← Back to Blog
The 15-Minute Project Estimate: How Founder-Led Agencies Use AI on Past Project Data to Quote New Work Without Underpricing It

The 15-Minute Project Estimate: How Founder-Led Agencies Use AI on Past Project Data to Quote New Work Without Underpricing It

November 11, 2026·7 min read

I quoted a website redesign from memory last month, and by week three I was working for free. The number was not the problem. The problem was that I pulled it from the best-case version of the last project, the one where the client answered email and nobody changed the homepage after launch.

Underpricing is a memory problem, not a rate problem

I have never met a founder who did not know their day rate. The rate is the easy part and it is rarely where the money goes. The money goes in the gap between what you thought the work would take and what it actually took.

I went back through one package we sell about three times a quarter and found the same pattern in 9 of the last 12 projects. We estimated 60 to 70 hours. We delivered between 85 and 100. Nobody flagged it, because the invoice had already gone out at a fixed price. At $150 an hour, a 25-hour miss is $3,750 of labor I donated. Four of those in a quarter and I have effectively hired someone and given their salary away.

This is why the conversation about how to price client projects usually starts in the wrong place. People want a better rate. What they need is a better memory of what the work actually cost them. Project estimation for agencies is maybe 20 percent pricing and 80 percent recall.

The data you need is already sitting in three systems

Founders assume AI for project scoping requires a big data project and a consultant. It does not. You need about 10 rows.

For most founder-led agencies, the history lives in three places: the time tracker (Harvest, Toggl, or the timer built into ClickUp), the invoice history (QuickBooks or Xero), and the scope document in Google Drive or Notion. None of this is new. It is just scattered, and nobody has ever put it in one row per project.

Pull 10 to 12 projects from the last 18 months that resemble what you are quoting now. For each one, capture six things: type of work, estimated hours, actual hours, revision rounds, number of people on the client side who gave feedback, and days spent waiting on approvals.

Those last three columns predict overruns better than the deliverable does. A project with one decision maker and two revision rounds comes in close to estimate almost every time. A project with a committee and a shared inbox does not, and now you can see that in writing instead of feeling it in your stomach at week five.

The 15-minute estimate, step by step

Here is the version I run now.

  1. Build the comparable table. Ten rows in Google Sheets or Airtable, six columns from the list above. If you have an assistant, this is the first task I would hand them, because it pays for itself on the very first quote.
  1. Write the new scope in five bullets. Not a paragraph, five bullets. If you cannot get it to five, the scope is not ready to be priced, and that is the most useful thing this exercise has taught me.
  1. Feed both into Claude or ChatGPT with a prompt that forces a range instead of a guess. Mine looks like this: "Here are 11 past projects with estimated hours, actual hours, revision rounds, stakeholder count, and days waiting on approvals. Here is the new scope in five bullets. Give me a low, base, and high hour range. Name the two closest comparable projects and explain why they match. Then list the three assumptions that would break this estimate."
  1. Add a buffer tied to named risk, not a flat 15 percent. If the new client has three approvers, look at what the three-approver projects actually took and price that. A flat buffer becomes noise everyone ignores, including you.
  1. Write the assumptions into the proposal. The three assumptions the model names become your exclusions section. That one paragraph has saved me more arguments than any contract clause, because it turns a surprise into something we both agreed to in advance.

What the AI should never decide

The model does not know that this client's marketing lead has quit twice, or that the last project with this brand took six weeks to approve a logo. It does not know you want this logo in your portfolio and would take the work at a lower number on purpose. All of that is your call.

So I never let it give me one number. I want a range, the two closest comps, and the assumptions. Then I decide what I say out loud.

It also does not set your rate. Your rate is a positioning decision, and sometimes a relationship decision. This is the most useful thing AI for consultants and agencies does right now, and it is not writing proposals. It is remembering. The judgment stays with you. If you ever catch yourself forwarding the raw output to a client, you have skipped the only part that required a human.

Make it a quoting system, not a one-time estimate

The estimate is not the valuable part. The habit is.

An agency quoting system is three small pieces: a six-question intake form in Tally or Typeform that maps to your risk columns, the comparable table in Airtable or Google Sheets, and ten minutes at every project closeout to log actual hours and revision rounds before the final invoice goes out. That last piece is the one everyone skips, and it is the only one that compounds.

After three months you have 20 real data points instead of a feeling. After a year you can quote a new project in 15 minutes with a straight face, and you can show a client exactly why the number is what it is. This is the boring part of founder operations systems, and boring is the whole point. The people who scale are not better estimators by instinct. They built something that remembers for them.

I keep a version of the intake form in the workshop library at https://workshop.mastermindshq.business, and there is a comparison of how this fits into a broader ops build at https://www.mastermindshq.business/mastermindvsl if you want to see the full shape of it.

If this resonates and you want to build this kind of infrastructure in your own business, the Mastermind is where we do the work live. You can learn more at mastermindshq.business.

Ready to put this into practice?

Join Joe Che's AI Business Mastermind, a small cohort for founder-led service businesses that want to systematize their operations with AI.