restaurants — just going back to what had worked before.

The savings goal#

I want to save $2,000 in 3 months. Based on my spending, is that realistic?

"Possible but tight." Current surplus: $582. Goal requires: $667. Gap: $85/month. But with the identified cuts already implemented, the surplus jumps to $874/month — hitting $2,000 in 3 months with buffer.

The diagnostic that explained everything: "In January your surplus was $1,584 — more than enough. The issue is dining grew from $458 to $748 in 4 months, consuming almost all of your savings cushion. Fix dining and this goal is easy."

Day 30: The final report#

Create a spending report I can share with my partner.

The 30-day report in numbers:

3-month total: $12,071
Monthly average: $4,024 (trending up)
Total saved over 3 months: $1,729 (~$576/month)
Savings rate trajectory: 33% in January → 12% in April

Areas to watch (Claude's words, not ours): dining growing fast, $96/month on unused subscriptions easy to cancel today, savings rate slipping and needs course correction.

The 30-day results#

Here's what actually changed, measured honestly.

Changes made during the experiment#

Spending: before vs after#

The savings rate recovery#

Not back to January yet. But the trajectory reversed.

What worked — and what didn't#

What worked:

The subscription and forgotten charges audits were the easiest wins. Zero lifestyle change, immediate savings. Anyone with 20 minutes should do this immediately.

The trend questions were more valuable than the snapshot questions. Knowing that dining was $748 last month is less useful than knowing it had grown 63% in four months. The trajectory is what triggers action.

Connecting two things Claude could see (the savings from cutting Adobe + DoorDash) to a third thing (funding the Amex payoff) was the kind of synthesis that's impossible when you're looking at data in pieces. That connection changed behavior more than any single insight.

What didn't work as well:

The weekend spending insight was clear, but behavior change was harder. Knowing you spend 73% more on weekends doesn't automatically make you spend less. It takes deliberate intention — planning cheaper Saturday activities, cooking Sunday instead of brunching. The data surfaces the problem; you still have to solve it.

Dining reduction took about 2 weeks to actually stick. The first week it felt like deprivation. The second week it started feeling like January again — which had been fine.

Can AI help you save money?#

The honest answer: yes, but not by telling you things you don't know. You already know food delivery is expensive. You already know restaurants add up.

What AI does is make the magnitude undeniable. You didn't know you'd ordered DoorDash 11 times in April. You didn't know Adobe was costing you more than Netflix and Spotify combined while sitting unused. You didn't know your savings rate had quietly collapsed from 33% to 12% in four months.

The information was always in your bank statements. You just couldn't see it.

Connect your bank once, ask a few questions, and within an hour you'll know things about your financial behavior that most people never figure out. Whether you act on them is up to you — but at least the decision is informed.

Over 30 days, one person with one bank connection found $430/month in ongoing savings and changed a debt trajectory that would have cost $891 in unnecessary interest. That's not a financial transformation. It's just information that was always there, finally made visible.

Try it yourself: the 5 questions that found the most money#

If you only ask five questions, ask these — in this order:

"Give me a full financial summary for last month" — the uncomfortable starting point
"List all my subscriptions and their monthly cost" — the easiest money you'll find
"Find recurring charges I might have forgotten about" — catches the rest
"Show all late fees and interest charges this year" — pure waste, immediately fixable
"What could I cut to save an extra $300/month?" — the action plan

Setup takes about 5 minutes at openbudget.sh. The first question takes about 30 seconds to answer. The rest of the hour is just deciding what to do with what you learn.

OpenBudget connects your bank to Claude via MCP — every transaction synced automatically, every question answered from your real data. Get started at openbudget.sh →