Dashboard

How to Make a Price List With AI

AI cannot know your costs and will invent them with total confidence. Bring your own floor prices, let it do the packaging, and watch for the three lines that leak money.

Manuele Estivo
Manuele Estivo
Growth & SEO Lead
12 October 20261 min read

How to make a price list with AI, in one sentence: do the numbers yourself and let the model do the packaging. An assistant has no idea what your materials cost, how long a job takes you or what your market pays, and it will invent all three with total confidence. What it is genuinely good at is the part most people find hardest, which is deciding what to bundle, what to name things, where to put an anchor and which lines invite haggling.

So the method is to arrive with costs and leave with structure.

Work out your floor before you open a chat

Every line on a price list needs a number underneath it that you can defend. For a service, that is your time at a rate you can live on plus anything you buy in. For a product, it is unit cost plus the share of overhead that unit has to carry. Neither of those is a thing a model can estimate for you, and a plausible guess is worse than no guess, because it anchors you.

Write four columns for each item before you start: what it is, what it costs you, how long it takes, and the lowest price you would accept without resenting the work. That last column is the one that stops a price list becoming a trap, and it is the number you never show anyone.

If you are pricing project work rather than a catalogue, the breakdown approach in pricing a one-page website with AI is the better starting point, because it splits the job into blocks and asks which blocks AI actually shrinks.

What to hand the model

Give it your floor prices and your constraints, and ask only for packaging decisions. Keep the costs out of the prompt if you would rather not share them, and give it the floor price alone, since that is all the structure work needs.

text
You are helping me package a price list. Use only these facts.

Business: [what you do, who buys]
Items, with my minimum acceptable price:
1. [item] - [minimum]
2. [item] - [minimum]
3. [item] - [minimum]

Constraints:
- Currency [X]. Prices shown to [consumers / businesses].
- [Include / exclude] VAT in displayed prices.
- I will not discount below the minimums above.

Produce:
- Three tiers, with a name and one line on who each is for.
- Which items belong together and why, in a sentence each.
- Two places my current list invites haggling, and the rewording.
- Anything a customer would ask that the list does not answer.

Do not invent prices, costs, timings, discounts or guarantees.
If something is missing, write NEEDS INPUT and say what is missing.

The last two lines earn their place. A model asked for a price list will fill gaps rather than flag them, and the NEEDS INPUT convention turns an invented number into a question. Phrase the constraints positively too: Anthropic's prompting guidance notes that telling a model what to do beats telling it what not to do, so "use only the prices I gave you" does more work than a list of prohibitions, with the prohibitions as a backstop.

The three-tier structure, and when it is wrong

Three tiers is the default for a reason: it gives a customer a frame rather than a single number to accept or refuse, and the middle option does most of the selling. A worked example for a two-person cleaning service with floors of 60, 110 and 180:

Tier

Contains

Price

Role

Essential

Standard clean, two hours

65

Makes the next tier look reasonable

Standard

Standard clean plus kitchen and oven

125

The one most people buy

Full

Standard plus windows and a deep bathroom clean

195

Anchors the top, sells rarely

Notice that none of the three sits on its floor. The spread between the floor and the price is what absorbs a difficult job, and a tier priced at its floor has nowhere to go when the oven turns out to be worse than described.

Three tiers is wrong when your work genuinely varies per job, because publishing fixed tiers for variable work means either absorbing the bad jobs or renegotiating every time. In that case publish a starting price with what it includes, and quote individually. The drafting method for that is in using AI to write customer quotes.

A price list for goods sold to consumers is a regulated document in Great Britain, and a price list for services is not. AI does not know which one you are writing, and will format both identically.

The GOV.UK Price Marking Order guidance is explicit about scope: "The PMO covers goods, not services, and is limited to sales between traders and consumers." Where it does apply, the selling price is defined as "the final price to be paid for a unit of a product, or a given quantity of a product, including VAT and all other taxes", and an indication of selling price must be "unambiguous, easily identifiable, clearly legible, and displayed using a font which is clear and of reasonable size". Prices also have to be placed where consumers can see them without asking staff.

The practical reading for a small business: if you sell goods to consumers, the number on the list is the number including tax, and a footnote adding VAT later is not compliant. If you sell services, you have more freedom, and showing a tax-exclusive price to consumers is still a reliable way to generate arguments at the point of payment. Business buyers usually prefer tax-exclusive prices, which is a reason to keep separate lists rather than one hedged one.

Three places a list leaks money

First, the word "from". A price list that says "from 50" has told the customer the lowest number you will ever accept and invited them to aim at it. Either publish what the 50 version actually includes, or publish a tier instead.

Second, undefined inclusions. Any line a customer could reasonably interpret two ways will be interpreted the cheaper way, and you will do the work. "Includes a revision" is a negotiation; "includes one round of revisions, up to two changes per round" is a price. Ask the model specifically to find lines that could be read two ways, since this is the task it is best at, being a careful reader of ambiguous text with no stake in the outcome.

Third, a fixed price wrapped around a variable cost. Travel, materials bought at the going rate, delivery, or anything billed to you by the hour will move, and a tier that swallows it is a tier that quietly loses money in exactly the months the cost rises. Price those as a separate line at cost, or state a radius, a quantity or a cap in the inclusion. A model will not spot this for you unless you tell it which of your costs vary, so add that to the constraints and ask which tiers are exposed to it.

If you are considering where to put prices for a software product rather than a service, the variable-cost question changes the answer, and pricing an AI product with variable costs is the relevant version of this exercise. The general ground is covered in AI for small business.

Before you publish

  1. Check every number against your own floor column. Nothing on the list may sit below it.

  2. Read each line as a hostile customer and mark anything you could be argued into for free.

  3. Decide tax treatment once and apply it to the whole list.

  4. Put a review date in your calendar, because a price list nobody revisits becomes an old price list quietly.

FAQ

Can AI tell me what my competitors charge?

Only if it can read their published prices, and then it is a summary rather than an analysis. Anything it offers about unpublished pricing is a guess, and the figures sound exactly as confident as the real ones.

Should I show prices at all?

For standardised work, yes, because the main cost of hidden prices is the enquiries you answer from people who were never going to buy. For bespoke work, publish a starting point so people can self-select.

How do I handle a discount request?

Decide the answer before the list is published, not in the conversation. A fixed policy, such as a percentage for a year paid in advance, is a price. An ad hoc discount is a revision to your floor.

What about rounding?

Pick one convention and apply it everywhere. The inconsistency is more noticeable than the convention: a list of 65, 125 and 195 reads as considered, while 65, 124 and 200 reads as guessed.

How did this land?

About the author

Manuele Estivo
Manuele Estivo

Growth & SEO Lead

Manuele covers distribution: SEO, content strategy, and how AI-built products find their first thousand users. He tests everything he recommends.

Share

Get the next post in your inbox

One email a month. Product updates, engineering posts, and the best of Built with Swarmz.

I agree to receive emails about AI building tips and Swarmz product news. Unsubscribe any time.