Turn your P&L into plain-English answers you can ask anytime

Upload your profit-and-loss statement once and interrogate it like a person. "Why did margin drop in March?" "Which costs grew faster than revenue?" Answers in seconds, in English.

What you'll have when you're done

A finance workspace where your own numbers live, and a habit of asking them questions instead of staring at them. Instead of waiting for your finance person to build a custom analysis, you upload the P&L, ask the question in plain language, and get a grounded answer with the math shown. It is the difference between owning a spreadsheet and being able to talk to it.

Reading my own financials used to be slower than it should be

I can read a P&L. Most operators can. But reading it fast enough to act is a different skill, and the honest truth is that the real questions ("what actually drove the margin dip, and is it a trend or a blip?") take a custom pull that I would have to ask someone for and wait on. By the time the answer came back, the moment to act on it had often passed. Worse was the version where I did not even ask. The number looked roughly fine at a glance, I moved on, and the slow leak kept leaking until a quarter later it was big enough to be obvious and expensive. The questions I never asked cost me more than the ones that came back late.

The unlock is mundane and large: a model can hold your whole P&L in its head and answer follow-ups instantly. You ask, it answers, you ask the sharper version of the question, it answers again. No ticket, no wait, no re-explaining. The catch is that an AI will confidently mislead you if you feed it the statement without context, so the setup matters more than the prompt.

What you need first

• Your P&L exported as CSV or Excel, ideally a trailing twelve months by month so trends are visible.
• A Claude Project on a business plan. Your financials are sensitive; keep them off a personal account (is your data safe in AI).
• One fact about your own books: whether they are cash or accrual basis. This single fact changes how every number should be read.

Step-by-step

Step 1Export a trailing-twelve-month P&L

Pull your profit-and-loss as a flat file, monthly columns across a trailing twelve months. Monthly granularity is what lets the AI see "March looks different" instead of just a yearly lump. As with the close workflow, export clean detail, not a report with subtotal rows mixed into the data.

Step 2Brief the Project before you ask anything

In your finance Project instructions, state the three things the model cannot guess:

• Your accounting basis. "These books are accrual basis." On accrual, a margin dip is about earned revenue and incurred cost; on cash basis it might just mean a big customer paid late. Get this wrong and every answer is subtly wrong.
• Which rows are subtotals. "Gross profit, operating income, and net income are subtotal rows, not line items. Do not add them into category sums."
• The reviewer rule. "Only use numbers from the file I uploaded. If a number is not in the file, say so. Never estimate."

Step 3Ask the question you actually have

Upload the P&L and ask in plain English. The good questions are the ones you would put to a CFO:

- Why did gross margin drop in March versus February? Show the lines that moved.
- Which expense categories grew faster than revenue over the last 6 months?
- What is my real gross margin if I exclude shipping costs?
- Which three line items are my biggest month-over-month swings, and are they one-offs or trends?

Because it is a conversation, the second question is where the value is. "Okay, that vendor line jumped, is it growing every month or was March a spike?" Keep pulling the thread.

Here is the shape of a grounded answer to "why did gross margin drop in March?", illustrative:

Gross margin dropped from 38.1% in February to 32.4% in March.

Revenue was actually up 6% ($410K to $435K), so this is a cost story, not a demand story. Three COGS lines moved:

Line	Feb	Mar	Change
Fulfillment	$94K	$128K	+$34K (+36%)
Payment processing	$12K	$13K	+$1K
Cloud hosting	$8K	$19K	+$11K

Fulfillment is the main driver: it grew $34K against only $25K of new revenue, meaning each order cost more to ship. Hosting more than doubled. Of the 5.7-point margin drop, fulfillment explains roughly 4 points and hosting about 1.5.

One caution: I am reading "Cloud hosting" from the COGS section as you labeled it. If it belongs in operating expense instead, the gross-margin math changes. Confirm that classification.

Notice what makes that useful rather than dangerous: it shows the lines that moved so you can check them, it does the subtraction for you, and it flags its own assumption about where hosting belongs instead of burying it. The next prompt writes itself: "Is fulfillment cost per order actually trending up, or was March a one-month spike?" That follow-up is what tells you whether to renegotiate with a shipper or just absorb a blip.

Step 4Verify one number, then trust the pattern

Before you act on an answer, check one figure it cited against the source statement. The fastest check is to take one number from its answer (the $128K fulfillment line above) and find that exact cell in your P&L. If it matches, the model is reading the file correctly and you can trust the pattern-level answers. If it is off, the file structure confused it and every number is suspect. The point of verifying is not paranoia; it is calibration. After a few checks you will know how much to trust it, and the answer is usually "a lot, for analysis; never for numbers it wasn't handed."

Make it yours. A SaaS CEO asks different questions than an agency owner. If you run on recurring revenue, the sharp ones are about retention and expansion ("which cohort is churning fastest, and what did that do to MRR?"). If you run a services business, they are about utilization and project margin ("which client work came in under my target margin last quarter?"). Tell the Project what kind of business you are in once, in its instructions, and the same workflow starts asking the questions that fit your model instead of generic ones.

How you'll know it's working

You start asking financial questions in the moment a decision needs them, instead of parking them for the next finance sync. The lag between "I wonder why" and "now I know" collapses from days to seconds. And you catch things earlier, the slow-creeping cost line that would have shown up in a quarterly review now surfaces the first month you ask.

When it breaks

• The numbers do not add up. Subtotal rows are being counted as line items. Re-state which rows are subtotals (Step 2), or re-export cleaner.
• It draws a wrong conclusion about a dip. It does not know your accounting basis. Tell it cash or accrual explicitly.
• It invents a figure. Tighten the "only use numbers from the file" instruction. If it persists, the file may be malformed and it is guessing at structure.
• Answers feel generic. You uploaded one month. Give it the trailing twelve so it has something to compare against.
• It gives you a percentage with no dollar figure. Always make it show both. "Marketing up 38%" is meaningless until you know it is $4K versus $4M. Add "show the dollar change next to every percentage" to your Project instructions.
• It compares the wrong months. If your export ends in a partial current month, the model may treat the stub as a full month and report a fake collapse. Tell it which column is incomplete, or trim the partial month before uploading.
• You are about to put its number in front of the board. Stop. For anything external, verify every figure against the source. This workflow is for your understanding and speed, not for generating numbers you present as audited.

Where this fits in your harness

This is the middle of the finance chain. It assumes your books are closed and clean, because a P&L Q&A is only as honest as the ledger under it. Once you can interrogate the past, point the same Project forward with a cash-flow forecast you trust, and when board season comes, let it draft board-ready financial summaries from the same numbers.