13-Week Cash Forecast: A Five-Prompt Chain in One Claude Conversation

What This Builds

A week-by-week picture of your cash for the next 13 weeks: what you start with, what comes in, what goes out by category, and what is left. You build it in one Claude conversation, one prompt at a time, and you check the numbers after every prompt before you go on. The last prompt produces a one-page summary you can hand to your accountant. You end with a forecast you understand, because you supplied each assumption and Claude had to defend or revise it.

Claude writes the table and the questions. You decide what the numbers are, and your accountant tells you what they mean for your taxes and your books.

Prerequisites

• A Claude account. A free account can run this chain, though a long conversation may hit usage limits. A Pro subscription ($20/month) gives more room. That is the whole cost of the build.
• Rounded figures for your starting cash, expected weekly sales, and your regular bills with the week each is due. Round to the nearest hundred or thousand. Exact cents do not matter in a forecast.
• Ten minutes to check Claude's arithmetic after each step.

The Concept

A prompt chain is a conversation where each question builds on the last answer. Instead of asking Claude for a finished forecast in one go, which tends to bury guesses inside confident-looking numbers, you walk it through in stages. First it repeats your assumptions back so you can fix them. Then it builds the table. Then it argues against its own assumptions, shows you the thin weeks, and writes the summary. Each stage gives you a chance to catch a mistake while it is still small.

Keep it all in one conversation. Claude can then see the earlier answers, and your corrections stick.

What to keep out of the chat

Use rounded totals only. Leave out bank account numbers, the EIN, loan account numbers, staff names and anything that ties pay to a person. Write "payroll" as one line, not as a list of people. Before you start, check the data settings for your Claude plan, because a consumer chat plan may use conversations to improve models unless you turn that setting off. If you have a partner or lender who expects the numbers to stay private, tell them before you paste.

Build It Step by Step

Part 1: Gather your inputs

Write these in a note, all rounded:

• Starting cash today across your operating accounts, as one number.
• Expected sales for each of the 13 weeks, or a weekly pattern (for example, "about the same every week, with a lift on holiday weeks").
• When sales cash arrives. Card sales reach your bank on the schedule your payment processor sets, so check your own deposit pattern.
• Supplier spending per week and your payment terms.
• Payroll per pay period and the week each pay date falls in.
• Fixed bills with the week due: rent, utilities, insurance, loan payments, software.
• Tax payments and one-off items that your accountant has told you to expect.
• The lowest cash balance you are comfortable with.

Part 2: Prompt 1, restate the assumptions

Open a new Claude conversation and paste the prompt below. Fill each square bracket with your own rounded information.

I run an independent full-service restaurant. I want to build a 13-week cash forecast. Do not build the table yet. First restate my assumptions back to me as a numbered list, grouped as: starting cash, cash in, cash out by category, timing of each payment, and one-off items. Mark any assumption I did not give you as MISSING, and do not fill it in. Then list the questions you need answered before a forecast would be reliable.

Starting cash today (rounded): [starting cash]
Expected weekly sales, weeks 1 to 13: [weekly sales or pattern]
How and when sales cash reaches my bank: [deposit timing]
Supplier spending per week and payment terms: [supplier amounts and terms]
Payroll per pay period and pay dates: [payroll amount and pay-date weeks]
Fixed bills and the week each is due: [rent, utilities, insurance, loans, software]
Taxes and one-off items I expect: [tax payments and one-offs]
The lowest cash balance I am comfortable with: [cash floor]

Checkpoint 1. Read the numbered list. Does every figure match your note? Is anything marked MISSING that you do know? Reply to Claude with corrections and answers to its questions. Do not move on until the list is right, because every later table repeats it.

Part 3: Prompt 2, build the week-by-week table

Using only the corrected assumptions above, build a 13-week cash table. Use these columns: Week number, Starting cash, Cash in, Suppliers, Payroll, Rent and utilities, Loans insurance and other, Total cash out, Ending cash. Add a Total row at the bottom that sums Cash in and each cash-out column over the 13 weeks. Each week's Starting cash must equal the previous week's Ending cash, and Ending cash must equal Starting cash plus Cash in minus Total cash out. Round every figure to the nearest hundred. If you had to assume anything that I did not give you, list it under the table.

Checkpoint 2. Check these by hand, or paste the table into a Google Sheet and total it there:

1. In three or four rows, add up the cash-out columns and confirm the Total cash out.
2. Confirm Starting cash plus Cash in minus Total cash out equals Ending cash.
3. Confirm each Ending cash equals the next week's Starting cash.
4. Spot-check the Total row against the 13 weeks.
5. Read the list of assumptions Claude added. If it invented one, correct it with a reply.

Here is what a correct stretch looks like. These are invented example numbers for practice.

Check week 2 as an example. Suppliers 10500 plus payroll 13500 plus rent and utilities 9000 plus other 2500 is 35500. Starting cash 31000 plus cash in 36000 is 67000, and 67000 minus 35500 is 31500. Week 3 starts at 31500, the same number week 2 ended with.

