Large Kuwaiti IT systems integrator · Digital platforms

A governed procurement platform replaces pricing by email and judgement, built in 20 days.

  • AI document reading
  • Microsoft 365
  • Exchange Online
  • Claude
  • Approval workflow
A distributor's reply read into quote lines, with a confidence label on every field and nothing saved until a person confirms
  • 20 daysfrom first commit to a working procurement desk
  • 1,012commits
  • 7default roles, from sales rep to general manager
  • 4procurement reports with Excel export
Before

Salespeople asked procurement for prices directly. Procurement emailed the distributors it knew, margin was set by judgement, and approvals happened in conversation.

Quote history lived in individual mailboxes and left when the person did. Turnaround couldn't be measured, distributors couldn't be compared, lost deals taught nothing, and margin decisions couldn't be audited.

What we built
Sourcing workspace
Suggested distributors, generated RFQ text, and chases, replies and prices recorded in one place.
AI-assisted reply capture
A pasted distributor email or spreadsheet is pre-filled into price lines with a confidence score per field. Nothing is saved until a coordinator confirms.
Margin-checked release
Pricing can't be released to sales until every quoted line has a recommended margin and sell price.
Shared-mailbox intake
A Microsoft 365 inbox is synced every 15 minutes, and a person turns each message into a request or a reply.
Reports and audit
Price comparison, turnaround, distributor responsiveness and price position, with Excel export, and an audit record for every change.
Sourcing queue listing open pricing requests and their status
Sourcing queue: every open pricing request in one place.
Suggested distributors to ask for a quote, drawn from the vendor map
Who to ask: distributors suggested from the vendor map.

Where AI works in the solution

reading distributor quotes

From a pasted reply to checked price lines

Coordinators paste a distributor's reply, from an email or a spreadsheet. Claude reads it into structured quote lines against a fixed schema, with a confidence label on every field. If the AI doesn't answer in time, a rule-based reader takes over, and nothing is saved until a person confirms.

Integration

Built around the shared mailbox the team already used

  1. existing systemDistributorsRFQs and replies by email
  2. existing systemMicrosoft 365 mailboxthe team's shared inbox
  3. we builtProcurement platformsourcing, release, reports, audit
  4. AI modelAnthropic Claudereply text to quote lines
SystemDirectionMethodWhat flows
Microsoft 365 · Exchange OnlinebothMicrosoft Graph · webhook and 15-minute syncIncoming requests and distributor replies; RFQs and notifications sent from the shared mailbox
Anthropic ClaudeoutSDKPasted reply text in, structured quote lines out
IdentitybothSDK · webhookSign-in, multi-factor authentication and user lifecycle
Microsoft ExceloutFile exportPriced sheets and procurement reports
next case studyExecutive project profitability analytics
Read it →