Concept buildA worked example of how we would build this for a typical business. Not client work: no client, no client data, no achieved results.
AI Agents
Concept buildA concept build for the purchasing team of a UAE trading and distribution company that chases overseas suppliers by hand for order confirmations, ship dates and shipping documents. Four agents split the work: a planner picks the orders that need chasing, a researcher gathers the facts, a writer drafts one email per supplier and a reviewer checks every figure against the sources. A buyer approves each email before it goes, so nothing reaches a supplier on its own.
The brief
A typical trading company in the UAE buys from suppliers several time zones away and resells to retailers and contractors across the region. A small team of buyers looks after a long list of open purchase orders, and each one needs the same chasing: has the supplier confirmed it, when will it ship, where are the packing list and the commercial invoice, has the booking moved. Buyers do it from their own inboxes when they have a moment, so follow-ups are uneven, the expected dates in the ERP go stale, and sales promises customers delivery dates the supplier changed a week ago. The work is repetitive enough to automate and risky enough that nobody wants software writing to suppliers unchecked: one wrong date, or a sentence that sounds like a commitment on price, costs more than the time it saves.
Whether an order needs a follow-up is a rule the buyers set and can read. The models group orders, gather facts and write, so the number of emails each morning is predictable and never decided by a model.
Text in a supplier’s reply is extracted into fields and never followed as an instruction. A reply that asks for a new recipient, a different bank account or changed terms is flagged for a person instead of answered.
Each order’s run has a ceiling on steps, tool calls and model spend. A run that reaches it stops and appears in the queue with its trace, rather than retrying quietly in the background.
Drafts follow each buyer’s own templates and sign-off, so suppliers keep hearing from the person they already deal with, and the approving buyer’s name is on every email that goes out.
Explainer
Explainer · 7 slides · about 48 s
Concept buildConcept build: Supplier Follow-up Agents for a Trading and Distribution Company
Concept build, explained in 7 slides: a concept build for the purchasing team of a UAE trading and distribution company that chases overseas suppliers by hand for order confirmations, ship dates and shipping documents.
Slide 1 of 7
01 · AI agents
Supplier Follow-up Agents for a Trading and Distribution Company
Concept build: Supplier Follow-up Agents for a Trading and Distribution Company.
02 · The problem
What a typical trading and distribution company is up against
A typical trading company in the UAE buys from suppliers several time zones away and resells to retailers and contractors…
03 · The approach
How we would build it, in 5 steps
We would build it in 5 steps, from write the follow-up rules with the buyers to replay past orders, then a supervised pilot.
04 · Architecture
How the pieces connect
How the parts connect, from the first input to the person who reviews the result.
05 · Demo · sample run
One supplier, three open orders, one approval
A sample run, step by step, including where a person steps in.
06 · Typical targets
What we would measure it against
Typical targets, each with the basis it would be measured against.
07 · Next step
Scope a build like this for your business
Book a strategy call to scope a build like this.
Architecture
Architecture of the Supplier Follow-up Agents for a Trading and Distribution Company concept: ERP open orders → Planner agent (open orders); Planner agent → Researcher tools (one task per supplier); Purchasing mailbox → Researcher tools (thread history); Forwarder booking → Researcher tools (booking status); Researcher tools → Writer agent (facts + sources); Writer agent → Reviewer agent (draft); Reviewer agent → Buyer approval (checked draft); Buyer approval → Approval send gate (approved email); Approval send gate → Follow-up tracker (sent + logged); Researcher tools → Finance team (bank-detail stop).
Diagram with 11 parts: ERP open orders (source), Purchasing mailbox (source), Forwarder booking (source), Researcher tools (processing), Approval send gate (processing), Planner agent (model), Writer agent (model), Reviewer agent (model), Follow-up tracker (system of record), Buyer approval (human), Finance team (human).
Connections:
Sample run
Planner selects three ordersRoutine
Three open orders with one supplier: no confirmed ship date five days after issue.
Researcher gathers the factsRoutine
Order lines, the last reply in the thread and the forwarder booking, each with a source.
Writer drafts one emailRoutine
One email for the three orders, asking for ship dates and the packing list.
Reviewer returns the draftNeeds care
The draft says 12 May; the order says 21 May. Rewritten, then passed.
Buyer approves the emailA person takes over
The buyer edits one line and approves; it is sent in the supplier’s thread.
A reply mentions new bank detailsStopped
The run stops and finance is alerted; the agents do not answer it.
