Claude, By the Numbers — The Data Plugin

Instead of learning spreadsheet formulas, you can point Claude at your revenue and expenses and ask plain questions, turning an unopened spreadsheet into a five-minute monthly financial check-in.

5:05 video5 min readWatch on YouTube

A folder called Finances sits on a lot of solo operators' desktops, twelve months of revenue inside it, never opened. Not because the numbers are unanswerable, but because opening the file means formulas, and formulas mean guessing at syntax and still not trusting the answer that comes out. Claude's data plugin removes that step entirely: point Claude at the same spreadsheet and ask a question the way you'd ask a colleague, and it explores, cleans, visualizes, and compares the numbers back to you in plain sentences.

The loop the plugin replaces

Without help, understanding your own numbers tends to follow a familiar, unproductive loop: export a CSV, open it, stare at rows and columns, get a vague sense that things are up or down, and close the file. The questions that actually matter, whether a client is fading, whether the mix has shifted, whether the business made money last month, go unanswered not because they're unanswerable but because answering felt too hard. Decisions end up riding on gut feeling instead of the data sitting right there.

Asking plainly, no formula required

The alternative is asking directly, the same way you'd ask a colleague sitting across the desk: which clients bring in the most revenue, and has the mix changed over the last two quarters? No formula, no pivot table. Claude reads the data, runs the analysis, and hands back a sentence you can act on, faster than it takes to write a single SUM formula. That covers straightforward questions too: total sales by month, which client led last quarter, whether software spending has crept up over six months. Each question gets turned into the right analysis and answered in a sentence, not a dump of raw numbers.

That kind of question can turn a vague feeling into something specific and actionable. "I think business is going okay" becomes "three clients carry 60 percent of revenue, and one has been sliding since September," which is the real gap between reacting to a problem after it's already hurt you and seeing it coming early enough to do something about it.

Four kinds of work

Beyond just answering questions, the plugin does four distinct kinds of work: exploration, cleaning, visualization, and comparison. Not every question needs all four, but together they turn a raw export into something you can actually trust and act on.

Exploration comes first: point Claude at an unfamiliar export with column headers that don't quite make sense, and it describes the structure, flags outliers, and suggests questions worth asking, often surfacing a better question than the one you started with. Cleaning handles the tedious reality that messy data is the norm rather than the exception: inconsistent date formats, duplicate rows, impossible values like negative sales. The plugin does that work and reports exactly what it changed, which is what actually lowers the barrier to working with your own numbers in the first place. Visualization turns a request for monthly revenue over the past year into an actual chart, a line chart for trends or bars for comparisons, ready for a deck, a partner conversation, or just your own sense of where the business is heading. Comparison is what separates a snapshot from a trend: how this quarter stacks against last, whether acquisition cost is climbing or falling. Totals tell you where you are; comparisons tell you where you're heading.

Making it a monthly ritual

Every business has a question it's been avoiding, and the plugin is built for exactly that kind of question. Point Claude at your invoices and expenses and ask directly, "am I profitable month to month?" It reads both files, correlates them, and gives you a real read on financial health, often surfacing more in a five-minute conversation than a year of meaning to look ever would. The suggested habit is to make it a ritual: on the first of the month, spend ten minutes pointing Claude at last month's revenue and expenses and asking for a health check, total revenue, total expenses, margin, how it compares to the month before, and anything that looks off. Doing that consistently means each review stacks on the last, until a trend becomes visible well before it turns into a real problem.

Revenue concentration is one specific question worth asking directly for any service business: who are your top clients, what share rides on the top three, and is anyone quietly spending less than before? If a single client makes up 40 percent of your income and you don't know it, you can't plan for the day they leave, and the plugin makes that check trivial to run, along with cheap follow-ups: of the top clients, which are declining, and what changed last quarter that might explain it?

What it can and can't tell you

The honest caveat runs through the whole approach: the plugin's analysis is only as trustworthy as two things, the data fed into it and the assumptions underneath it. Clean inputs and sound assumptions produce a reliable read; bad inputs produce confident, well-formatted nonsense. That's exactly why the plugin reports what it cleaned rather than cleaning silently, so the change log becomes an audit trail you can verify before trusting anything built on top of it. And there's a hard limit worth naming directly: the plugin can tell you a client is declining, but it can't tell you whether to call them, offer a discount, or let them go. It surfaces the number; the judgment stays yours.

Key takeaways

  • Claude's data plugin answers plain-English questions about a spreadsheet directly, without formulas or pivot tables.
  • It performs four kinds of work: exploration, cleaning, visualization, and comparison, and reports exactly what it changed during cleaning.
  • A monthly ritual, pointing Claude at last month's revenue and expenses for a health check, catches sliding clients or rising costs early.
  • Revenue concentration, how much income rides on your top few clients, is one of the most valuable single questions to ask.
  • The analysis is only as reliable as the data and assumptions behind it; the plugin surfaces numbers, but the decision stays with you.

Try it yourself

Paste in a spreadsheet of your own invoices and expenses for the year and ask Claude what's in each file, flag anything messy, then request a business health check: whether you're profitable month to month, your strongest and weakest months, whether any expense is growing faster than revenue, and one follow-up question to ask next. This walkthrough is part of Humanitarians AI's Claude Cowork Plugins series.

