Custom GPT Proposal and SOW Generator: One Assistant That Already Knows Your Template and Rate Card

What This Builds

Every new deal right now starts the same way: a blank document, your SOW template open in another tab, and an hour or more spent retyping sections that barely change from client to client. This build fixes that by loading a Custom GPT once with your firm's SOW structure, your rate card categories, and the exclusions you always mean to write down and often forget. From then on, you feed it one filled-out intake form per deal and it hands back a full first-draft SOW, not a blank page.

Prerequisites

• Plus subscription ($20/month), since building a Custom GPT requires a paid plan
• Your firm's SOW template, rate card structure, and standard exclusions list, each as a short reference document
• A completed discovery intake form from one real or sample deal, to test the build

The only ongoing cost is the Plus subscription itself. Once it's running, every future proposal uses the same GPT at no extra cost.

The Concept

A Custom GPT is a version of ChatGPT you configure once and reuse forever, the way you might hand a new associate a binder of your standard templates before their first day. You write the binder's contents into its instructions and knowledge files a single time. After that, every conversation with the GPT starts already knowing your SOW sections, how you structure pricing, and which clauses you always exclude, so it drafts from that shared understanding instead of guessing.

Build It Step by Step

Part 1: Gather your inputs

Before opening ChatGPT, assemble three short documents:

1. SOW template structure, with section headers only: project overview, scope of work, deliverables, timeline, pricing structure, assumptions and exclusions, payment terms, signature block. No client data in this file.
2. Rate card structure, describing your service categories and how each is priced (hourly, blended rate, or fixed fee per deliverable type), in your own format.
3. Standard exclusions list, the clauses you add to nearly every SOW (after-hours support, hardware costs, third-party licensing fees, scope changes requiring a change order).

Keep each file focused on one topic. A single mixed document confuses retrieval later.

Part 2: Build the GPT

1. In ChatGPT, click your profile in the bottom left, choose My GPTs, and then click Create a GPT.
2. Switch to the Configure tab and fill in a name (something like "SOW Drafter") and a one-line description.
3. Paste instructions like this into the Instructions field:

You draft first-pass Statements of Work for an independent IT consulting practice.

For every request, you will receive a filled-out discovery intake form describing
one client engagement. Use the uploaded SOW template, rate card, and exclusions
list to produce a complete draft with these sections, in this order: Project
Overview, Scope of Work (numbered deliverables), Timeline, Pricing Structure,
Assumptions and Exclusions, Payment Terms.

Rules:
- Pull every deliverable directly from the intake form. Never invent a
  deliverable the form does not describe.
- Apply the rate card's categories to price each deliverable. If the intake
  form doesn't specify enough detail to price a line item, write "Needs
  consultant input" instead of guessing a number.
- Always append the full standard exclusions list unless the intake form
  explicitly says an item is in scope.
- If the intake form is missing information you need (client size, timeline,
  systems involved), list those gaps as questions at the top of your response
  instead of filling them in with assumptions.
- Write in plain, confident language a business owner can read without a
  technical background.

1. Under Knowledge, upload the three files from Part 1.
2. Under Capabilities, turn off web browsing, image generation, and code interpreter. This GPT only needs the files you gave it.
3. Set visibility to Only me and save the GPT.

Part 3: Test and refine

Feed the GPT a real or sample intake form and read the draft line by line. Two things to check specifically:

• Does every deliverable trace back to something the intake form actually said?
• Does the exclusions section match your standard list word for word, not a paraphrase?

If the GPT invents scope or drops an exclusion, tighten the instructions rather than fixing the individual draft. A single wording fix in the instructions carries forward to every future deal.

Real Example: Network Redesign and Migration Proposal

Setup: The GPT built above, loaded with a real SOW template, rate card, and exclusions list.

Input: A filled intake form for a 30-person accounting firm needing a network redesign and a Microsoft 365 migration, describing current pain points, rough timeline, and the systems in play.

Output: A complete draft SOW: project overview restating the firm's stated goals, eight numbered deliverables pulled straight from the intake answers, a phased timeline, a pricing section following the rate card's categories, and the full standard exclusions list appended at the end. Two flagged questions at the top note that the intake form didn't specify current mailbox count or existing backup vendor.

Time saved: Turns a task that used to take four or more hours of drafting into a 20 to 30 minute review and edit pass.

Before Anything Goes to the Client

Nothing this GPT drafts should reach a prospect's inbox unedited. Before you send a proposal, read every line item against the intake form it came from and check every rate against your actual rate card. Answer the flagged questions at the top of the draft yourself; the GPT raises them, but only you know the real mailbox count or backup vendor. You are the one who copies the finished text into your proposal document and sends it. The GPT never emails, files, or shares anything on its own.

What to Do When It Breaks

• The GPT invents a deliverable not in the intake form → Point to the exact sentence you expected it to use. Then add a stronger instruction telling it to quote the intake answer behind each deliverable.
• It ignores your exclusions list → Knowledge retrieval gets less reliable past a handful of files. Keep the exclusions document short and separate from the rate card rather than combining them.
• Pricing stops matching your current rate card → Knowledge files don't update themselves. Re-upload the rate card any time you change it, and remove the old version first.
• Formatting drifts away from your template's section names → Instructions that say "use these section headers exactly" hold up better than instructions that only describe the sections loosely.

Variations

• Simpler version: Skip the knowledge file uploads and paste the rate card and exclusions directly into the instructions field if they're short enough to fit comfortably.
• Extended version: Build a second Custom GPT for change orders, reusing the same rate card and exclusions files but with instructions scoped to a single added deliverable instead of a full SOW.

What to Do Next

• This week: Run two or three real intake forms through it and compare the drafts against what you'd have written by hand.
• This month: Add a conversation starter for the most common deal types you see, so you don't retype the same opening prompt each time.
• Advanced: Once you have a working proposal chain that ranks a prospect's pain points from a call transcript, feed that ranked list into this GPT's intake form field as your real starting input.

Advanced guide for IT consultant professionals. These techniques use more sophisticated AI features that may require paid subscriptions.