Build plan
We go through a month of supplier threads with the buyers and write down when an order needs chasing: no confirmation a set number of days after issue, a ship date coming up with no documents, a booking that moved. Those rules, not a model, decide which orders get a follow-up, and the buyers own the thresholds.
The planner reads open orders from the ERP and groups them by supplier. The researcher gets read-only tools for the order lines, the supplier’s email thread and the forwarder’s booking status. The writer can only create a draft in the shared mailbox. The reviewer has no tools at all: it compares the draft with the researcher’s facts and passes or returns it.
The reviewer’s checklist is written with the buyers: every order number, date and quantity in a draft must appear in the facts, the email must not commit to prices, payment terms or penalties, and it must go to the contact the ERP holds for that supplier. A reply that mentions changed bank details or asks for payment stops the run and goes to finance.
Drafts land in one queue each morning with the facts and the reviewer’s result beside them. The buyer edits, approves or rejects; only approved drafts are sent, from the shared purchasing mailbox and in the supplier’s existing thread. Rejections are kept with the buyer’s reason and become test cases for the next change.
Before anything is sent, a few months of past orders run through the agents and their drafts are compared with what the buyers actually wrote. A pilot with two buyers and a handful of suppliers then runs for three to four weeks, and more suppliers join only when the review results hold.
Typical targets
Typical target
50–70% less buyer time spent chasing suppliers
Basis: Chasing time logged by the buyers for two weeks before the pilot, compared with time spent reviewing drafts during it
Typical target
A draft ready the same morning for every order past its confirmation window
Basis: Design property: the planner’s rules run on every open order daily; checked as the share of due orders with a draft in the pilot
Typical target
No email reaches a supplier without a buyer’s approval
Basis: Design property: the send tool accepts only drafts that carry an approval record
Typical target
Every date and order number in a sent email traced to a source
Basis: Design property: the reviewer returns any draft with a figure missing from the facts; audited on a sample of sent emails
Your project
Every action the agent may take, with its read or write scope and which ones need a person’s approval first.
Each run records the plan, every tool call with its inputs and outputs, and the approval decisions.
Real examples from your backlog scored before every prompt, tool or model change goes live.
Step and cost ceilings per run, so an agent that loops or meets bad data stops and asks instead of carrying on.
Where this fits
FAQ
No. This is a concept build: a worked example of how we would set up follow-up agents for a typical UAE trading and distribution company. The orders, suppliers and emails in the sample run are sample data, and the time-saving range is a typical target that a pilot would measure against your own baseline.
Because the roles need different permissions. The researcher may read the ERP and the mailbox, the writer may only create drafts and the reviewer has no tools, so a mistake in one role cannot reach a supplier by itself. Smaller roles are also easier to test one at a time. The cost is more model calls per order, which is why we split the work only where the permissions differ.
Three layers. The reviewer returns any draft whose order numbers, dates or quantities are not in the researcher’s facts; the buyer reads the draft beside those facts before approving; and the send tool refuses anything without an approval record. Each run also has step and spend limits, so an agent that loops stops and flags the order instead.
Not in the first version. When a supplier replies with a new ship date, the researcher records it in the follow-up tracker with the email it came from, and the buyer updates the ERP. Writing back to the ERP is a second phase, with its own narrowly scoped API key, once the pilot shows the extracted dates are reliable.
The buyers set an escalation ladder in the rules, for example a second follow-up after a few days and then an alert to the buyer and to the sales owner whose customer order depends on that shipment. The agents never threaten penalties or change terms; a supplier who stays silent becomes a phone call for a person.
Worked examples
Concept builds are worked examples, not client work.
Concept buildAI Agents
A concept build showing how we would build an AI agent that qualifies property leads on WhatsApp and website chat at any hour. It would ask about budget, area and timing, score each lead against the agency’s own rubric and book viewings for qualified buyers, so agents spend their time with people who are ready to view.
Concept buildAI Agents
A concept build for a US online marketplace where sellers upload products with thin titles and half-empty attributes. An agent reads each new listing and the supplier’s spec sheet, fills category attributes and a description from that data only, validates the result against the category schema and sends it to a merchandiser for approval. Nothing publishes without a person.
Concept buildAI Agents
A concept build for a research agent that prepares a cited company brief before a sales call, so reps start from checked facts, not a pile of browser tabs. It plans its own research, works only through allow-listed tools, links every line to a source, and drafts a CRM note that a rep approves before anything is saved.
Next step
Book a free 30-minute strategy call. Bring one process you want to automate and we will talk through scope, integrations and a sensible first step.
Free 30-minute call · no commitment
