I had AI check my household finances every morning, and small daily fixes started adding up

Good morning. It’s まさきん.

Let me start with the conclusion. For household finances, one small daily improvement beats one big overhaul. It’s easier to keep up without forcing it.

My job involves visualizing daily numbers and turning them into systems. So I wondered if I could apply that to household budgeting too.

I combined the AI assistant Claude with data from the accounting app freee. The setup automatically reviews my card spending every day. I designed it to surface exactly one improvement suggestion at a time.

Here’s what’s inside that setup, what it’s been like to actually use it, and the tools involved.

It started because I never had time to look back

I’d been logging card usage in freee, the household/accounting SaaS. But logging was as far as it went. I never got around to reviewing the history and actually using it to improve anything.

I don’t have time on weekday mornings or evenings to carefully look back through past records. Honestly, it felt like too much of a hurdle.

So I wondered if I could hand off the “looking back” step itself to AI. That’s what led to this setup.

The tools and how it works

The foundation is simple. It combines these two things.

• Claude’s (Anthropic’s AI assistant) scheduled task feature: runs an automatic exchange with the AI at fixed times, 7am and 9pm every day
• freee’s API integration: via MCP (Model Context Protocol, a way for AI assistants to access external service data directly), Claude can read freee’s card transaction records (wallet_txns) directly

I’m not manually opening freee to check things myself. Claude goes and pulls freee’s data automatically at set times, then summarizes anything worth flagging.

The actual routine is split between morning and evening.

7am: one improvement suggestion from yesterday’s spending

1. Pull yesterday’s card transaction records from freee
2. Check for three things
   • Any single large expense
   • Subscription-like recurring or fixed payments
   • Spending in an unexpected category (like a string of late-night convenience store visits)
3. Cross-check against the log of past suggestions to avoid repeating the same point recently
4. Pick just one theme from that and present it as “one concrete thing to do today”

Here’s the actual instruction I give Claude, generalized with any account-specific details removed.

You are a household finance improvement assistant.
Using yesterday's card transaction data, create exactly one
"today's improvement suggestion" following these steps.

1. Pull yesterday's card transaction records
2. Check for the following
   - Any single expense of ¥5,000 or more
   - Subscription-like recurring or fixed payments
   - Spending in an unexpected category, like late-night convenience store use
3. Check the suggestion log from the last 14 days to avoid repeating the same point
4. Pick exactly one theme from the above and present it as
   one or two sentences describing a concrete action to take today

Summarize the output in three lines: "Theme," "Suggestion," "Reasoning."
Don't flag multiple things — narrow it down to one.

Keeping the prompt itself as text makes it easy to tweak later. For example, I can nudge it with something like “check harder for duplicate subscriptions.” That flexibility has been a real benefit.

9pm: reviewing the day’s spending

It tallies up the day’s spending total and category breakdown, and checks it against the rough monthly target.

Then it offers one point to watch out for tomorrow. This also doubles as a light check on whether I actually followed through on the morning’s suggestion.

Either way, the key is never flagging more than one point at a time. Too many suggestions at once means none of them get acted on. So I deliberately narrow it to “just this one thing today.” It’s much easier to actually follow through that way.

The output is automated too

Whatever the AI notices gets automatically logged in two places.

• A task management app: registered as an actionable “thing to do today”
• My daily note: recorded as the day’s insight, so I can look back on it later

This makes it harder to forget a suggestion. It’s also easy to look back later and see what got flagged during a given week.

Fixed, recurring payments are one of the check categories. Phone bills are the same kind of fixed cost, in that sense.

Once you sort that one out, you don’t need to keep touching it. That pairs well with a daily-review system like this.

What I’ve noticed doing this

Keeping it up, I’ve noticed small spending habits more often. For example, forgetting to cancel a subscription I wasn’t using anymore. I also found an add-on option I’d left signed up for without really meaning to.

Each one is a small amount on its own. But they add up to a bigger effect on the household budget than I expected.

One thing that’s worked especially well: handing off that first “look back and notice” step to AI.

I get to focus purely on “should I act on this or not.” The whole step of opening data and staring at it to notice something has disappeared. That alone lowered the psychological barrier a lot.

That said, I don’t take the AI’s suggestions at face value. I still make the final call myself.

Canceling a fixed cost or changing a contract especially needs to account for the household’s actual situation. Someone else might be using it, or it might be something we’ll need again later. So I treat AI as something that creates the “trigger to notice,” not the decision-maker.

Points I’m careful about

Running this setup, I keep the following in mind.

• Never leave the final call on canceling or changing a contract to AI, regardless of the amount: treat it strictly as a trigger for noticing
• Manage the credentials used for the API integration carefully: never share or expose the tokens used to connect to freee
• Don’t let a failed automation day become a “do nothing” day: on days the data pull fails, it falls back to general advice based on past trends, so the habit doesn’t break

Who this suits

• Anyone using freee or a similar household/accounting SaaS but not finding time to review it
• Anyone who’d rather stack small improvements than overhaul everything at once
• Anyone reasonably comfortable with AI assistants like Claude and API integrations

Wrap-up

I’ve come to feel that improving household finances is less about big decisions and more about stacking small realizations. I connected Claude and freee through MCP. Handing off the “look back” step to AI let me focus purely on making decisions.

A fixed cost like a phone bill is a good example too — sort it out once, and the effect keeps paying off.

I switched to Rakuten Mobile myself to cut my own monthly fixed costs. That ended up being one of the triggers for turning my whole household budget into a system like this.

This article contains affiliate advertising. If you sign up for a product or service through a link on this site, we may receive compensation from the partner company. The site operator is also an employee of Rakuten Group and may receive compensation through an employee referral program. The content and opinions here are based on the operator’s own experience and research regardless of any advertising relationship, but please read with the above in mind. See the Disclaimer & Affiliate Disclosure for details. Disclaimer & Affiliate Disclosure

This article includes machine-translated content. Please check the official Rakuten Mobile multilingual page for exact terms.

If you have concerns about Rakuten Mobile coverage or signal quality, you can consult through the official signal improvement request form .