How to automate quotes from email without losing control
You can automate quotes from email safely if you split the job three ways: the AI only decides which catalog item each request line refers to, code calculates every price, total and lead time from your price sheet, and a rep approves every quote before it is sent. Anything unclear is flagged, not guessed. That keeps the speed of automation and the accountability of a human signature.
Why are teams nervous about automating email quotes?
Because a wrong quote is expensive. Quote a price too low and you either eat the margin or have an awkward call. Quote the wrong part and you lose credibility. Most people have also seen AI tools produce confident, wrong answers, so the idea of a model writing prices to customers is reasonably uncomfortable.
The answer is not to avoid automation. It is to design it so that the model never does the things it is bad at, like arithmetic and inventing facts, and always shows its uncertainty.
There is a quieter risk too: a tool that sends on its own trains the team to stop looking. Keeping a rep approving every quote, with flagged lines shown first, means attention goes where it is needed and nobody is surprised by what a customer received.
Who decides what? The three-way split
| Task | Who does it | Why |
|---|---|---|
| Read the email and attachments, split into request lines | AI | Free-form language is what language models handle well |
| Match each line to a catalog item | AI, with confidence check | Customers rarely use your part numbers |
| Unit price, quantity break, customer price | Code, from your price sheet | Prices must be exact and traceable |
| Line totals, subtotal, lead times | Code | Arithmetic should never be generated text |
| Unclear or unmatched lines | Rep (flagged) | Guessing is how errors reach customers |
| Approve and send | Rep | A person stays accountable for every quote |
Worked example: one email, five lines
Here is an illustrative RFQ from an existing customer to a fasteners and fluid power distributor. All prices below are invented for the example.
| Request line | Match | Price rule applied | Line total |
|---|---|---|---|
| 500 x 1/2-13 x 2 hex bolt, grade 8, zinc | Catalog hex bolt, gr. 8, zinc | 250–999 break: $0.36 each | 500 × $0.36 = $180.00 |
| 1,200 x 1/2 flat washers | Catalog flat washer 1/2 | 1,000+ break: $0.05 each | 1,200 × $0.05 = $60.00 |
| 40 x JIC 37° male adapter -8 | Catalog JIC adapter -8 | Customer-specific price: $8.75 | 40 × $8.75 = $350.00 |
| 10 x hose assembly, 1/2 in, 36 in | Catalog hose assembly | List price: $46.20 | 10 × $46.20 = $462.00 |
| "the usual O-rings" | No confident match | Flagged for rep | Not priced |
Illustrative. Subtotal of priced lines: $180.00 + $60.00 + $350.00 + $462.00 = $1,052.00, taxes extra. The fifth line waits for the rep.
What happens with the flagged line?
The rep opens the draft, sees four lines priced and one flagged with the reason ("no confident match"), checks the customer's history and picks the right O-ring kit. The quote goes out with all five lines after one quick decision instead of a full rebuild.
Because the rep chose a part for that wording, Merlin stores it as a learned alias for that customer. Next time "the usual O-rings" arrives from them, it matches automatically, and the rep still approves the quote.
Which guardrails should you set?
- Amount threshold: quotes above a value you choose always get flagged for a closer look.
- First-time customers: flagged, so someone checks terms and credit before quoting.
- Low-stock items: flagged, so you do not promise a lead time you cannot meet.
- Missing specs: a line without size, material or grade is flagged rather than matched to the nearest item.
- Permissions: the owner decides which teammates can approve and send.
- Audit log: every draft, approval and send is recorded with who did it and when.
What about language and data privacy?
In Québec, many customers write in French and others in English, sometimes in the same week. Merlin replies in the language the customer wrote in, and part numbers and descriptions come from your price sheet so they stay consistent.
Your email and pricing are sensitive. Merlin is hosted in Canada, your data is never used to train AI models, and the product was built with Québec's Law 25 in mind. The security page has the details. Mailbox connection is Gmail and Google Workspace today, with Microsoft 365 and Outlook coming soon.
How do you start?
Connect the mailbox where RFQs arrive, upload your price sheet, and run in shadow mode for one to two weeks so your team can compare drafts with the quotes they actually sent. Nothing can be sent during that period. See how the Quote Desk works, or read more about why human approval matters for AI quotes.
What does a rep's review actually look like?
Reviewing a draft should take minutes, not a rebuild. The rep opens the draft next to the original email and checks a short list, then approves, edits or rejects it. Everything they change is recorded in the audit log.
- Read the customer's email and the draft side by side, in the language the customer used.
- Check flagged lines first: each shows why it was flagged, such as no confident match, missing spec, low stock or first-time customer.
- Glance at matched lines: part number, quantity and the price rule applied (quantity break or customer price).
- Confirm lead times and terms, then approve. The quote and PDF go out from your mailbox.
- If you swapped a part, that choice is saved as a learned alias for next time.
Teams in fasteners and fluid power, where one RFQ can mix bolts, fittings and hose, often find most of the review time goes to the flagged lines. See our page on fasteners and fluid power distributors.
Frequently asked questions
Can the AI change a price?
No. In Merlin, prices, totals and lead times are calculated in code from your price sheet. The AI only selects which catalog item matches a request line, and a rep approves the quote.
What if the customer's email has no part numbers at all?
That is the normal case. The AI matches descriptions to your catalog. When it is not confident, or a spec is missing, the line is flagged for a rep instead of guessed.
Do web-form RFQs count as email?
Yes, if your website emails form submissions to you. Those messages arrive in the same mailbox and are drafted like any other RFQ.
Can quotes ever go out without approval?
No. A human rep approves every quote before it is sent.
Can I see why a line was flagged?
Yes. Each flagged line shows its reason, such as no confident match, a missing size or grade, low stock, a first-time customer or a quote above your amount threshold, so the rep knows exactly what to check.
