Give Your AI Your Cost Records So Estimates Start From Your Real Numbers.

By · Published · Updated · About 7 min read

On this page (7)
  1. Why a generic assistant cannot estimate your jobs
  2. What you need from the assistant: cost records for AI estimate accuracy
  3. Build the context file
  4. Before and after: cost records for AI estimate accuracy
  5. What changes when the assistant knows the operation
  6. What this is not: cost records for AI estimate accuracy
  7. Stop points

You have a kitchen remodel to bid: 200 square feet, gut to the studs, quartz counters, custom cabinets. You ask ChatGPT for a scope of work.

It returns generic line items. "Demolition: a per-square-foot range." "Electrical: consult a licensed electrician." "Cabinets: varies by material and finish."

It does not know that you use the same cabinet supplier on every job. It does not know that your electrician charges a flat rate for kitchen circuits. It does not know that you built three nearly identical kitchens last year and know what each one cost.

You end up converting generic output into real numbers from records you already keep. This page shows how to let the assistant read those records first.

Why a generic assistant cannot estimate your jobs

Every conversation starts blank. ChatGPT does not know:

You can paste a material list or describe a scope. When the next kitchen request arrives, you explain your process again. The knowledge exists in your past estimates, supplier invoices, and subcontract agreements. The assistant does not see it.

What you need from the assistant: cost records for AI estimate accuracy

You need an assistant that knows your operation without a briefing on every bid.

It should price from your costs. When it drafts an estimate, it uses your labor rates, your suppliers, and your markup, not national averages or ranges.

It should know your crew. It knows which electrician does your residential work, which plumber handles commercial, and which HVAC company gives you the best terms. When it drafts a schedule or an estimate, it uses those relationships.

It should reuse past scopes. You have built 30 decks, remodeled 50 bathrooms, and finished a dozen basements. The assistant should pull line items from similar completed jobs, not generic work descriptions.

It should match the client. Some clients want a detailed breakdown. Some want one total. Some need a payment schedule and a warranty explanation. The assistant should use the format that works for your business.

Build the context file

Keep a folder of Markdown files, such as an Obsidian vault. Claude Code reads a file named CLAUDE.md at the start of each session. That file tells the assistant who you are, how you price, and where the records live. You stop typing "I'm a general contractor specializing in..." at the start of every session.

The file and its links hold five things.

Your pricing structure: labor rates by trade, material markup percentages, overhead method, and target margin.

Your subcontractor network: electrical, plumbing, HVAC, drywall, and painting, with rates, availability patterns, and specialties.

Your past project library: links to completed job folders with photos, scopes, final costs, and change orders.

Your material specifications: preferred suppliers, the brands you install, and the products you warranty, such as your cabinet maker, flooring distributor, and window supplier.

Your client templates: proposal format, payment schedule structure, warranty language, and change order process.

## Pricing (owner-maintained)
Labor rates by trade: pricing/labor-2026.md
Markup and overhead: pricing/markup.md
Material prices: pricing/materials.md (each line dated)
Rule: if a material price is older than 30 days, mark it "verify" in the draft.
Rule: list quantities and unit costs; do not present a total as final.

## Subs
Electrical (residential): SUB-E1, flat rate per kitchen circuit set; 2-week lead
Plumbing (commercial): SUB-P2 | HVAC: SUB-H1 (best terms on ductwork)

## Past jobs
jobs/2025-K-07 kitchen, jobs/2025-K-11 kitchen, jobs/2025-K-15 kitchen
jobs/decks/ (30 builds, composite and pressure-treated)

The file changes as the business changes. Add a new supplier once. Link a finished job folder. Update the labor rates when they change. The assistant sees each change at the next session, with no retraining.

Before and after: cost records for AI estimate accuracy

A bathroom estimate

Before: you describe the scope, paste the rough square footage, and list the fixtures. The assistant returns national ranges and generic line items. You spend 30 minutes replacing every number with your costs and supplier prices.

After: you prompt, "Draft an estimate for job B-2609, standard bathroom scope, upgraded fixtures." The assistant pulls your bathroom template from past jobs and applies your labor rates and markup. It uses your fixture suppliers with the dated prices on file. You check quantities, confirm any price marked "verify," and total it in your estimating sheet.

A change order

Before: a client asks to add a deck to a kitchen remodel. You search past jobs for a similar deck, price materials from current lumber costs, estimate labor hours, add your markup, and write the change order.

