Inventory Match Agent
It checks each incoming order against the stock in three warehouses. If an item is short, it suggests a replacement before anyone starts packing.
Example business · Regional food-service distributor, 3 warehouses
PO #48213 → live stock
94% of lines matched without a human. Substitution texted to buyer at 6:02 AM, approved 6:04 AM.
94%
of orders matched with no one touching them
−61%
stock-outs on delivery
22 hrs
saved per week
Time to build
4 weeks
Typical build price
$6,000–$12,000
Running cost
$300–$500 a month
Pays for itself in
about 2 months
See it running
The console the team works in.What it looks like on a phone.
A recreation of the live dashboard, playing one real job from start to finish.The texts the agent sends and gets back, playing one real job from start to finish. Names redacted.
[Distributor] Orders
Text message · 6:02 AM
Morning! Your PO 48213 is matched. One item is short: Basil 1 lb (short 4 of 10).
Suggested swap
Thai basil 1 lb × 4
You ordered it in Aug and Sep · same price
Yes, swap
Done. All 5 lines are on the 7:30 truck. ✓
In plain English
Like a clerk who has memorized all 11,400 products and every customer’s habits.
Think of a very fast order clerk who never gets tired. Every order that arrives, by email, PDF or voicemail, gets read, checked against what is on the shelves in each warehouse and typed into the order system. When something is short, the clerk suggests a swap the customer has liked before and asks the buyer by text.
The problem
Before: orders came in by email, PDF and phone, in five different formats. Two staff typed each one into the order system by hand, checked stock themselves and called customers about swaps after the truck was already loaded.
Works inside the tools they already had
What the agent does, step by step
- 1
Reads every order
It reads orders from email, PDFs and voicemail transcripts and turns them into one clean format: item code, quantity and delivery time.
- 2
Matches against live stock
It finds the right item among 11,400 products in three warehouses, even when the name is slightly off. It also tracks batch numbers and expiry dates.
- 3
Suggests replacements
When an item is short, it picks substitutes based on what that customer has bought before, then texts the buyer for a yes.
- 4
Enters the order
Clean orders go straight into the order system. Tricky ones go to a Slack channel with a note explaining why.
What goes in, what comes out
It reads
- Order emails and PDF purchase orders
- Voicemail transcripts from the order line
- Live stock counts from 3 warehouses
- Each customer’s order history
It produces
- ✓Clean orders entered in the order system
- ✓Swap suggestions texted to the buyer
- ✓A Slack note for every order a person must see
- ✓A daily list of items running low
What it actually sends
A real-format example of one message from the agent. This is what the team sees, in the tool they already use.
Inventory Match Agent AGENT
5:42 AM
PO 30418 · [Customer] restaurant · delivery Thu 7:00 AM
Holding these 2 lines until the buyer answers. Truck loads at 6:30.
A normal day before
With the agent
5:30 AM: two clerks start retyping overnight orders
5:30 AM: overnight orders already entered
7:00 AM: shortages found while packing
5:45 AM: shortages flagged with a swap ready
Afternoon: calls to customers about wrong items
Afternoon: clerks handle new accounts and the odd exception
How we built it
Every build follows the same path: learn the job, connect the tools, run it next to your team, then switch it on.
- 1Week 1
Map the orders
We sat with the two order clerks, collected 300 real orders and listed every format and every mistake.
- 2Week 2
Connect and teach
Connected email, the stock sheet and QuickBooks. Taught it the product list and each customer’s past swaps.
- 3Week 3
Run side by side
The agent prepared every order while the clerks still did their own. We compared and fixed the misses.
- 4Week 4
Go live
Clean orders go straight in. Anything unclear goes to Slack with the reason.
Your people still decide
- ●Any swap over $200 or on a contract item
- ●New customers and credit holds
- ●Price changes
The agent asks before it acts on any of these. Every action it takes is logged, and you can switch it off at any time.
Builder’s note
The hard part was not reading the orders, it was the names. Customers write “lemmons”, “rom hearts” or an old item code. We matched on the customer’s own history first, which got most of these right, and sent anything under 90% sure to a person.
Phil Wood, founder, PWOOD GROUP
Example build based on a common workflow, not a named client. Figures show what the agent is designed to deliver.
Have a job like this? Book a free 30-minute call and we’ll tell you if an agent can do it.
Next: Claims Intake Agent →Give us 30 minutes. We’ll show you where the week goes.
A free video call. Tell us how your team spends a normal week. We’ll name the jobs an agent could take over, what that saves and roughly what it costs. If it isn’t worth it, we’ll say so, and you’ve spent nothing.
Example week