An AI purchasing agent that does the legwork. The buyer decides.
EM Polar's buyers work out what to order from the ERP, look for suppliers, compare quotes, negotiate prices and chase delivery dates. We built an AI purchasing agent that does this work continuously and leaves every decision to a person.
Case study – EM Polar- ERPconnection to the company's ERP – read directly, nothing written back
- 7parts of the buying work the agent takes on
- 100%of numbers calculated and checked by an algorithm
- 0actions sent without a person's approval
The problem
What to buy is in the ERP. Working it out is not.
Every purchase starts with a sum nobody writes down: what production needs, minus what is in stock, minus what is already on its way. Then a supplier, an enquiry, a quote in a PDF, a price to negotiate and a delivery date to chase.
When it is urgent the delivery date decides; otherwise the price. Either way, it is hours of copying before anyone makes a decision, and a better supplier or a better price is rarely looked for.
See it work
What to buy, from whom, at what price.
The agent works the requirement out of the ERP, writes to the suppliers, reads what comes back and negotiates. The buyer approves the order.
What we built
Seven things the agent does.
Everything that can be computed is computed and checked by an algorithm. The language model only does what needs language – writing to suppliers and reading their replies – and every action waits for a person.
- 01
Live stock tracking
Stock, open orders and what production needs, read from the ERP in real time, so the net requirement is always current.
- 02
New suppliers
Beyond the suppliers the buyers already know, the agent finds and contacts new potential suppliers for each item.
- 03
Negotiation
Counter-offers worked out with game-theory models: what to ask for, when to push and when to accept.
- 04
The best price, all the time
Prices are tracked continuously, not only when an order is due, so a better offer is caught when it appears.
- 05
Capacity limits
The company's own capacity limits are factored into what to buy, how much and when.
- 06
A person approves every step
Every enquiry, counter-offer and order goes to a buyer for approval before it is sent.
- 07
Numbers checked by algorithm
Every quantity, price and metric is calculated and verified by an algorithm, never estimated by the language model.
The rules it works by
The company's way of buying, kept.
The agent fits how EM Polar already buys rather than asking the buyers to change.
- 01
Read only
The agent reads the ERP. It does not write into it.
- 02
Date or price
When it is urgent the delivery date decides; otherwise the price.
- 03
Certificates noted
Item groups that need material certificates are marked as such.
- 04
Approval stays
The approval rule in the ERP and the owner's sign-off stay exactly as they are.
- 05
Their data stays theirs
Nothing is passed on to third parties.
How it works
From a requirement to an order a buyer signs.
The agent runs in a secure cloud and reaches the ERP over the company's VPN.
- 01
Requirement
Demand, stock and open orders come live from the ERP.
- 02
Sourcing
Known and new suppliers are asked, with the items and dates needed.
- 03
Negotiation
Quotes are read, PDFs included, compared and negotiated.
- 04
Order
A buyer approves the proposal; the owner signs it off in the ERP as before.
The result
The buyers decide. The agent does the legwork.
The requirement, the search for suppliers, the enquiries, the negotiation and the reading of quotes are done before a buyer opens the order. What is left is the decision: which supplier, which date, which price.
Nothing is sent without a person's approval, and nothing is written into the ERP.
The same approach works for
- Purchasing from your ERP – SAP, ABRA or another
- Finding and negotiating with new suppliers
- Reading supplier quotes and order confirmations from PDF and e-mail
- Chasing delivery dates against the production plan
Contact
A similar problem on your desk? Let's look at it together.
Thirty minutes with a founder: your process, your data, and whether a system like this one would pay for itself in your plant.
Or reach us directly
- Matyáš ProkopiusCo-founder
- Adam BulkaCo-founder