Discuss a project

DL-039 · Case study

LLM product matching for Amazon Business sourcing

A B2B supplier matched quote lines to products by hand. In a paid research phase we built a prototype that scores Amazon Business matches for a person to approve.

Client
Name withheld
Location
US East
Industry
B2B procurement and MRO supply
Capabilities
01Audit03Build
01

Challenge

  • Quote requests arrive as free-text item lines, not product identifiers.
  • Buyers searched for each item by hand before quoting a price.
  • Automated ordering needed confidence the matched product was the right one.
02

Environment

  • Procurement team quoting industrial and office supplies to business customers.
  • Amazon Business API as the sourcing and ordering channel.
  • Buyers who must stay accountable for every quoted item.
03

Intervention

  • Used an LLM to turn each quote line into search keywords.
  • Scored candidate products against the request, with criteria and doubts shown.
  • Built a review screen where a buyer accepts, rejects or corrects matches.
  • Wrote the PRD and target architecture for search, scoring and ordering.
04

Outcome

  • A working prototype matching quote lines to Amazon Business products.
  • A PRD and architecture ready for a production build.
  • Every match carries a score and reasons a buyer can check.
Each quote line is turned into keywords by an LLM, searched in the Amazon Business API, scored against the request and shown to a buyer, who approves the match before ordering.candidatesnext phaseQUOTE LINESFree textLLM KEYWORDSSearch termsAMAZON BUSINESSProduct searchMATCH SCORINGScore / criteria / doubtsBUYER REVIEWApprove / correctORDERINGPlanned
  1. 01QUOTE LINESFree textLLM KEYWORDSSearch terms
  2. 02LLM KEYWORDSSearch termsAMAZON BUSINESSProduct search
  3. 03AMAZON BUSINESSProduct searchMATCH SCORINGScore / criteria / doubtscandidates
  4. 04QUOTE LINESFree textMATCH SCORINGScore / criteria / doubts
  5. 05MATCH SCORINGScore / criteria / doubtsBUYER REVIEWApprove / correct
  6. 06BUYER REVIEWApprove / correctORDERINGPlannednext phase
Fig. 01Quote line to approved matchEach quote line is turned into keywords by an LLM, searched in the Amazon Business API, scored against the request and shown to a buyer, who approves the match before ordering.Scale NTS

Stack and services.

Stack05

  • Amazon Business API
  • Large language models
  • Keyword generation
  • Match scoring
  • Human-in-the-loop review UI

Services used02

Tell us what has to work.

A senior engineer reads every brief. Describe the system, what it runs on today and what happens when it fails.

Discuss a project