Chapters

  1. 0:00The naive framing: "you'd need formulas"
  2. 0:09Why the spreadsheet stays closed
  3. 0:40Ask it plainly — no formula required
  4. 1:13The anchor: three clients, sixty percent, one sliding
  5. 1:25Four kinds of work — explore, clean, visualize, compare
  6. 2:27The avoided question: am I profitable month to month?
  7. 2:40Make it a monthly ritual
  8. 3:06How concentrated is your revenue?
  9. 3:30The anchor returns — caught early
  10. 3:44Only as trustworthy as your data and your assumptions
  11. 4:08What it can't tell you
  12. 4:30Carry-out
  13. 4:41Your turn
  14. 4:59Outro
Full transcript(auto-generated, with timestamps)

The naive framing: "you'd need formulas"

[0:00]People assume understanding a year of revenue means learning spreadsheet formulas. It doesn't. You just ask in plain English. So, it's actually sitting in that file you never open.

Why the spreadsheet stays closed

[0:09]Solo operators often avoid their own numbers. A folder called finances, 12 months of revenue inside it, never opened because opening it means formulas and formulas mean guessing its syntax and still not trusting the answer. Without help, understanding your numbers looks like a loop. Export a CSV, open it, stare at rows and columns, get a vague sense things are up or down, close the file. So, the questions that actually matter go unanswered, not because they're unanswerable, but because answering felt too hard. Decisions ride on gut feeling instead. Point Claude at that same file

Ask it plainly — no formula required

[0:42]And ask the way you'd ask a colleague, which clients bring in the most revenue and has the mix changed over the last two quarters? No formula, no pivot table. Claude reads the data, runs the analysis, and hands back a sentence you can act on faster than it takes to write a single SUM. That's natural language querying. What were total sales by month? Which client led last quarter? Am I spending more on software than 6 months ago? Claude turns each question into the right analysis and answers in a sentence, not a table of raw numbers.

The anchor: three clients, sixty percent, one sliding

[1:13]It turns I think business is going okay into something specific. Three clients carry 60% of revenue and one has been sliding since September. That's the gap between reacting to problems and seeing them coming.

Four kinds of work — explore, clean, visualize, compare

[1:25]Beyond answering, there are four kinds of work: exploration, cleaning, visualization, and comparison. Not every question needs all four, but together they turn a raw export into something you can trust and act on. Exploration first. Point Claude at mystery export, column headers that don't quite make sense, and it describes the structure, flags outliers, and suggests the questions worth asking. Often the best question isn't the one you started with. Cleaning. Messy data is the norm. Inconsistent date formats, duplicate rows, impossible values like negative sales. The plugin does this tedious work and reports exactly what it changed. Automate that and the barrier to actually working with your numbers falls away. Visualization, ask for monthly revenue over the past year and you get a chart for a deck, a partner, or just to see your own trajectory. Line charts for trends, bars for comparisons, and you can always ask for something different. In comparison, how does this quarter stack against last? Is acquisition cost climbing or falling? Totals tell you where you are, comparisons tell you where you're heading.

The avoided question: am I profitable month to month?

[2:27]Every business has a question it's been avoiding. Start there. Point Claude at your invoices and your expenses and ask, "Am I profitable month to month?" It reads both files, correlates them, and gives you a real read on your financial health.

Make it a monthly ritual

[2:40]You'll know more about your finances after this 5-minute conversation than after a year of meaning to look. So, make it a ritual. First of the month, 10 minutes, point Claude at last month's revenue and expenses and ask for health check. Total revenue, total expenses, margin, how it compares to the month before, and anything that looks off. Do it every month and you're never surprised by your own numbers again. Each review stacks on the last until a trend is visible before it becomes a problem.

How concentrated is your revenue?

[3:06]For a service business, one question matters most. How concentrated is your revenue? Ask directly, "Who are my top clients? What share rides on the top three? And is anyone quietly spending less than before?" Concentration is one of the most important things a solo operator can know. If a single client is, say, 40% of your income and you don't know it, you can't plan for the day they leave. The plugin makes that check trivial.

The anchor returns — caught early

[3:30]And the follow-up is cheap. Of those top clients, which are declining? What changed last quarter that might explain it? The plugin makes each next question almost free, so ask it. Remember the client sliding since September? This is how you'd catch that early.

Only as trustworthy as your data and your assumptions

[3:44]Here's the honest part. The plugin's analysis is only as trustworthy as two things: the data you fed it and the assumptions underneath it. Clean inputs and sound assumptions give a reliable read. Bad inputs give confident, well-formatted nonsense. That's exactly why it reports what it cleaned, so you can verify the fixes before trusting the numbers built on top of them. The change log isn't paperwork, it's the audit trail.

What it can't tell you

[4:08]The plugin can tell you a client is declining. It can't tell you whether to call them, discount, or let them go. It surfaces the number, you own the decision. Four habits make it sing. Ask a specific question, not analyze my data. Iterate, your first question breeds a better one. Compare, don't just total. And export on a schedule, because the plugin can only work with the data you actually give it. The data plugin turns

Carry-out

[4:31]The spreadsheet you're avoiding into a 5-minute conversation, but the answer is only as trustworthy as your data and your assumptions. So, the judgment stays yours.

Your turn

[4:41]Your turn. Paste this into Claude. Here's a spreadsheet of my invoices and one of my expenses for the year. Tell me what's in each file and flag anything messy. Then, give me a business health check. Am I profitable month-to-month? Which months were strongest and weakest? Is any expense growing faster than revenue? And end with the one follow-up question I should ask next. Claude by

Outro

[5:00]The numbers, the data plugin. Liam in for Bear.

More from Claude

Humanitarians AI Lyrical Literacy Project