Quote Every RFQ In Under 5 Minutes With This AI Agent (Claude + monday.com)

What Should A Quotes Board In monday.com Track?

Every RFQ becomes one item on the quotes board, added by hand or automatically. Ours holds the customer name, company, email, phone number, date received, requested items, status, quote total and a link to the quote document.

Each quote then moves on to accepted or declined. The sooner a quote reaches the customer, the better the chance they buy, so the whole build is about cutting the time between an RFQ arriving and the quote going out.

What Product Data Does An AI Agent Need To Price A Quote?

A product list with three key data points: the name, the SKU and the unit price. It can sit in an Excel spreadsheet, Google Sheets, monday.com or another system, as long as the agent can reference it.

Our demo board also holds the unit of measure and stock level, and links each product to an optional suppliers board showing who supplies it. The quotes board and the product list are the parts you can't do without.

How Does The RFQ Agent Work?

It reads incoming emails and works out which ones are RFQs, since not every email will be. For each RFQ it adds an item to the quotes board, checks the requested products against the products board and builds a quote with those line items, laid out like your standard quote.

It then saves the quote to a Quotes folder in Google Drive, adds the link to the item and marks it as prepared, ready to send. Without it, a person reads each email, types it into the system and starts the quote by hand.

This is the process we built for the video. The agent can be set up for a different process, or for RFQs that arrive through more than one channel.

Should An AI Agent Send Quotes Automatically?

Not to begin with. In this build the agent stops at a prepared quote, and a salesperson checks the document, makes sure every requested item is on it and nothing has been missed, then sends it themselves.

The agent can get things wrong, especially while you're still testing and refining it. Once you're confident in it, sending can be automated too, and the time from an RFQ arriving to the quote going out comes down to a matter of minutes.

How Do You Write The Instructions For A Quoting Agent?

Have Claude write them from a spoken description of the workflow rather than writing a system prompt by hand, and give it plenty of detail. Our description had the agent check whether each email is an RFQ and, if it is, create an item on the quotes board.

From there it takes each product's unit price, unit of measure and supplier from the products board, builds the quote, saves it to the Quotes folder and changes the item's status to quote prepared, with the link added. We asked Claude for two things back: the system prompt and a skill for building the quote.

1. In the Claude Console (you need admin access), open the Agents tab and create an agent.
2. The console's summary box has no dictation, so dictate the workflow into a normal Claude chat and attach a PDF of a quote you send day to day.
3. While Claude works, create a folder called Quotes in Google Drive. OneDrive or SharePoint would do the same job.
4. Paste what Claude wrote into the summary box, then press Generate. It fills in the agent's name, description and system prompt.

Claude returned the RFQ quote builder skill and the system prompt, column IDs included, as our Claude is connected to monday.com via MCP. It also noted details such as unit prices showing in dollars and free delivery always being printed: fine for a demo, but they'd need getting right for real use.

The generated system prompt was much shorter than the instructions we pasted in. That's what we want, as shorter instructions are easier for the agent to follow correctly.

Which Model, Tools & Skill Does The Quoting Agent Need?

We used Opus 5.5 with the effort level on low, as quoting isn't overly complicated. Opus is probably overkill and costs a little more, but a less capable model is more likely to hit problems or get stuck, and for anything more sophisticated we'd raise the effort level or the model.

It connects to monday.com for the boards, Gmail to read the RFQs and Google Drive to save the quotes, sharing one vault for the credentials. We set every tool to Always allow, as Always ask needs approval at every step and we see little risk here. Single actions, such as creating items, can still be switched off.

The RFQ quote builder skill, added from the zip file Claude produced, keeps every quote consistent. If it doesn't appear after you create it, press Create Skill and click back again. An advisor running a stronger model than the agent, such as Fable 5.1, is optional for when it gets stuck, and this agent needs no workflows or sub-agents.

How Should The RFQ Agent Be Triggered?

Manually from the console while you're testing, then through the API once it's live, so each new email in the inbox triggers the agent with that email's data. We wouldn't put it on a schedule: keeping the time to quote down means checking the inbox very often, which costs money, and pulling the whole inbox each time gets messy.

The deployment runs in an environment (we recommend Anthropic's cloud environment) and needs the vault holding the tool credentials. Networking can be limited to the MCP servers, packages and hosts you allow, though we weren't worried about what this agent might do online.

For testing we kept the trigger on manual and the session budget off, with a short initial message such as "Please begin work". If a session runs through its budget, such as $10, the agent stops where it is and the work is lost, so we'd add one only once the agent is refined, by which point the risk is small anyway.

How Do You Test An RFQ Agent?

Send an RFQ to the inbox the agent reads, asking for products that are on your products board. Ours was a test request for a new office fit-out for 12 people, listing six products, sent from our own address. In real use it would come from a customer.

What If An RFQ Includes Products You Don't Stock?

We'd build the agent around the products you sell most to begin with, such as the 100 or 200 lines you sell constantly. For anything it can't find on the products board, one option is to have it flag the item as beyond what it can handle and pass it to a person.

Supplier sourcing can be automated too: when a product isn't on the board, an agent could find a supplier and email them for a price on the quantity requested. We'd make that a second agent, called by this one as a sub-agent, and it wasn't part of this build.

How Long Did The Agent Take & What Did It Cost?

About two minutes, at a cost of $1.37, for the six-item RFQ. We started it with Run now on the deployment and followed the session under View sessions.

It stopped almost straight away because the email came from an internal address, so we replied that it was a test and to carry on. The whole run, including that pause, took two minutes and 17 seconds.

Its summary reported a duplicate check (the quotes board was empty, so there was nothing to clash with), a new quote item with the company, contact name, email, phone, today's date and request details, and the status set to prepared. All six items matched the products board and were in stock for the quantities asked. It raised no review flags.

Did The AI Agent Get The Quote Right?

As far as we could tell, yes. The quotes board item had every detail filled in, from the quote number and total to the requested items and customer name, and its link opened the quote in the Quotes folder in Google Drive.

The quote looked almost identical to the example we gave Claude, with perhaps a few small differences. Checked against the RFQ email, all six requested products were on it. It's ready for a person to review and send, and the next step would be more testing, tweaking and edge cases.