Wholesale and distribution automation

Buyers raise purchase orders when stock falls to its reorder point, then spend part of every week chasing suppliers for confirmations and delivery dates, often by email to a rep who replies from a phone. Those dates rarely make it back into the ERP, so customer service quotes lead times from memory. Raising the purchase orders, sending them and chasing until each is confirmed involves no judgement, so it runs from your reorder points on a set schedule. Reading the replies does. AI picks out confirmed dates, part shipments, substitutions and price changes, and the ERP is updated from what it finds. A changed price or a substituted product goes to the buyer rather than straight into the system.

Supplier price increases arrive as spreadsheets, PDF catalogues and portal downloads, each in its own layout and each with its own effective date. A purchasing or product officer maps the columns, updates cost prices in the ERP, recalculates sell prices and pushes the changes to the website, usually late the night before they apply. Where a file follows a known layout, fixed mappings and validation do the whole job and apply your margin rules to every line. AI is used only for files in odd formats, reading them into the same structure. Lines that break a rule, such as a cost jump beyond your tolerance or a product code nobody recognises, wait for the purchasing officer to approve.

Each import shipment brings a commercial invoice, a packing list, a bill of lading and sometimes a certificate of origin, and they seldom agree with each other or with your purchase order on the first pass. Someone in purchasing or logistics checks them, sorts out the discrepancies with the supplier and assembles the pack for your customs broker. Supplier documents arrive in every layout, so AI reads them into structured data. Fixed checks then compare quantities, product descriptions and declared prices against the purchase order and flag any mismatch, and the pack assembles in the format your broker asks for. Your logistics coordinator reviews the flagged differences and approves the pack before it goes.

Supplier rebate agreements pay on volume thresholds, product ranges, growth targets and marketing contributions, each with its own terms and claim period. The detail lives in a PDF the owner or purchasing manager negotiated and nobody else has read closely, so claims go in late or not at all. AI reads each agreement and extracts the terms, and the purchasing manager checks that extraction before any rule is built on it. From then on, purchases are tracked against each agreement, what you are owed is calculated, and each claim is prepared on schedule. The purchasing manager can also see when a threshold is close enough to be worth adjusting an order for.

Capturing supplier invoices is well served by packaged tools, and many distributors already use one. The work that stays manual sits after capture. An accounts payable officer matches each invoice to the purchase order and the goods receipt, finds that the supplier billed the old price or a full pallet when a part pallet arrived, and emails the buyer to ask what to do. That matching is rule work. Each invoice is checked against order and receipt within your price and quantity tolerances, approvals are routed, approvers who sit on them are chased, and supplier statements are reconciled. AI reads only the invoices that capture tools fail on. This layer earns its keep at high supplier volumes.

Carrier invoices arrive as long weekly statements, each line a consignment with its weight, zone, fuel levy and surcharges. Checking them against agreed rates is tedious, so most businesses pay them as they come and absorb overcharges such as a wrong zone, a residential surcharge on a business address, or the same consignment billed again. We build this without AI from start to finish. Each line is matched to the consignment you booked, rated against your carrier’s rate card and checked for duplicates. Differences are listed with the evidence attached, ready for the accounts team to dispute with the carrier.

An order ships in part, the rest goes on back order, a freight surcharge applies to one delivery, and the customer signs the docket for fewer cartons than were picked. Someone in accounts pieces this together before an invoice can go out, and invoices sit while they wait. Fixed software builds each invoice from the source records, meaning the dispatched quantities, the signed proof of delivery and the agreed freight terms, then checks it for completeness before release. AI has little to do here beyond reading a free-text note scrawled on a docket. Invoices can then go out when the goods leave, rather than when someone catches up.

Your accounting software already sends basic reminders. The slow part is cash application, and the follow-up that goes beyond a template. Trade customers pay in batches, with one transfer covering many invoices less a deduction for a short delivery, and the remittance arrives as a PDF or not at all. Fixed matching rules apply payments to invoices and log every step, while AI reads remittance advice and classifies debtor replies as a promise to pay, a dispute or a query about a missing proof of delivery. Reminders follow a cadence you set. Disputes and hardship go straight to the credit controller, who also approves anything off-template.

