Should AI Be Allowed to Send Your Quotes? An Approval Matrix
AI belongs in drafting your quotes long before it belongs in sending them without a human checking the number first — the risk isn't the drafting step, it's the moment a wrong price becomes a commitment your business has to honor. The fix isn't "AI or no AI," it's separating quoting into four distinct steps and deciding, deliberately, which ones a system can own outright and which ones still need a person to say yes.
Why Is "Should AI Send Quotes" the Wrong Question?
Framed as one yes-or-no decision, this question is unanswerable, because quoting isn't one step. It's drafting, price calculation, discount approval, and customer delivery — four distinct decisions with four very different risk levels. Treating them as one decision means either blocking useful automation everywhere, or allowing risky automation everywhere. Neither is right.
What Does the Approval Matrix Actually Look Like?
| Step | Can AI Own This Alone? |
|---|---|
| Drafting the quote document/language | Yes — low risk, easily reviewed before anything is final |
| Price calculation from your rate card | Yes, if the rate card and rules are current and version-controlled |
| Discount or exception approval | No — this is where margin actually gets lost; keep a human gate |
| Sending the quote to the customer | Depends entirely on whether the previous three steps were verified |
Where Does This Actually Go Wrong in Practice?
The realistic failure isn't an AI system inventing a number from nothing. It's an AI system correctly calculating from a price list that's one version out of date, or applying a standard discount to a customer who shouldn't get one, and then sending that quote before anyone reviews it. The AI didn't "hallucinate" — it did exactly what it was told, using stale or wrong inputs, and nobody caught it because the send step had no gate.
What Belongs in the Discount and Exception Gate?
Any quote that deviates from your standard rate card — a loyalty discount, a rush surcharge waiver, a bundled-service price — should require a named person's explicit approval before it goes out, logged with a timestamp. This isn't about distrust of automation; it's about keeping a record of who approved what, which matters the first time a customer disputes a price six months later.
How Do You Build This Without Slowing Down Every Quote?
Most quotes are standard: no discount, no exception. Those can move through drafting and sending with minimal friction once the rate card is verified current. The gate only needs to trigger for quotes that deviate from standard pricing — which, in most businesses, is a minority of quotes, not all of them. That's the design goal: fast for the routine case, gated for the risky one. Our Agentic AI builds are scoped exactly this way, step by step, not as one all-or-nothing automation.
What Should You Ask Before Automating Any Part of Quoting?
Who currently owns pricing exceptions, and would they actually see a flagged quote before it sends? Is your rate card a single source of truth, or does it live in three places that can drift out of sync? If a wrong quote went out tomorrow, how would you find out — a customer complaint, or a system alert? If the honest answer to that last one is "a customer complaint," that's the gap to close before adding any automation at all.