If a row fails a check, tell Claude which row and which check, and ask it to redo the table.

Part 4: Prompt 3, challenge every assumption

Now challenge your own forecast. Go through each assumption in the table one at a time. For each, say what would make it wrong, whether the error would push cash up or down, and what question I should answer to firm it up. Do not change the table yet. Then add one stress case: show only the Ending cash line again for all 13 weeks if weekly sales come in [lower sales percentage] lower than my forecast and every cash-out line stays the same.

Pick a lower-sales percentage from your own history. Look at your worst recent week or month compared with normal and use something near it. Do not borrow a number from a statistic.

Checkpoint 3. Answer the questions Claude raised. Spot-check the stress line: Ending cash in week 1 should equal Starting cash plus the lowered Cash in minus Total cash out. Then reply: "Update the table with my answers, keep the same layout, and show the stress case again." Run Checkpoint 2 on the updated table.

Part 5: Prompt 4, mark the thin weeks and list options

Using the updated table, mark the three weeks with the lowest Ending cash, and mark any week where Ending cash falls below my cash floor of [cash floor]. For each marked week, say which lines drive it. Then list options I could weigh for those weeks, grouped as: timing of payments, sales, costs, and conversations to have with my accountant, landlord, lender or suppliers. Write each option as a question or a trade-off, with no recommendation. Do not give tax, legal or lending advice.

Checkpoint 4. Are the marked weeks really the lowest ones in the table? Does each option make sense for your restaurant? Cross out any that do not. Some options, such as asking a supplier for different terms, involve another party. Decide on your own whether to raise them.

Part 6: Prompt 5, the one-page summary for your accountant

Write a one-page summary of this cash forecast for my accountant. Include: the key assumptions in a short list, a compact version of the 13-week table showing only Week, Starting cash, Cash in, Total cash out and Ending cash, the thin weeks and what drives them, what changed after the challenge step, and a list of open questions for the accountant. Use rounded figures, plain language and no staff names or account numbers.

Checkpoint 5. Compare the compact table to the full table, row by row. Check that the summary contains no name, account number or detail you did not intend to share. Copy the result into a Google Doc, add your own notes, and send it to your accountant yourself.

Part 7: Update it each week

Each week, open the same conversation or start a new one. Paste the last table and your actual cash balance and say: "Roll this forecast forward one week. Replace week 1 with actual figures below, add a new week 13, and keep the same layout." Re-run Checkpoint 2.

Real Example: A Slow Stretch Caught Early

Setup: An owner with an invented starting cash of 20000 and flat weekly sales.

Input: The rounded note from Part 1, pasted into Prompt 1.

Output: Claude restates the assumptions and flags "insurance payment week" as MISSING. The owner checks the policy schedule and answers. The table shows cash dipping in the weeks when rent and a quarterly bill land together. In the challenge step Claude points out that the sales assumption ignores a quiet week after a holiday. The stress line shows one week close to the owner's cash floor. The options list includes a question about moving a supplier payment by a week, and a question to bring to the accountant about when a tax payment falls. The owner sends the one-page summary and asks both questions before the thin week arrives.

Time saved: Building the first version by hand can take an evening. The chain takes an hour or two, most of it checking, and the weekly update is short. The main gain is seeing a tight week early enough to act.

What to Do When It Breaks

• The table does not add up. AI models make arithmetic slips in long tables. Tell Claude the row and the check that failed, and ask it to redo the table. For a lasting fix, paste the table into Google Sheets and let Sheets do the sums.
• Ending cash and next week's Starting cash disagree. Say so in plain words and ask Claude to correct the chain from that week forward.
• Claude invents a number. Ask it to list every figure that did not come from you. Replace or remove them.
• The conversation gets long and Claude starts forgetting the early assumptions. Ask Claude for the current assumptions list and the current table, open a new conversation, and paste both in.
• The forecast drifts from reality. This is not a bug. Actuals will differ from a forecast. Roll the forecast forward each week with real balances, and note which assumptions missed so you can adjust them.

Variations

• Simpler version: Run prompts 1, 2 and 4 only, and skip the challenge and the summary.
• Extended version: Run the chain for two scenarios, a normal one and a slow one, and ask Claude to put the two Ending cash lines side by side. Put the finished chain's prompts into a Claude Project so you can reuse them.

What to Do Next

• This week: Run the chain once with rounded real numbers and send the summary to your accountant.
• This month: Update the forecast every week for a month and compare it with the real balance.
• Advanced: Pair it with the weekly numbers digest Zap, which gives you the prime cost trend in your inbox.

Advanced guide for restaurant owner-operators. A forecast is a planning aid built from your own assumptions. It is not tax, legal or financial advice, so your accountant reviews it before you act on it. Claude plans and features change, so check claude.ai for current details.