Why every AI-drafted quote should be approved by a person

What can go wrong in an AI-drafted quote?

Most errors in an automated quote are not dramatic. They are the same small misreadings a new rep would make: the wrong thread size, an old part number the customer still uses, a quantity in boxes instead of units. The difference is that software can make the same mistake at 2 a.m. on fifty quotes in a row if nobody looks.

What did the Air Canada chatbot case show?

In Moffatt v. Air Canada (2024 BCCRT 149), decided by British Columbia's Civil Resolution Tribunal on February 14, 2024, a customer relied on the airline's website chatbot, which told him he could apply for a bereavement fare within 90 days of ticketing. The airline's own policy page said otherwise.

Air Canada argued it should not be liable for what the chatbot said. The tribunal called that ‘a remarkable submission’ and wrote that ‘it should be obvious to Air Canada that it is responsible for all the information on its website.’ The airline was ordered to pay $812.02 in damages, interest and fees.

The amount was small, but the principle is not. Customers will treat what your automated tools send as coming from your company. For a distributor, that means a quote drafted by AI is your quote. A person reviewing it before it leaves is the simplest way to make sure it says what you mean, and it gives you a clear record of who approved what if a customer ever questions a price.

This is a summary of a published decision, not legal advice.

Doesn't approval slow everything down?

Only if the approver has to rebuild the quote. When the draft is complete, with every line matched, priced and totalled, and with unclear lines clearly flagged, approval becomes a review, not a rewrite.

Speed still matters a great deal. A 2020 Paperless Parts survey found that 67% of part buyers expect a quote within 24 hours. A draft that is ready in minutes and approved within the hour meets that expectation with room to spare, without giving up control.

Worked example: review time vs. build time

An illustrative example with round numbers: a distributor receives 40 RFQs a day. Building each quote from scratch takes a rep about 20 minutes. Reviewing and approving a complete draft takes about 3 minutes.

Building from scratch: 40 × 20 = 800 minutes, or 13 hours 20 minutes a day. Reviewing drafts: 40 × 3 = 120 minutes, or 2 hours. Human approval costs the team about 2 hours a day in this example, and saves 680 minutes, or 11 hours 20 minutes, compared with doing everything by hand.

In other words, approval is not the expensive part. Re-keying is. Your own times will differ, so try them in the ROI calculator.

What should be flagged for extra attention?

A good approval flow does not treat every quote the same. It points the reviewer straight at the lines and quotes that deserve a closer look.

• Lines that are unclear, unmatched or missing a spec, flagged instead of guessed.
• Quotes above a dollar amount you set.
• First-time customers, where pricing and credit deserve a second look.
• Low-stock items, where the quoted lead time may not hold.
• Substitutions, where the system matched a different part than the one named.

How should approval work in practice?

1. The RFQ arrives by email and the AI reads each requested line.
2. The AI chooses which catalog item matches each line, and nothing more.
3. Prices, quantity breaks, customer pricing, totals and lead times are calculated in code from your price sheet.
4. Unclear lines, large quotes, new customers and low-stock items are flagged.
5. A rep reviews the draft, corrects any match, and approves it.
6. The quote goes out from your mailbox, and the draft, approval and send are logged.

How does Merlin handle approval?

In Merlin's Quote Desk, a human rep approves every quote before it goes out. The AI only picks which item matches a request line; prices, totals and lead times are calculated in code from your own price sheet, so the AI cannot invent a price.

When a rep picks a different part than the draft suggested, Merlin remembers it as a learned alias, so the same request is matched correctly next time. Every draft, approval and send is recorded in an audit log, and owners set per-teammate permissions; details are on our security page. Before going live, teams can run shadow mode for 1 to 2 weeks, where Merlin drafts on real RFQs but nothing can be sent. For the full picture, read the RFQ automation guide.