Procurement Agent
Compares quotes properly and keeps a record of who actually delivers.
What is procurement agent?
A procurement agent receives purchase requests, routes them for the approvals your policy requires, collects and compares quotes on genuinely comparable terms, and records what each supplier delivered against what they promised.
Buying looks simple until you look at the paperwork. A request arrives by email or WhatsApp, somebody chases three quotes, the cheapest is chosen, and nobody notices the cheapest excluded delivery.
Six months later the same supplier is chosen again for the same reason, because the delivery problem was never written down anywhere a buyer would see it.
An agent routes the request properly, normalises quotes so the comparison is genuine, and keeps a running record of supplier performance so the last three orders inform the next one.
Why supplier selection stays with people
Duara builds procurement agents from Nairobi for organisations buying often enough that requests, approvals and quote comparison have become an administrative burden rather than a commercial activity.
We are deliberate about not letting an agent select a supplier. Choosing who you buy from involves relationships, credit terms and judgement about reliability that no scoring model captures honestly, and pretending otherwise produces defensible looking decisions that are quietly worse.
Who this is for
- Requests arrive by email and WhatsApp
- Quotes are compared on price alone
- Nobody records which suppliers deliver late
- Approvals happen informally or not at all
Industries we serve
- Construction and contracting
- Manufacturing
- Hospitality groups
- Healthcare facilities
- NGO operations
- County government suppliers
Problems this solves
These describe buying that is functioning but leaving money and evidence on the table.
- Quotes compared on headline price. The cheapest number wins even when it excludes delivery, warranty or a payment term the others include.
- Supplier performance never recorded. Nobody writes down who delivered late, so the same supplier is selected again for the same reason.
- Requests arriving informally. A purchase request in a WhatsApp message has no approval trail and no record that it happened.
- Approvals applied inconsistently. Policy says two signatures above a threshold and urgency regularly overrides that without anybody deciding to.
What we deliver
Requisitions captured in one place, approval routing that follows your own policy, quotes normalised so comparison is real, and supplier delivery tracked against what was promised.
Purchase requests captured in one place
Approval routing that follows your policy
Quotes normalised onto comparable terms
Supplier delivery tracked against promises
A record of every approval and who gave it
Reporting on spend by category and supplier
Features
- Like for like comparison. Quotes are normalised for delivery, warranty and payment terms, because a headline price that excludes delivery is not the lower price.
- Supplier history attached. Past delivery performance appears next to each quote, so the buyer sees reliability alongside cost.
- Policy driven approval. Routing follows the thresholds you have written down, which usually reveals how often they were being bypassed.
- No supplier selection. The agent presents and compares. Choosing who you buy from involves judgement no scoring model captures honestly.
- Spend visibility. What you buy, from whom, and at what rate, which is frequently the first time anybody has seen it aggregated.
Benefits
- Better decisions from the same quotes. Normalised comparison changes which quote wins more often than buyers expect.
- Poor suppliers stop being reselected. Written delivery history prevents the repeat that memory allows.
- Approval policy actually applied. Routing enforces the thresholds you already agreed rather than relying on people to remember them.
- Spend becomes negotiable. Seeing aggregated volume with one supplier is what makes a rate conversation possible.
Technologies we use
- Requisition capture. A single intake for purchase requests, replacing email and messaging threads.
- Approval workflow. Routing by value, category and requester according to your written policy.
- Quote normalisation. Adjusting quotes onto comparable terms for delivery, warranty, tax treatment and payment.
- Supplier performance tracking. Recording promised against actual delivery dates and quantities on every order.
- Spend analysis. Aggregating purchases by category and supplier so volume and rates become visible.
Normalising quotes is the technically dull part and where the value is. Most quote comparison fails on inclusions rather than on arithmetic.
How the process works, step by step
A procurement agent usually takes six to eight weeks, since approval policy often needs writing down before it can be encoded.
-
Step 1
Free buying review
We look at how requests arrive today, who approves what, and whether supplier performance is recorded anywhere.
-
Step 2
Policy written down
Approval thresholds are documented, which frequently reveals that the written policy and the practice differ.
