From inquiry to quote with AI: 4 steps to decide what AI handles
Split quoting into 4 steps: AI extracts the request and drafts the scope; you pick the questions and finalize the quote.
Key points
- AI extracts the request and lists gaps
- You pick the questions AI drafts
- AI drafts the scope, exclusions included
- You finalize the quote; AI does the math
Contents
AI extracts the request and lists the missing items
A&A perspective
Extracting the request is the step in quoting that is easiest to hand to AI. Extraction here means pulling the inputs for a quote, such as items, quantities and the deadline, out of the inquiry text you received.
From the sources
In December 2024, Anthropic, the US company that makes the conversational AI Claude, published an article on patterns for building AI into work. One pattern it describes splits a task into fixed steps in order, with each step taking the result of the one before.
From the sources
According to the article, this pattern suits tasks that can be cleanly split into fixed, smaller steps. It takes more time, but each AI call becomes an easier task, which raises accuracy. The article adds that you can place checks on intermediate steps to confirm the process is still on track.
From the sources
The same article also recommends starting with the simplest approach and adding complexity only when it is needed.
A&A perspective
Automate & Augment (A&A), the company writing this article, thinks quoting fits this pattern. The work from inquiry to quote splits into 4 steps: extracting the request, checking for missing items, setting the scope and finalizing the quote. Extraction is the first of them.
From the sources
Stripe, the US company that provides online payment infrastructure, publishes a Japanese-language guide to the requests for quote that companies send out. The guide says a request should give the supplier all the information needed to price accurately. The items it lists include the scope of work, quantities and delivery timing.
A&A perspective
Inquiries to a small company often arrive as a few lines in a web form, not as a formal request like that. In the extraction step, have AI return three things side by side: the items the inquiry stated, the items it left out, and the inquiry text each finding rests on. With the source text next to each item, you can quickly confirm that something really is missing.
Hypothetical example
Take a made-up example. A commercial print shop with 20 employees receives an inquiry that says only: "We want panels and flyers for a trade show next month."
Hypothetical example
AI picks up "items: panels and flyers" and "deadline: before next month's trade show." It then returns the missing items: panel size and count, flyer quantity, whether design files exist, and installation at the venue.
You pick the questions to send from AI's drafts
A&A perspective
Let AI draft the questions about the missing items, and choose yourself which ones to send. Checking for missing items is the step where you go back to the customer about anything extraction found missing.
A&A perspective
AI stops at the draft because the number and order of questions depend on your relationship with the customer. Ask about every missing item at once and some customers stop replying. With a first-time customer you might ask only about budget and deadline first, and save the rest for a meeting. Making that call is the owner's job.
From the sources
Stripe's guide to requests for quote lists, among the supplier's steps, using the set question period to ask the requesting company about anything unclear.
A&A perspective
A&A reads this as saying that even a formal request leaves a step where the supplier sends questions back. With an inquiry of a few lines, going back to the customer is all the more unavoidable. A&A expects that having AI draft the questions cuts the time spent writing them from scratch.
Hypothetical example
At the print shop, AI turns the four missing items into one question each. With the trade show close, the owner decides to send only two first: panel size and count, and whether design files exist. Flyer quantity and installation at the venue stay on the list as items to ask about in the meeting.
A&A perspective
Do not delete the questions you hold back; keep them on the list of missing items. When answers arrive, fill in the list and pass the completed list on to the scope step. Each step still takes the result of the step before it.
A&A perspective
This step also draws a line: AI does not send the questions to the customer. A question reaches the customer under your company's name, so a badly worded one reflects directly on your company's credibility.
| Step | AI handles | The owner keeps |
|---|---|---|
| Extracting the request | Returns the items stated, the items missing and the source text, side by side | Confirms against the source text that an item really is missing |
| Checking for missing items | Drafts one question for each missing item | Chooses which questions to send first, and sends them |
| Setting the scope | Writes the scope draft as included, excluded and on hold | Decides the on-hold exceptions and clears the group before the quote is final |
| Finalizing the quote | Calculates a draft amount from price-list unit prices and quantities | Decides discounts, rush fees and exceptions, then finalizes and sends the quote |
AI drafts the scope, down to what it excludes
A&A perspective
Once the answers are in, have AI write a scope draft that lists what is excluded as well. Scope means the line between the work you take on in this quote and the work you do not.
