Turn a three-hour bank reconciliation into a 15-minute review with Claude Cowork

Bank reconciliation is one of the most recurring jobs in any business, and one of the easiest to put off. It happens every month, the rules barely change, and until recently a person still had to sit with two exports and match them line by line. Some people put it off for months, and one of the hosts admits skipping it entirely in a first business until taxes came due.

In EP 113 of The Biggest Goal Podcast, Alane Boyd and Micah Johnson explain why reconciliation resisted automation for so long, what changed, and the exact prompt that gets it done in Claude Cowork. Micah sets the target up front: "We can take these three hours and turn this into a 10, 15-minute job max."

Why reconciliation stayed manual for so long

Export a bank statement, export QuickBooks, and put them side by side. Nothing lines up. There are no matching IDs, and the descriptions for the same charge are not even written the same way. Traditional automation runs on exact logic, so it had no reliable way to walk through the two lists and decide which lines belong together.

Alane compares it to the matching worksheets kids do in school: a column on the left, a column on the right, and a line drawn between the things that go together. "That's literally what we're doing here with reconciliation," Micah replies. The harder versions of that worksheet pair a broccoli floret with a whole head of broccoli, and that kind of match is exactly what an if-this-then-that rule cannot see.

To automate it the old way, you would need a separate logic path for every variation of every vendor name: AWS, Amazon Web Services, Amazon Web SVCS. Every new charge would send you back into the automation to edit it, which defeats the purpose of building it.

Why AI handles the matching

AI is built for this kind of work: take a pile of data, analyze it, and make connections. It can read AWS, Amazon Web Services, and Amazon Web SVCS as the same vendor, and Salesforce and Salesforce.com as the same vendor. Then it checks whether the amounts match and whether the dates fall within three days of each other. If all three line up, the line item is a valid match, and it works through the whole file in seconds.

The hosts split the job into two halves. The matching work is low value for a person, even though it has to get done. Looking at what did not match and solving it is high value, and that is where a finance expert should spend their time: asking questions and making decisions, not ticking off rows that already agree.

They are also clear about what this is not. It is not chat. You could build it as an AI agent in n8n, which is how they would have described it in January 2026, but it would take longer to build. Claude Cowork is their default, and they mention ChatGPT Work as another option.

The prompt you can run today

Micah walks through the whole setup out loud in the episode. Gather the files you want to reconcile (two or more, since Alane has reconciled several at once), hand them to Claude Cowork, and ask for something close to this:

"I want you to please reconcile these files. I want you to look at the descriptions and find the ones that are similar, that have matching amounts, that have dates that are within three days."

Then ask for a set of findings: what matched, what did not, and why Claude thinks each item did not match. Micah calls those three rules a great rule set to start with.

Claude comes back with a full document listing every matching line item so you can scan down it quickly, plus the items that did not match and its reasoning for each one. The reasoning is what makes the review fast. An item might be a check written at the end of the month that has not cleared the bank yet, which is why it is missing from the bank statement. Another might be a bank fee that has not been entered in QuickBooks. You work down the list confirming each explanation.

Make it a skill after the first run

Alane uses the same approach on the statements from her investment property management company, where rents and building work all need to be accounted for. She drops everything in, including credit card charges and bank statements, and asks for a spreadsheet. The first time, Claude asked what format she wanted back. "I said a spreadsheet and then that's all I had to say," she recalls.

What came back was, in her words, "the most beautiful spreadsheet you've ever seen." It categorizes everything, adds notes explaining why it did what it did, and ends with a section of items it could not find, with a suggestion to raise them with the right person or a CPA. "I am not saying we don't need those experts, but it got me a whole lot farther than me having to do all that manually," Alane says.

For the months after that, Micah's advice is simple: "At the end of the first time doing this, just ask Claude, hey, make me a skill so that I can repeat this over and over." The skill keeps your preferences, your structure, and your spreadsheet format, so next month you run it with the new files and skip the explanation.

Before Cowork, Alane says the work took her eight to 10 hours to assemble, go through, and double check. With the skill, it takes her about 30 minutes. She keeps refining the rule set too. Some rules only become obvious once you see the first results, and each one she adds to the skill means less time next month.

Where the person stays in the loop

When someone says they automated reconciliation, it sounds like nobody is watching. The hosts reject that. AI should not be touching money, and it should not be making the final calls. Its job is the heavy lifting of line-item matching, delivered in a format a finance professional can read quickly, so that person spends five to 30 minutes reviewing instead of seven to nine hours matching.

The hosts' own bookkeeper can run the same process. Their results land in a Google Sheet, so when she has a question, she leaves a comment and tags Alane or Micah.

The same method works anywhere two lists should agree. Micah points to companies that do media buys, where the credit card statement has to be reconciled against spend that the team, not necessarily the bookkeeper, is executing.

Start small with your own reconciliation

The hosts close with a short checklist: give Claude the files, keep the decisions with a person, and, as Micah puts it, "Don't over engineer. Start small and start simple." Get the first results, bank the time saved, and then improve it. After a run or two you will notice what you wish it did next, and you can ask Claude to add it. Alane's more advanced setup, where Claude Code pulls the monthly statement from her email and files it in Dropbox under the right name, came after that first simple run, not before it.

If you want to get your own reconciliation done with AI this month, working at your own desk, the AI bank reconciliation course walks you through it. It is a three-lesson course you do inside Claude, and you finish with a reconciliation system running on your own books, saved as a skill you can rerun every month.

Bank reconciliation is the task everyone wants to automate, and before today's AI tech, nobody could. Alane Boyd and Micah Johnson explain why it stayed manual for so long, then walk through the exact prompt that turns eight to ten hours of line-item matching into a short 10 min review.

When you export your statements from your bank and QuickBooks (or another accounting platform), there are no matching IDs, no shared descriptions, and nothing a traditional automation could key off; somebody had to sit down and compare them line by line, and that somebody was expensive.

In this episode, you’ll learn:

• Why rules-based automation could never reconcile your books and what changed when AI could read AWS, Amazon Web Services, and Amazon Web SVCS as one vendor
• The three-rule starting prompt that matches on description, amount, and a three-day date window
• How to read the exceptions when AI explains that a check has not cleared or a bank fee has not hit QuickBooks yet
• How to turn the first run into a reusable skill so you never re-explain your setup again
• Where the human stays in the loop, because the matching is low-value work and the exceptions are not
• How the same approach applies to investment property statements, media buys, and anything else with two lists that should agree

Reconciliation might be a boring task, but that's exactly why it is a great place to point AI first. Press play and start automating your reconciliation workflow this week.