AI for Wholesale: Where It Helps a Maker Sell to Shops

A ceramicist sells four hundred mugs a year through her own site and at two markets. A shop in another city emails: they like the work, do you do wholesale, what are your terms. She has never written a line sheet, has no idea what a minimum order should be, and is fairly sure that whatever number she says first will be the number she is stuck with.

This is the moment most small makers meet wholesale, and it is a genuinely different business from the one they have been running. Retail sells one item to a person who already wants it. Wholesale sells thirty items to a buyer who is calculating whether they can sell them on at a margin, and who will judge you on paperwork before they judge you on the product.

The question worth answering is narrow: where does a language model actually help with that, and where does it quietly cost you money.

Why does wholesale reward paperwork so heavily?

Because the buyer is not present. The conversation that used to happen at a trade show now happens through documents that have to answer every question in advance, and the seller who leaves a gap loses to the one who did not.

The scale of that shift is easy to underrate from the small end. Digital Commerce 360 reported on McKinsey's ninth annual B2B Pulse survey of 3,942 decision makers, which found 73 percent of buyers willing to spend over 50,000 dollars per order through self service or remote channels, up from 59 percent two years earlier. Thirty nine percent would go over 500,000 dollars and 20 percent over a million. Those are not figures about ceramics, and that is the point: the expectation of buying without a conversation is now normal across the whole of trade, and it arrives at your inbox in the form of a buyer who wants a document rather than a call.

What follows from that is unglamorous. The seller who converts is the one whose line sheet answers the six questions before they are asked.

What should never be generated?

Your wholesale price. This is the single firmest rule in the piece and the one most often broken, because asking a model what to charge feels like research and produces a confident number in four seconds.

Wholesale pricing is arithmetic on facts only you hold. Craftybase sets out the standard structure plainly: cost of goods times two gives the wholesale price, times four gives retail, so an item costing eight dollars to make sells at sixteen wholesale and thirty two retail. The multipliers are conventions that exist because retailers need roughly half the shelf price to survive. They are not a valuation of your work.

The failure mode is specific. A model asked for a wholesale price for handmade candles will return a plausible market figure, which bakes in somebody else's material costs, somebody else's batch size and somebody else's labour rate. If your true cost of goods is higher than theirs, you have just agreed to lose money on every unit, and you will not notice until the third order because wholesale losses arrive in volume.

The same logic governs minimum order quantities. Craftybase's guidance points at an opening order value in the 150 to 200 dollar range with a lower reorder threshold around 75 to 100, and unit minimums such as six per line. But the number that matters for you is the point where packing, shipping and the time spent managing the relationship stop eating the margin, and that calculation needs your packing time, not an average.

Where it genuinely earns its place

Three jobs, and they are all the sort of work that stops a maker from shipping.

The first is the line sheet itself, treated as a completeness problem. A buyer wants product name, image, wholesale price, recommended retail, case pack, minimum, lead time, materials, dimensions, care instructions and a code they can key into their own system. Assembling that for forty products is tedious in exactly the way models are useful: give it your product data and the required field list, and it will tell you which of the forty rows are missing which fields. That is a checking task, not a creative one, and it is the same discipline we described for getting a product catalogue into a consistent structure.

The second is the trade description, which is a different piece of writing from the retail one. Retail copy sells the feeling. Trade copy sells the sell through: who buys this, what it sits next to on a shelf, what the margin looks like, why it does not need explaining to a customer. Most makers write one description and use it for both, which reads as naive to a buyer. Rewriting forty retail descriptions into forty trade descriptions is a genuine use of generation, provided you supply the positioning. The method is the one we set out for writing product descriptions without producing forty identical paragraphs.

The third is response speed on enquiries. A trade enquiry that sits unanswered for four days usually converts to nothing, and small sellers miss them because they arrive during production. A drafted first reply that confirms receipt, attaches the line sheet, states the minimum and asks the two questions you always need to ask, ready for you to check and send in ninety seconds, is worth real money. Note the shape: the model drafts, you send.

What does a model do badly enough to cost you an order?

Lead times and stock commitments. This is where generated text becomes dangerous rather than merely unhelpful, because a trade buyer plans around the date you gave them and a missed date on a first order usually ends the relationship.

The mechanism is straightforward. A model asked to draft a quote will fill a lead time because the document has a lead time field, and it has no access to your kiln schedule, your supplier backlog or the fact that you are away for two weeks in August. It will write four to six weeks because that is what quotes say. If your real answer is nine weeks, you have promised something you cannot do, in writing, to a business customer.

Terms of sale carry the same risk in slower form. Generated trade terms will assert a returns position, a damages window and a payment arrangement that sound professional and may not be yours. Wholesale disputes are settled by the document, and the document said what the model wrote.

What about getting paid?