From the sources
Stripe's Japanese-language guide to writing quotes lists the information a quote should carry: item descriptions, prices, quantities, validity period, delivery timing and payment terms. It also lists other terms, such as special conditions, warranties and exclusions (what the quote does not cover).
A&A perspective
A&A thinks exclusions are easy to agree on in conversation and easy to leave unwritten. So have AI split the scope draft into three groups: included, excluded and on hold. On hold is for items you have no answer on yet, and for exceptions the owner decides.
Hypothetical example
In the print shop example, AI's scope draft looks like this. Included: printing the panels and flyers. Excluded: creating the design, on the assumption that the customer supplies the files. Delivery to the venue and installation: on hold.
A&A perspective
AI can list the exclusions because it has the list from extraction and the answers from checking for missing items. Have it write the scope draft from that list and those answers. Copy the agreed excluded lines into the other-terms section of the quote as exclusions. That way, the lines you drew in conversation stay in the quote.
A&A perspective
An on-hold group keeps AI from deciding exceptions on its own. It is enough to agree that customers you have discounted before, and rush jobs, go into the on-hold group and get flagged to the owner. Clear the on-hold group before you finalize the quote.
AI does the price-list math; you finalize the quote
A&A perspective
AI can do the math when it follows the price list, but finalizing the quote you send stays with the owner. Finalizing the quote means settling the scope, amount and terms, and sending them to the customer as a formal offer.
From the sources
Stripe's guide separates a quote from an estimate. A quote is a formal, fixed offer, and the amount on it is usually binding. If the buyer accepts, the guide says, the seller is obliged to provide the goods or services at the quoted price.
From the sources
An estimate, by contrast, is usually not binding. The guide says its figures are less precise and are often based on incomplete information.
A&A perspective
A quote you send stays with the customer as a promise. That is why A&A thinks AI should stop at multiplying price-list unit prices by quantities, while discounts, rush fees and exceptions stay with the owner.
Hypothetical example
In the print shop example, AI produces a draft amount from the unit prices and quantities of the panels and flyers. With the trade show coming soon, the owner decides whether to add a rush fee. The owner also decides whether to quote delivery and installation separately, then sends the quote.
From the sources
Stripe's guide also covers changes requested after a quote. It notes that what the customer wants to adjust is not always clear, and recommends asking questions to get the details. Rather than making the customer hunt for what changed, it suggests a short summary of the changes at the top of the quote.
A&A perspective
A&A's view is that you send changes back to extraction. Run a change request through the same 4 steps, starting from extracting the request, and the quote keeps a record of what changed and of the revised exclusions.
A&A perspective
That said, if your prices follow from items and quantities alone, a calculator built on the price list is probably faster than splitting the work into steps. If you get only a few inquiries a month, splitting the work may not be worth the effort.
Summary
A&A perspective
Hand extraction to AI and have it list the items stated and the items missing.
A&A perspective
Have AI draft the questions about the missing items, and choose yourself which ones to send.
A&A perspective
Have AI write the scope draft, down to the excluded and on-hold items.
A&A perspective
AI can do the math when it follows the price list; finalizing the quote stays with the owner.
Once the quoting work is split into 4 steps, each step has a clear owner: AI or you. The next thing to decide is a small test: run extraction alone on about 3 past inquiries and check whether AI's list of missing items matches what you actually had to ask.
FAQ
What if a customer asks for a rough price before you have the details?
Give an estimate, kept separate from the quote. Stripe's guide says an estimate is usually not binding and is used early, before the requirements are fully defined. A&A reads it this way: if you note that it is an estimate and state the assumptions behind it, you can explain any later difference from the quote.
Where should I start if I want to try this with ChatGPT?
The easiest start is to test extraction alone on about 3 past inquiries. Check whether AI's list of missing items matches the items you actually had to ask about. Before pasting in customer details, check your company's rules for handling them and the settings of the AI you use.
How should I measure the effect?
Compare two numbers before and after the trial: the time from opening an inquiry to sending the first reply, and the number of exclusions added after a quote went out. The sources for this article give no figures for faster replies or more orders won.
Sources & editorial note
Primary pages read for this article. Publication dates below belong to the sources; access dates record our research.
- Building effective agents
Anthropic · 2024-12-19
Accessed 2026-10-25 - 価格見積もり: 見積書を作成・管理する方法
Stripe · 2025-05-05 (最終更新日)
Accessed 2026-10-25 - 見積もりリクエスト (RFQ) の説明
Stripe · 2026-02-26 (最終更新日)
Accessed 2026-10-25