AI-Supported Quoting: Reading Inquiries, Drafting Responses, Human Sign-Off

In most SMEs that sell configurable products or services, a quote starts with an inquiry that arrives as free text: a rambling email, a forwarded PDF, a spec sheet with half the fields empty, sometimes a photo of a part. Before anyone can price anything, a skilled person has to decode what the customer actually wants, map it to your catalog or capabilities, and notice what is missing. That decoding step, not the calculation, is where inquiries sit for days while the one person who can do it works through a backlog.

That decoding step is precisely what language models are good at. A model can read the inquiry, extract the requested items, quantities, materials, deadlines, and delivery terms into a structured form, flag what the customer forgot to specify, and draft the clarifying questions. What it cannot do reliably is know that this customer always underestimates tolerances, that this material has a supply problem this quarter, or that this price needs headroom for negotiation. The split is clean: AI handles interpretation and drafting, your experts keep judgment and pricing.

A working setup usually has four stages. Incoming inquiries land in one place, whether that is a shared mailbox or a form. The AI reads each one and produces a structured summary: what is being asked for, in what quantity, by when, with open questions listed explicitly. It then drafts either a quote based on your price rules and product data or a clarification email where information is missing. Finally, and non-negotiably, a person who understands the business reviews the draft, adjusts it, and sends it under their own name.

The important design decision is that the AI produces drafts inside your existing process, not a parallel system. If your quotes live in the ERP, the draft should land there as a quote record, not in a separate tool someone has to copy from. If your sales team works from a shared inbox, the summary and draft should appear attached to the original email thread. Every copy-paste step you leave in the workflow is a place where adoption quietly dies.

It is tempting to treat human review as a temporary training wheel to be removed once the system proves itself. For quoting, resist that framing. A quote is a legally meaningful commercial commitment, and the model will sometimes misread a unit, miss a revision in a long email thread, or price an item your production team cannot actually deliver on the requested date. These errors are individually rare and collectively certain, and a person who knows the business catches most of them in under a minute per quote.

The sign-off also protects something commercially valuable: the relationship. Mittelstand customers often buy from you because a specific person understands their situation. When that person reviews and signs every quote, the customer experience is unchanged, just faster. The honest pitch internally is not that AI replaces the quoting expert, it is that the expert stops being a bottleneck and starts spending their time on the judgment calls only they can make.

Measure three things from the start. First, time from inquiry received to quote sent, because response speed is often the reason you win or lose against a competitor who answered first. Second, the correction rate: how often the reviewer changes the draft materially, not cosmetically. A falling correction rate over the first months tells you the extraction rules and product data are improving. A stubbornly high one tells you where your product data or price rules are ambiguous, which is worth knowing anyway.

Third, watch what happens to the inquiries you used to ignore. Most SMEs quietly triage: small or vague inquiries wait or never get answered because the quoting capacity goes to the big ones. When drafting becomes cheap, answering everything becomes feasible, and some of those neglected inquiries turn out to be real business. That widened funnel is often worth more than the time saved on the quotes you were already sending, but you only see it if you count it.