AI Prompts for Project Budgets: When 60% Done Costs 90% of the Money

AI prompts for project budgets do not tell you whether the work is healthy. They put your plan and your ledger in the same table, so the gap between them becomes visible. Most projects do not die in the last month - they die in the silence between two numbers nobody places side by side.

Picture week three of a six-week project. The second invoice has already gone out. Half of the deliverables are still sitting in "almost". If the client asked you today for the real status, what number would you say, and what would you compare it against? That question has a cheap answer, and it is not a better status report.

Why 60% Done and 90% Spent Is Normal

It is normal, and that is what makes it dangerous: the pattern gets explained away in a meeting, at no cost, until the last invoice.

The Ninety-Ninety Rule, Read Out Loud at an Invoice

The first 90 percent of the code accounts for the first 90 percent of the development time, and the remaining 10 percent of the code accounts for the other 90 percent of the development time. Tom Cargill of Bell Labs gets the credit; Jon Bentley popularized it in his September 1985 "Programming Pearls" column in Communications of the ACM, where it ran under the title "Rule of Credibility" (Wikipedia).

The joke lands because the final stretch is where the strange cases, the approvals, and the unpaid rework appear. Apply it to a deliverable, never to the whole project: one stubborn screen can eat a week while four easy ones take the same time.

And the tail is real. In a sample of 1,471 IT projects, the average cost overrun was 27 percent, but one in six projects was a Black Swan with a cost overrun of 200 percent on average and a schedule overrun of almost 70 percent (Flyvbjerg and Budzier). Read that as a warning about outliers, not as a forecast for your project.

Two Clocks That Never Meet

Progress is measured on a clock made of deliverables. Money is measured on a clock made of invoice dates. The management error is not being late; it is comparing two different clocks and calling the result "on track".

Step 1 - Weight the Scope Before the First Invoice

Six to Ten Lines, Weights That Add to 100

Write the deliverables as six to ten lines and give each one a weight in percentage of the total price. One rule matters more than the rest: the weight is the share of the work, not the share of the invoice. If one deliverable carries 30 percent of the price but only 10 percent of the effort, you have found a problem worth seeing before you start.

A Gate, Not a Percentage: Draft, Approved, Paid

Replace the percentage with four states: not started, draft delivered, client approved, paid. "80 percent done" is not a state, it is a conversation. A gate either passed or it did not, and that is the whole point.

Step 2 - AI Prompts for Project Budgets: Build a Plan You Can Weigh

What the Prompt Returns

The first prompt is Project Manager Assistant, and it returns a plan in seven parts: the scope, with what is included and what is explicitly out; five to eight milestones with deliverable, owner and date in dependency order; a breakdown of the next two weeks into tasks of at most one day; a RACI matrix; a risk register with five risks; a weekly status report ready to send; and a thirty-minute sync agenda. It carries a written rule to flag unrealistic deadlines honestly, which most plans skip.

The Milestones Are Already Your Weights

Those five to eight milestones are the weight list from Step 1. You do not need a second document, you need to assign each milestone its percentage and make them add to 100. The status report is the other half: it is the text where progress turns into an opinion, and that is precisely why it should always be read next to the weight table, never on its own.

One framing instruction before you run it: declare the budget and the deadline in the prompt. With both numbers in hand, the plan gets measured against a real constraint instead of a wish.

Step 3 - Read the Cost Side Without Inventing a Number

What the Prompt Returns

The second prompt is Claude Financial Statement Analyst, pointed at your project ledger instead of a company's annual report. It returns a sanity check that says whether the numbers add up, key ratios with one line of interpretation each, trends across three years, red flags where every flag carries the number that proves it, a verdict in plain language, and three questions to answer before going further. Days sales outstanding is the one to watch here, because it is the column of money you billed and have not collected.

Why This Prompt Is Safe for a Project Ledger

Two written rules make it usable: calculate every number by showing the formula and the inputs, and never invent data that is not in what you pasted. Say it plainly: if you paste neither real hours nor real invoices, the output is not a diagnosis, it is a template with your project's name on it.

Step 4 - The Ten-Minute Reconciliation

The Six-Column Table

Now put the two sides in one table, one row per deliverable, and do the arithmetic yourself. It takes five minutes of typing and no AI at all.

Deliverable	Weight	Gate	Earned	Spent to date	Delta
Example: checkout flow	30	Approved (1.00)	30	34	-4
Example: admin screens	20	Draft (0.50)	10	18	-8

Earned value is the budgeted cost of the work actually performed, which exists because "percent complete" by feel cannot be measured. Weight times gate gives you earned; spent to date comes from your ledger. The delta is the only number that matters, and the ratio of earned to actual cost is what the discipline calls CPI, while earned against planned is SPI (Earned value management).

One Decision Per Row

Every row where spend runs ahead of earned allows exactly one decision: cut scope, rename the deliverable (no, renaming is not the same as cutting), invoice the approved milestone, or say out loud that this stretch was quoted badly. This article stays on the arithmetic and does not recommend rates or prices.

What the Prompts Cannot Know

Four things live only with you: your real hours, your rate, the revisions you gave away, and when you will actually be paid. If you do not log hours, the model can only reorder your memory, which is not the same as measuring it. The commercial hours never make it into a timesheet either, and they are the quietest cost in the whole table.

What to paste instead of opinions: the hour export and the invoice list with issue date and payment date. Then tell the model to mark as estimate anything that does not come from a document.

Reforecast at Every Milestone, Not at the End

When a milestone closes, readjust the weights for what is left rather than for what is done, and run Steps 2 and 3 again with the smaller remainder. Reweighting is not renegotiating the price: if the scope itself moved, that conversation belongs with the client, separate from the arithmetic.

FAQ

What is earned value in plain terms?

It is the part of the budget that your finished work has already earned. Multiply each deliverable's weight by its gate, add the rows, and you have the number to compare against what you spent.

Can ChatGPT track my project budget on its own?

No. It has no access to your invoices, your hours or your bank account. It can structure the plan, read the ledger you paste, and rewrite the status report with the delta in front.

How do I know a project is losing money before it ends?

Compare spend against earned on each row, not at the project level. When one row shows spend ahead of earned and the gap keeps widening across two milestones, the project is already paying for itself.

What do I paste into the financial prompt if I only have invoices?

Paste the invoice list with issue date, amount and payment date, and say that hours are missing. You will get a weaker but honest reading of the collected column, which is often the number that explains the cash problem.

The next status report you send can carry two columns beside the paragraph: earned and spent. If they disagree, you now know which row to open first, and the two prompts above do the reading and the rewriting around numbers you can defend (a business plan that adds up takes the same approach one step earlier).