AI + semantic search · B2B procurement
Intelligent quoting engine: from a pasted free-text list to a formal quotation in seconds
"2 hammer drills 1/2, electrical tape x10 and a white helmet size M" — that is how a real customer writes, and that is how this system takes it: paste the list in free text and out comes a formal quotation with real market prices. No templates, no codes, no salesperson spending hours searching item by item.
Problem
Quoting a purchase list takes hours: interpreting what the customer wrote, finding each item, comparing prices and building the document.
Result
The customer pastes their list in natural language and receives a formal quotation with market prices in seconds.
The context
In B2B distribution, quoting is the silent bottleneck: the order arrives by email or chat, written in the customer’s language — quantities in words, half-remembered brands, approximate sizes — and someone has to interpret it, find each product, compare prices and build a presentable document. Hours per quote, while the customer compares against whoever answered first.
The hard part is not finding a product: it is understanding an order. "A desk and a chair with wheels" means an office chair, not a wheelchair. That class of ambiguity, obvious to a human, is where traditional search engines fail.
How it works
The flow has three AI stages. First, a language model parses the complete order — splits items, interprets quantities written in words, preserves brand and size, and uses the context of the whole order to resolve ambiguities. Second, each item is searched by meaning against a semantic index of hundreds of thousands of products with market prices.
Third — and this stage is what separates a demo from a trustworthy system — another model validates every match: is this THE product that was requested? Cooking oil is not motor oil, and it is not a deep fryer either. If there is a single valid candidate, it auto-selects the best price; if genuinely different products compete, the system asks with simple options; and if something was not found, it says so — it does not sneak in a substitute. The output is a formal quotation document: itemized table, taxes, terms, ready to print or send.
Engineering decisions
Decision 01
Ask only on real ambiguity
The same product in several stores is not ambiguity — pick the best price and move on. The system only interrupts when genuinely different products are at stake (brand, tier, type). Every unnecessary question is friction sending the customer back to their spreadsheet.
Decision 02
Honest outputs
If an item has no reliable match, the quotation says so and offers to rewrite or remove it. A quoting engine that pads with similar-looking substitutes generates returns and distrust — the cost shows up later, multiplied.
Decision 03
The LLM validates what embeddings bring close
Semantic search retrieves candidates close in meaning; a language model confirms which ones are actually the requested product, with a strict category rule and a flexible size rule: if there is no 1-liter presentation, the 900 ml one is a valid answer — a deep fryer never is.
What changes for the business
It is a new sales channel, not an upgrade to the existing one. In traditional e-commerce the customer browses the virtual aisles; here they hand over their entire shopping list and the system identifies in your stock exactly what they want, in their own jargon: "that white tape for joining threads," "that liquid for cleaning rust." Quoting stops being a chore and becomes an answer.
- From hours per quotation to seconds, formal document included.
- Answering first: the cheapest commercial advantage there is.
- Every quotation is recorded: price history and traceability.
- The sales team quotes volume without growing headcount.
Where else it applies
Distributors & wholesalers
Customer purchase lists turned into instant quotations.
Purchasing organizations
Comparing the market for every line of the requirement.
Industrial hardware
Orders written "the way the foreman talks."
Corporate procurement
Budgeting internal requirements without an email chain.
FAQ
Frequently asked questions about this case
Where do the prices come from?
From a market index built according to the industry: public catalogs, the business’s own price lists, or both. The index updates continuously and every price keeps its source.
What if the customer writes with typos or ambiguously?
The parser is built for real language: quantities in words, half-remembered brands, typos. And when ambiguity is genuine — different products that fit — the system asks with simple options instead of guessing.
Does the quotation come out in a presentable format?
Yes: a formal document with an itemized table, taxes, totals and terms, ready to print, save as PDF or send. Every quotation is recorded for follow-up.
Can it quote against our own catalog instead of the market?
Yes — the semantic index is built from whatever sources you define: your catalog, the market, or both compared. It is the same engine as the semantic search, pointed at the source that serves you.
Have a similar challenge?
Tell us your problem. We'll tell you how to solve it — and how long it takes.