A retailer selling online and in store runs across a storefront, a marketplace or two, a point-of-sale system, a warehouse or fulfilment partner, a shipping app and the accounts. The customer service team sits in the middle, answering questions none of those systems answer on their own, while buyers and the finance team work through supplier price files, returns, stock counts and rebate claims in spreadsheets. We automate the predictable steps as fixed, tested software that applies your pricing, returns and matching rules the same way every time. AI is used only where something has to be read or drafted, such as a customer’s email or a supplier’s catalogue, and its output is checked before anything leaves the system.
These are patterns we see in retail and e-commerce businesses, not client case studies. Your process gets its own map in the assessment.
Wholesale and stockist orders that arrive by email
Many retail brands also sell to stockists, and stockists rarely use the wholesale portal. Orders come in as an email listing styles by name, a PDF purchase order from a department store, or last season’s order spreadsheet with the quantities changed. Someone keys each line into the inventory system and hopes the colour codes are right. AI reads the email or attachment into order lines and maps the stockist’s descriptions and their own codes to your SKUs. Each line is then checked against the stockist’s account, your wholesale price list and available stock before the order is written and a confirmation goes out. Lines the AI is unsure of go to whoever keys these orders today, and nothing is posted until they clear them.
Supplier price files loaded without retyping
Brands and suppliers send new price lists and range sheets in their own formats, with their own product codes, sometimes as a PDF catalogue. A merchandiser maps each file to your product catalogue, updates cost prices, works out retail prices and checks nothing has fallen below margin, then updates the website and the point-of-sale system. Files in odd formats, the PDF catalogue included, are read by AI into rows first. Every file then gets the same treatment, with a stored mapping for each supplier, validation on every row, your margin and rounding rules, and a flag on any cost rise that breaks your margin floor. New products and exceptions wait for the merchandiser’s approval before they go live on the site.
Supplier bills, rebates and expenses
A stock invoice should match the purchase order and the goods received at the warehouse. Often it does not. Short shipments, substituted sizes, a cost price that differs from the agreed price file and freight charges nobody expected all turn up on the bill. Capture tools read most invoices well, and we use them where they work. The part that stays manual is the exceptions. Matching rules pair each bill with its order and receipt within the tolerances you set, send mismatches to the buyer who placed the order and reconcile supplier statements. AI reads the invoices the capture tool cannot. This pays off when you buy from many suppliers.
Supplier agreements often carry volume rebates, marketing co-op contributions, promotional support and settlement discounts, each with its own terms, thresholds and claim deadline. The buyer who negotiated them knows roughly what is owed, and claims that are not made on time lapse. AI reads each agreement to pull out the terms, and your buyer checks those terms before any rule is set up. From then on, fixed software calculates entitlements from your sales and purchase data and keeps a claim schedule so every claim goes in on time with its supporting figures. Claims above the thresholds you choose are approved by the finance manager before lodgement.
Receipt capture and expense apps handle most of this well, and for most retailers we would tell you to use one and leave it there. What sometimes still runs by hand is the work around it. Store managers across several locations forget to upload receipts for petty cash and store cards, and platform, marketplace and advertising charges arrive as monthly statements with no receipt at all. A scheduled chase for missing receipts and rule-based coding for the recurring platform charges leave the bookkeeper only the exceptions to review. Unless those two gaps cost you real time each month, this is not where we would start.
Customer complaints and returns
The same question arrives by email, chat, social media and the marketplace inbox. To answer it, someone finds the order in the storefront, checks whether the warehouse picked it, opens the carrier’s tracking page and types a reply. The lookup across store, warehouse and carrier runs by rule, and when a parcel stalls the customer gets a notice before they ask. AI reads each enquiry to find the order and work out what is being asked, then drafts a reply from the recorded status only. Routine replies go out on templates you have approved, and anything the AI drafted outside a template is checked by your customer service team first. Lost parcels, damage and angry customers go to your customer service lead.
Complaints reach a retailer through email, chat, product reviews, social comments and marketplace messages, and marketplaces track how quickly sellers respond. Each channel is watched by a different person, if it is watched at all, and nobody can see the whole list. Every complaint lands in one log with a response deadline for its channel, and anything close to that deadline is escalated. AI drafts a response from the order history and your policies, and your customer service lead approves it before it is sent. The owner gets a regular view of what customers complain about and which products or suppliers keep appearing, so the fix can happen upstream.
A customer brings back a faulty blender or a pair of headphones, and the store refunds or replaces it under your returns policy. The claim to the manufacturer is a separate job, with its own form for each brand, a serial number, photos and a fault description, and the credit note may arrive weeks later with nobody matching it to the claim. The return authorisation is issued under your rules, each manufacturer’s claim form is filled in, and every claim is tracked until the credit arrives and is matched. AI reads the customer’s description and photos and drafts the fault field for the returns officer to check. Whether to refund, repair or replace stays with your returns team.
Stock, dispatch and store staff
After a stocktake or cycle count, the variance report lists lines where the shelf and the system disagree across stores and the warehouse. Somebody then works back through receipts that were never booked, transfers still in transit, returns not put back into stock and marketplace orders that sold the same unit twice. We run that investigation as fixed software, with no AI involved. It compares counts to the system, pulls every transaction behind each variance line and lists the likely causes, so the investigation starts with evidence. Your inventory manager reviews the findings and approves each adjustment before stock on hand changes.
Shipping apps already book carriers, print labels and send tracking emails for standard parcels, and if yours does that well, we leave it alone. The manual work sits at the edges. Bulky items go on pallet freight booked by phone, split shipments leave customers with half an order, proofs of delivery for pallet freight have to be chased, and damaged-in-transit claims need photos, invoices and forms for each carrier. For those edges we build rules that book pallet and oversize jobs through the carrier’s portal or email, link split shipments to one order, chase PODs and assemble each transit claim. Accepting a carrier’s claim offer stays with your operations manager.
Rostering and workforce software handles most of this, and if you already use it well, we would not build over it. The work that tends to stay manual is what happens before a request reaches that software. Leave requests arrive by text to a store manager, casual staff change their availability around exams, and managers copy it all in when they get time. One intake collects those requests, checks them against the roster rules you already have and passes approved changes into your rostering system. Approval stays with each store manager, and pay interpretation stays in your payroll software.
Store, marketplace and accounts kept in step
A new product is created in the storefront, listed again on each marketplace, set up in the point-of-sale system and mapped in the accounts. Every price change repeats the trip. Payouts from the storefront and marketplaces arrive net of fees, refunds and chargebacks, and someone reconciles them to orders by hand at month end. We treat this as one process and measure it that way. Nearly all of it runs as fixed software, with mappings between systems, a scheduled sync, a payout reconciliation and an exception queue for anything that does not agree. AI is used only to map free-text product notes to fields, and those mappings are reviewed before they are saved.
If one of these costs you real money every month, describe it in four answers.
Thinking about what a process actually costs you?
The assessment is small, fixed-price, and tells you the real number, whether or not you ever build.
Worth a 30-minute conversation.