Skip to content
itsorted.online
AI guide

Generating Customer Quotes With AI From Your Price List, Safely

AI can turn a messy customer enquiry into a neat quotation in seconds: it reads "3 aircons, one leaking, Tampines condo" and picks the right line items from your price list. The danger is letting it do the maths or invent prices. This guide shows a safer design: a structured price list, AI that only chooses items and quantities, code that calculates totals and GST, a person who approves, and sending through your CRM or email. It works with ChatGPT, Claude or Gemini and tools like Make or Zapier.

By IT Sorted teamReviewed 7 min readHow we write guides
in

Key takeaways

  • AI models can misread numbers or invent prices; OWASP lists misinformation among the top LLM risks.
  • Store the price list as a sheet with item codes, so the AI picks codes and code looks up prices.
  • Structured output features make the AI return fixed JSON fields your workflow can check.
  • IRAS requires GST-registered businesses to quote GST-inclusive prices, with fines up to S$5,000.
  • A one-click human approval step catches wrong items before the customer sees them.
In this guide · 8 sections
  1. 01Why AI should not calculate the quote
  2. 02Step 1: turn your price list into structured data
  3. 03Step 2: let AI pick items in a fixed format
  4. 04Step 3: calculate totals, discounts and GST in code
  5. 05Step 4: human approval and sending
  6. 06Tools to build it with
  7. 07Do it yourself or get help?
  8. 08Frequently asked questions

Short answer

Keep your price list in a structured sheet, let the AI choose item codes and quantities in a fixed JSON format, let code calculate prices, discounts and 9% GST, and have a person approve every quote before it is sent. Never let the AI type a price or total itself.

Why AI should not calculate the quote

Large language models write fluent text, but they predict words rather than look things up or do arithmetic. They can round wrongly, use a price from an old list they saw earlier, or invent an item that sounds right. OWASP's Top 10 for LLM applications lists misinformation (LLM09) and excessive agency (LLM06) among the main risks, and both apply to a quote that goes straight to a customer.

A wrong quote is a commercial problem, not just an embarrassment: the customer may hold you to the price. Our guide to AI hallucinations as a business risk covers why this happens.

TaskWho should do itWhy
Understand the enquiryAIGood at reading messy text and photos
Pick items and quantitiesAI, from a fixed list of codesLimited to valid choices
Look up unit pricesCode, from your price sheetAlways the current price
Apply discounts and GSTCodeExact arithmetic, consistent rules
Write the cover messageAIFriendly, specific wording
Approve and sendA personFinal check before commitment

Step 1: turn your price list into structured data

Most quoting mistakes start with a price list in a PDF or someone's head. Put it in a sheet with one row per item.

  • Item code, such as SVC-GEN or PART-GAS, unique and short.
  • Description written the way customers describe it, plus common synonyms.
  • Unit and unit price, stated as before or after GST, and used consistently.
  • Rules, such as minimum quantity, call-out fee or "only with item X".
  • Valid from date and an owner who updates it.

Add a short rules tab for things like weekend surcharges, bundle discounts and areas you do not serve. These become checks in code, not instructions you hope the AI follows.

Step 2: let AI pick items in a fixed format

Give the AI the enquiry and the list of valid item codes and descriptions, and ask it to return only codes, quantities and questions. Most AI platforms now support structured outputs: OpenAI says these make responses follow a JSON schema you define, and Anthropic offers the same for Claude.

  • items: a list of item code and quantity pairs, where the code must be one of yours.
  • assumptions: anything the AI guessed, such as unit count or building type.
  • questions: what to ask the customer if the enquiry is unclear.
  • confidence: high, medium or low, so low-confidence quotes go to a person first.

Your workflow then rejects any code that is not in the sheet, and any quantity outside a sensible range. If the AI is unsure, the quote becomes a reply asking the customer a question instead.

Step 3: calculate totals, discounts and GST in code

A spreadsheet formula, Make or Zapier step, or a small script looks up each code's unit price, multiplies by quantity, applies your discount rules and adds GST at 9%.

If you are GST-registered, IRAS's page on displaying and quoting prices says written and verbal quotes must show the GST-inclusive price. Showing both is fine if the GST-inclusive price is at least as prominent, for example "S$109 w/GST (S$100 w/o GST)". "Prices are subject to GST" on its own is not acceptable. IRAS says non-compliance can mean a fine of up to S$5,000.

Step 4: human approval and sending

  1. Create a draft, not a sent quote

    Write the quote into your CRM, accounting app or a Google Doc template as a draft, with the AI's assumptions shown in a note for the approver.

  2. Notify the approver

    Send a short summary to email or WhatsApp with Approve and Edit links. Most approvals take under a minute.

  3. Check the three risky points

    Right items, right quantities, assumptions that hold. Prices and GST are already correct because code did them.

  4. Send from your normal system

    Approved quotes go out from your CRM, accounting software or email, with your quote number and terms, so they are recorded like any other quote.

  5. Log and follow up

    Record the quote and set a follow-up reminder. Our guide to automating enquiry follow-up shows how.

Once the approval rate is consistently high, you can auto-send a narrow class of simple quotes, such as a single standard service, while keeping approval for everything else.

Tools to build it with

You do not need custom software for a first version.

  • Make or Zapier to connect the enquiry form, the AI step, the price sheet and your CRM or email. Our Zapier vs Make comparison explains which suits which.
  • Google Sheets or Excel for the price list and calculations.
  • An AI model with structured outputs, such as OpenAI, Claude or Gemini, through the automation tool's built-in step.
  • Your accounting or CRM tool for the final quote document, so numbering and records stay in one place.

Customer enquiries contain personal data. Use a business AI plan or API that does not train on your data by default, and mention AI-assisted processing in your privacy notice.

Do it yourself or get help?

A comfortable Make or Zapier user can build a first version in a day or two, starting with the five most common jobs.

Safe AI quoting checklist

If you want it built for you, our Workflow Automation service (S$390 per workflow, up to 5 steps) connects your form, price sheet, AI step and CRM in your own account. To answer customer questions on your website before they ask for a quote, see AI Website Assistant (S$790).

Frequently asked questions

It can draft them, but it should not set the prices. Let ChatGPT or another model read the enquiry and choose item codes and quantities in a fixed format, then use a spreadsheet or automation step to look up prices and calculate totals and GST. A person should approve each quote before it is sent.

Language models predict text rather than calculate, so they can make arithmetic slips or use outdated prices that appeared earlier in a conversation. Code always uses the current price list and exact maths. Keeping calculations in code also makes every quote auditable.

If you are GST-registered, yes. IRAS says written and verbal quotations must show GST-inclusive prices. You may show the GST-exclusive amount as well, as long as the GST-inclusive price is at least as prominent. Phrases like prices are subject to GST on their own are not acceptable. GST is 9%.

Only for narrow, simple cases after a trial period, such as a single standard service with no options. Keep human approval for anything with several items, discounts or assumptions. The approval step usually takes under a minute and protects you from honouring a wrong price.

A price list in Google Sheets or Excel, an automation tool such as Make or Zapier, an AI model that supports structured outputs, and your CRM, accounting software or email to send the final quote. No custom software is needed for a first version.

Rather have it done for you?

Fixed price, set up in your own accounts.

Order online, get a written scope, and keep full ownership of everything we set up. Questions first? Ask us in writing.