-
Step 3
Requisition intake
Requests move to a single channel so there is a record before anything else changes.
-
Step 4
Quote normalisation
Comparison rules are agreed with your buyers, since what counts as comparable depends on what you buy.
-
Step 5
Performance tracking
Delivery against promise begins accumulating, and becomes useful after roughly a quarter of orders.
Quality assurance and how we work
The commitments attached to a system touching purchasing decisions.
- The agent never selects a supplier. It compares and presents. Choosing who you buy from involves relationships and credit judgement that no scoring model captures honestly.
- Comparison rules are agreed with your buyers. What counts as comparable depends on what you buy, so normalisation rules are set with the people who buy rather than imposed.
- Every approval is recorded. Who approved what, at what value and when, kept as a trail rather than living in a mailbox.
- Supplier scores are descriptive, not predictive. We record what happened on past orders. We do not claim to predict future reliability from it.
- We report policy bypasses. Where urgency routinely overrides approval thresholds, that is reported rather than quietly accommodated.
Frequently asked questions about procurement agent
Does it choose the supplier?
No. It compares and presents with supplier history attached. Choosing involves relationships and credit judgement that no scoring model captures honestly.
Can it place orders directly?
Once approved, yes, within limits you set. The approval itself stays with the people your policy names.
What does normalising quotes mean?
Adjusting them onto comparable terms for delivery, warranty, tax and payment, because a lower headline price that excludes delivery is not lower.
How does it track supplier reliability?
It records promised against actual delivery on every order, which becomes useful after roughly a quarter of trading.
Does it predict which suppliers will fail?
No. It records what happened. Presenting past performance as prediction would overstate what the data supports.
Will it enforce our approval policy?
Yes, once the policy is written down, which is often the point at which organisations discover the written version and the practice differ.
What if urgent purchases bypass approval?
That is reported rather than accommodated silently, because a bypass nobody sees is how thresholds stop meaning anything.
How long before it is handling requisitions?
Six to eight weeks, with the approval policy work usually taking longer than the software.
Can it handle tenders?
It supports quote collection and comparison. Formal tender processes have procedural requirements we scope separately.
Does it integrate with our accounting system?
Yes, so approved orders and receipts flow through rather than being rekeyed.
Can it flag price increases?
Yes. A supplier quoting materially above their own history is flagged for the buyer to question.
What about suppliers we are contracted to?
Contract terms are applied so quotes are checked against agreed rates rather than compared as if the contract did not exist.
Does it help with negotiation?
Indirectly. Seeing aggregated volume with one supplier is usually what makes a rate conversation possible in the first place.
Can requests come from several sites?
Yes, and consolidating requests across sites frequently reveals buying the same item at different prices.
What if a requester wants a specific supplier?
They can say so and the reason is recorded. Preference is legitimate and hiding it is what causes problems.
Does it handle import purchases?
Yes, and duty and clearing costs are part of normalisation, since they change comparison materially.
Will our buyers resist it?
Some do initially. It works better when framed as removing quote arithmetic rather than as oversight of their judgement.
Can it report on spend by category?
Yes, and for many clients this is the first aggregated view of what they buy and from whom.
Do our supplier prices stay private?
Yes. Pricing and terms stay in your systems and are never used to train public AI models or shared with other clients.
What does requisition handling cost to run?
Requisition volume and integration count determine it, quoted against your own buying figures before anything is agreed.
Glossary of terms
Procurement vocabulary in plain terms.
- Requisition
- An internal request to buy something, before it becomes an order to a supplier.
- Purchase order
- The formal instruction to a supplier to deliver goods or services at agreed terms.
- Three quote rule
- A policy requiring several quotes before purchase above a value threshold.
- Total cost
- Price plus delivery, duties and other inclusions, which is what makes quotes genuinely comparable.
- Approval threshold
- The value above which additional sign off is required before buying.
- Supplier performance
- The record of what a supplier actually delivered against what they committed to.
Related services in AI Agents
Talk to us about procurement agent
Take your last five purchases above a hundred thousand shillings and check whether the quotes compared included the same things. They usually did not.
Start your project