This is the part retail sellers are least prepared for, because retail customers pay before the goods leave. Wholesale customers frequently do not, and the gap between delivery and payment is where small makers run out of cash while technically doing well.

The legal floor is worth knowing before you negotiate rather than after. The UK government's guidance states that a longer period than 60 days may be agreed for business transactions provided it is fair to both businesses, which sets the reference point for what counts as normal. The European position is the same order of magnitude and has been the subject of a long running attempt to tighten it.

That attempt is worth a sentence because it explains conflicting advice you will find online. EY's summary records that the Commission proposed a single maximum payment period of 30 days on 12 September 2023, with automatic interest and flat rate compensation of 50, 100 or 150 euro depending on the invoice, and that Parliament amended it on 23 April 2024 to allow up to 60 days by express agreement and up to 120 for seasonal or slow moving goods. The file then stalled in Council. So the thirty day cap you may read about is a proposal that did not land, and the operative number remains sixty.

For a small seller the practical posture is to write your own terms rather than accept the buyer's, and to make the first order the strict one. Payment before dispatch or a deposit on the opening order is normal, is not insulting, and is far easier to ask for at the start than to introduce later. A model is useful for drafting the polite version of that sentence and useless for deciding whether to hold the line, which is a business judgement about how much you need the order.

The stock problem wholesale creates

Selling to shops changes your inventory maths in a way nobody warns you about. Retail demand arrives smoothly, a few units a day. Wholesale demand arrives in cliffs: nothing for three weeks, then an order for sixty.

That lumpiness breaks the intuitive approach of making a bit more than you sold last month. It also breaks most simple forecasting, because the signal you need is not your sales history but your buyers' reorder rhythm, and with four stockists you have four data points rather than four hundred.

The honest answer for a maker at this stage is that the useful tooling is not forecasting at all. It is knowing your production capacity per week, holding a buffer of your two best sellers, and being willing to quote a longer lead time on everything else. A model can help you write down the capacity calculation. It cannot conjure demand signal that does not exist, and a forecast built on four orders will be confidently wrong in a way that costs you materials.

The buyer is using a model too

Worth sitting with for a moment, because it changes what a good line sheet looks like. A buyer at a chain of four boutiques who receives eleven supplier approaches a week is not reading eleven line sheets. They are increasingly pasting them into something and asking which of these fits our price band, which have a minimum we can meet, which can deliver before November.

That process rewards machine readable completeness and punishes charm. A beautiful line sheet as a flattened image, with the prices as part of the artwork, answers none of those questions when it is pasted into a chat window. A plainer document where every field is text, and where the numbers are labelled, survives the summarisation and stays in the shortlist.

The same applies to the email. A three paragraph introduction about your studio and your materials is the part that gets compressed away. The line that says minimum order 180 pounds, lead time six weeks, wholesale 40 percent of retail, delivery from stock on twelve lines, is the part that survives. Put it in the body, not in an attachment, and put it near the top, remembering that the rules on a cold approach differ depending on whether the shop is a company or a sole trader.

None of this means writing for machines instead of people. The buyer still decides. It means the filter in front of the buyer now reads text rather than looking at a design, and the seller who noticed that gets read by a human more often. It is the trade version of the argument we made about making a product page legible to the assistants that summarise it.

Running a trade side on the same shop

Practically, the question becomes where the trade information lives. A separate trade page with a password, a request form that captures the buyer's business details, and a line sheet that downloads rather than needing to be emailed, removes most of the back and forth that kills small orders.

None of that is technically hard and all of it is usually postponed, because the maker is busy making. If you are already rebuilding or setting up the storefront, adding the trade side at the same time costs almost nothing extra, which is one of the reasons our ecommerce website builder treats a trade area as a normal part of a shop rather than an enterprise feature.

The other piece worth getting right early is the supplier side of your own business, because wholesale volumes mean bigger raw material orders and longer supplier commitments. The terms in those agreements are worth reading properly, and using a model to review supplier terms is one of the better uses of the technology precisely because it is reading rather than writing.

A first hundred days of wholesale, honestly

If the enquiry has just landed and you have no line sheet, the order of work is this. Calculate your true cost of goods including your own time at a rate you would accept from anyone else. Set wholesale at twice that and check whether the resulting retail price is something a shop can actually sell. If it is not, the problem is your production cost and no pricing cleverness will fix it.

Then set a minimum that makes the smallest order worth packing. Then write the terms: payment, lead time, damages, returns, and whether you offer exclusivity by area. Then build the line sheet and have a model check it field by field against the list. Then reply to the buyer.

Four of those five steps are yours alone. The fifth is where the tool belongs, and keeping that boundary clear is most of what separates a wholesale side that adds profit from one that adds work.

Retail forgives a vague answer. A trade buyer reads the vagueness as risk and orders from somebody who wrote it down.