After: you prompt, "Add a 12x16 composite deck to job K-2604." The assistant reads three similar deck builds, applies your supplier prices and deck labor rate, and fills your change order template. You review it and send it for the client's signature.

A sub schedule

Before: a whole-house renovation reaches rough-in. You check the electrician's availability, call the plumber about timing, and text the HVAC company about ductwork. Then you build the schedule.

After: you prompt, "Schedule subs for job R-2611: demo complete, ready for rough-in." The assistant reads each sub's availability pattern and proposes a sequence from past multi-trade jobs. It drafts a coordination message to each sub with the address and timeline. You confirm dates by phone and send. Dependencies across trades follow the same pattern as AI memory for project managers.

What changes when the assistant knows the operation

Estimates stop being research projects. When a client asks for a quote, the assistant starts from your pricing, your similar jobs, and your suppliers. You edit real numbers.

Mid-project changes get easier to price. The assistant references past work, lists real costs, and fills your change order format.

Sub coordination gets consistent. The assistant knows who you use, what they charge, and how each prefers to hear from you. It stops giving "hire a licensed plumber" advice.

Proposals carry the elements that persuade your clients: past project photos, specific materials, clear payment terms.

Job history becomes searchable. "Show me all deck builds over 200 square feet." "What did the last kitchen remodel cost?" "Which jobs ran over budget, and why?" The assistant reads the linked job folders to answer. When an estimate misses, write the reason into the job folder. How corrections become part of the system explains why that note matters more than the chat.

What this is not: cost records for AI estimate accuracy

This is not project management software or an estimating platform. You do not migrate data or learn a new system. You write files that tell the assistant what it needs to know about your business.

Claude Code needs a paid Claude plan or an API account; the free plan does not include it (Claude Code: Advanced setup, checked 2026.09.25). Obsidian is optional and free to download. Claude Code sends the content it reads to the model provider over the network. Claude Code: Data usage Keep client home addresses, gate codes, and payment details out of the folder; use job IDs.

You control what the assistant knows. The context is yours, in plain files that any assistant able to read files can use.

Stop points

The assistant drafts estimates, change orders, and messages. You approve every price, every scope, and every message before it reaches a client or a sub. Who approves what an AI agent sends shows how to write that rule down. Trade-specific records, such as panel schedules and code notes, can follow the electrician version of this setup.

Keep a record of cost records for AI estimate accuracy

Next steps for cost records for AI estimate accuracy

Topic map for cost records for AI estimate accuracy

Every page in this topic

  1. Give your AI assistant a business context file that ends the daily re-briefing
  2. Give your AI an appraisal practice file so comp selection follows your method
  3. Keep a project registry your AI reads so every answer fits the right job
  4. Measure the briefing and rework cost of using AI in your business.
  5. Keep a company context file so AI help starts from your real strategy
  6. Document your chiropractic practice approach so AI drafts sound like your clinic
  7. Build a church context file so AI drafts match each ministry and volunteer
  8. Keep client files your AI can read so session prep starts from the whole history
  9. Keep engagement files your AI reads so deliverables build on the work already done
  10. Write your brand voice into files so AI drafts sound like you on every platform
  11. Give your AI your cost records so estimates start from your real numbers
  12. Keep client voice files your AI reads so copy matches each brand from the first draft
  13. Keep a tech stack file your AI reads so advice fits the systems you run
  14. Give your AI your schema and metric definitions so its first SQL query runs
  15. Document your dental practice voice so AI patient messages need light edits
  16. Give your AI your catalog and customer file so product copy fits your store
  17. Keep one file per event so AI drafts never mix up clients, vendors, or dates
  18. Write your executive's operating manual into files so AI drafts follow the rules
  19. Give your AI your firm's compliance language so drafts reach review in better shape
  20. Build project files so AI drafts on-brand copy for each design client
  21. Log member, class, and campaign results so AI plans from your gym's history
  22. Document your HR policies once so AI drafts match your handbook
  23. Keep a project file per client so AI sources to their style and budget
  24. Keep property records so AI starts each spring with last season's facts
  25. Keep client session files so AI follow-ups reference real progress
  26. Split brand, audience, and campaign files so AI drafts on-brand copy
  27. Keep lender and product files so AI compares loans from your real options
  28. Record grants, donors, and board history so AI drafts reports with continuity
  29. Record your commission and state rules so AI answers from your jurisdiction
  30. Document roles, processes, and vendors so AI works from how your operation runs

More in AI by role

Neighboring topics

Services for this work