Order Agent
A web app for Keystone Supply, a building-supplies distributor, that turns incoming purchase orders into clean draft orders. Orders arrive as pasted email text, PDF text, spreadsheet rows or chat messages, in many different layouts. The agent extracts the customer, ship-to and line items, then maps each customer's part numbers to Keystone SKUs. It checks prices and stock and flags anything it is unsure about with a reason instead of guessing. A rep reviews every draft beside the original text and approves it. The agent never submits an order itself.
Agent Overview
Why We Built This
Every incoming order was keyed in by hand. An inside-sales rep read the email, PDF, spreadsheet or chat message, worked out what the customer meant and translated the customer's own part numbers into Keystone SKUs. They then typed each line into a new order and checked prices and stock as they went. That took about 20 minutes and cost about $11 per order. Because every customer uses a different layout and their own part numbers, mistakes crept in: wrong SKUs, outdated prices, and quantities the warehouse could not fill. These were often found only after the order was confirmed. - About 20 minutes of manual re-keying on every order - Customer part numbers translated to SKUs from memory or scattered notes - Prices that differ from the price list, caught only after confirmation - Quantities above available stock, discovered when the warehouse cannot fill them - Orders handled differently depending on which rep picked them up - No record of who changed what on an order, or why
The Solution
The agent takes over the transcription and checking: it extracts each order into a draft whatever layout it arrives in. Customer part numbers are mapped to SKUs through a per-customer map that the manager maintains. Unit prices are checked against the price list, and anything beyond the adjustable threshold (2% by default) is flagged. Quantities are checked against stock. Every uncertainty goes to an Exceptions queue with the reason and a concrete suggested action. That covers an unknown part number, a price difference, a quantity above stock or a customer on credit hold. People keep every decision. Reps review the original text side by side with the draft, can amend, swap, remove or add notes to line items, and approve each order. The agent never submits one. The agent learns nothing silently. A corrected mapping enters the map only when a rep clicks Save to map. Managers edit the part-number map and see the Value tracker and audit log. Every approval and edit is recorded.
How It Works
Takes in an order pasted as email text, PDF text, spreadsheet rows or a chat message, in any layout
Extracts the customer, ship-to address and each line item: part number, quantity, unit price and delivery date
Maps each customer part number to a Keystone SKU using that customer's saved map
Checks each unit price against the price list and flags any difference beyond the threshold
Checks each quantity against available stock and flags anything above it
Flags unknown part numbers, credit holds and anything it is unsure about, with the reason and a suggested action, and leaves those fields blank instead of guessing
Shows the draft beside the original text so the rep can amend, swap or remove lines and add notes
Saves a corrected mapping to the customer's map only when the rep clicks Save to map
Records every approval and edit in the audit log
Never submits an order: a rep approves each one
ROI & Impact
- Replaces a manual step of about 20 minutes and $11 per order with a reviewed draft, so reps fix flagged fields instead of typing whole orders - The Value tracker shows minutes and cost per order against that baseline, the share of orders approved without edits, pricing errors caught and rep hours returned - Price differences beyond 2% and above-stock quantities are caught before the order is confirmed, not after - Mappings and exception wording are manager-editable, and every change is audited, so the process stays consistent and reviewable instead of depending on individual reps' habits
Getting Started
Press "Test the agent" above and sign in with the demo inside-sales rep account
Open Incoming orders: sample orders in messy formats are already loaded, with statuses new, drafted, approved and exception
Open a drafted order to see the original text on the left and the draft on the right, with uncertain fields highlighted and the reason shown
Fix a highlighted field, such as a part number the agent could not map, and click Save to map so the next order from that customer maps it automatically
Approve the order, then sign in as the demo sales manager to see the change in the audit log and the order counted on the Value tracker
Visit Exceptions to see orders flagged for price, stock, unknown part number or credit hold, each with a suggested action
Related Agents
KYC Onboarding Case Preparer
A web app that prepares KYC onboarding cases for a bank. It assembles each customer’s details and documents, cross-checks them, flags mismatched, missing or expired items, rates risk by transparent rules, and drafts the summary and missing-items request. People make every approval decision.
Hotel Guest Line Agent
Hotel Guest Line Agent is a voice assistant that answers a hotel's in-house guest line for everything that isn't an emergency. It sounds like an unhurried, well-briefed front-desk host. Guests can order room service, request housekeeping or maintenance, book a table or a spa slot, arrange a wake-up call, ask about late checkout, and get answers about breakfast times, the gym, parking or what's worth walking to nearby. Every request lands on a live staff screen.
Arabic-first Employee HR Helpdesk
An AI HR assistant that answers employees' everyday HR questions instantly, in Arabic or English. It explains leave balances, payslips, benefits and working-hours rules, and drafts HR letters on request, using only the company's own HR records and policy handbook. Employees get answers in seconds instead of raising a ticket, and HR teams get their time back for the work that needs a person.
Ready to Start Your Project?
Get in touch with our team to discuss how we can help bring your AI-driven product vision to life.