Monday morning in a distribution business starts with a shared inbox holding orders, delivery complaints, supplier invoices, credit applications and statements, all mixed together. Someone reads each message and forwards it to the right person, and a message that sits unread becomes a missed order. AI classifies each email and pulls out the details that matter, such as the customer account, order number or invoice number. Fixed routing rules then send it to the order desk, accounts payable, credit or the warehouse, or straight into another automated process. Anything the classifier is unsure of lands in a review queue, so nothing is lost for being ambiguous.

Customer service in a wholesaler spends a large part of the day telling customers where their order is. Each call means opening the ERP, then the warehouse system, then the carrier’s tracking page, and reading back whatever turns up. Status lookups and proactive notices need no judgement. They pull from the order, pick, dispatch and freight records, so the customer hears about a delay before they ring. When an enquiry arrives by email, AI reads it, finds the order and drafts a reply with the current status, which is checked against the records before it is sent. A back order with no firm date stays with customer service.

A new trade customer fills in a credit application, often a scanned PDF with a director’s guarantee and a list of trade references. The credit controller checks the business details, emails the referees, runs a credit report and recommends a limit to the owner or financial controller. The legwork is automated. The application is checked for completeness, the business registration is looked up, the credit bureau report is requested, and reference requests go out and are chased until answered, with everything gathered into one file. The credit limit and payment terms remain the credit controller’s decision, made on the full picture instead of a checklist finished in a hurry.

Once credit is approved, the account still has to exist. Someone creates the customer in the ERP, sets the pricing tier, delivery address, freight zone and payment terms, adds the contacts to the accounting system and the online ordering portal, and sends the welcome email with logins. Miss a step and the first order stalls at the order desk. Approved details from the credit file flow into each system in turn by set rules, with the same checks every time, and the account manager is told when the customer can order. A mismatch between systems goes to accounts to resolve.

Big retailers, builders and national buyers send their suppliers questionnaires on emissions, modern slavery, safety and quality, often through a portal, with each one worded a little differently. Large companies’ own climate reporting is why the emissions questions keep coming. The owner or operations manager usually answers them from scratch, hunting for last year’s answers and the evidence behind them. We keep an answer library and an evidence register, both under version control, and AI drafts each response from approved answers only. The owner or operations manager approves every submission before it goes. A question with no approved answer goes to whoever owns that topic.

Freight management platforms already handle booking, labels and tracking for many distributors, and where one is in place we build around it. The manual part is usually the gap between that platform and everything else. Dispatch staff key consignments into a carrier site, print labels, and later hunt for a signed proof of delivery when a customer disputes receipt. AI plays almost no part here. The carrier is booked from the dispatched order, the consignment and label are created, and the proof of delivery is collected when it comes back and attached to the invoice. Customers get a delivery notice with tracking, and a failed booking goes to the dispatch lead.

A customer rings to return a faulty pump. Customer service issues a return authorisation, the warehouse receives the item, and someone then has to lodge a warranty claim with the manufacturer on the manufacturer’s own form, with the serial number, fault description and proof of purchase. Claims that are never lodged are credit you never recover. Fixed rules issue the return authorisation, track the item through receipt and fill in each manufacturer’s claim form from your records. AI reads the customer’s fault description and turns it into the manufacturer’s fault categories. Customer service checks that reading before the claim is lodged and takes back any claim the manufacturer rejects.

A cycle count finds fewer units in the bin than the ERP says, and someone has to work out why. The answer might be an unposted receipt, a pick from the wrong location, a return that was never put away, or a branch transfer keyed at both ends. Chasing that by hand is slow, so small variances are often written off unexplained. None of the tracing needs AI. Counts are compared with system quantities as they come in, and the movement history behind each variance, covering receipts, picks, returns and transfers, is pulled automatically. Likely causes are listed against each line for the warehouse manager, who decides what to adjust and signs off the write-off.