Run one finished job through the app before you pay, and check five things: the files that go in and come out, whether its part sizes and sheet count match your hand sums, the full monthly cost for your team, what you can export, and who owns your drawings and data. If any one fails, do not pay yet.
Joinery here means the workshop side of carpentry: doors, windows, stairs, fitted furniture and kitchens made to order. An AI app for that work usually does one of three jobs: it prices a job from drawings and schedules, it turns parts into a cut list and a sheet layout, or it drafts shop drawings. I have left vendor prices out on purpose. They change, they vary by plan, and a checklist should still work next quarter. Everything below is a test you run yourself.
Decide which job the app must do first
Judge each job separately, because a tool can be good at one and thin on another. Write down which of the three you are buying for, what it must produce, and how you would know it was right.
| Job | What it must produce | How you will know it is right |
|---|---|---|
| Quote from drawings and schedules | Priced lines you can send, traceable to rates and supplier prices | Total within a few per cent of what you really charged on a finished job |
| Cut list and sheet layout | Part sizes, quantities, grain, edging and a sheet count | Matches your hand sums, down to individual part sizes |
| Shop drawings | Dimensioned views a client or fitter can work from | Dimensions add up, reveals and hardware follow your standards |
Before you test anything specialist, run your messiest document through a general assistant such as ChatGPT or Claude. It sets a free baseline. If a general assistant already reads your door schedule correctly, a specialist app has to beat that on your own files, not just match it.
Data in: can it read your real files?
Vendors demonstrate with tidy files. Your inbox holds scanned schedules, drawings with a stamp over the dimensions and supplier price lists in three layouts. Put these checks to the app with files from your last three jobs.
- Every format you actually receive. List the formats on your last ten jobs: PDF, Excel, photos, CAD or model exports. Ask which the app reads natively and which need converting first. A conversion step you do for every job is a cost.
- Your largest file. Check page and size limits against your biggest drawing set. As a benchmark, Anthropic's help page says Claude accepts PDFs up to 1,000 pages but analyses drawings and other visual content only in PDFs of 100 pages or fewer, reading larger ones as text. It does not list DXF among its file types. An app built on a general-purpose model can inherit limits like these, so ask for the vendor's.
- Phone measurements and photos. If the app takes site photos or phone scans, remember that Apple describes its own Measure app's readings as approximate, and best on well-defined objects between 0.5 and 3 metres away. Joinery tolerances are tighter than approximate. Treat these as a first look and measure the opening properly.
- Supplier price lists. Give it a list in the awkward layout, with pack sizes and trade discounts. Check that it reads the unit, not only the number.
Whatever the app reads, make it show its working. The failure to watch for is the silently dropped row. Try this with a general assistant first, so you know what the problem looks like.
Here is a door schedule as a PDF. For each row list the door reference, width,
height, thickness, quantity and finish. Then state how many rows you read and
the total number of doors. Write "unclear" for anything you cannot read.A realistic (illustrative) reply starts with "Rows read: 13 of 14" and "Total doors: 31". Your schedule has 14 rows and 33 doors, because the row that straddles a page break was skipped. The fix is the habit, not the tool: always ask for a row count and a total, compare them with your own, and ask the tool to re-read the missing row. An app that gives you no way to reconcile its reading against the source is asking for trust you have not yet had a reason to give.
Data out: what comes back, and where can it go?
Outputs matter as much as inputs, because the app has to feed the rest of your week: the saw or CNC software, the invoice, the purchase order. Test the round trip.
| Check | Why it matters | How to test it in an afternoon |
|---|---|---|
| Quote leaves as an editable file | You will always want to change a line before it goes out | Export to Excel or an editable document and change three lines by hand |
| Cut list leaves in a format your saw or optimiser reads | Re-typing parts is where errors enter | Export a cut list and open it in the tool your workshop actually uses |
| Drawings leave as PDF and an editable file | Clients want PDF; your fitter or designer may want to edit | Send one to a colleague and ask them to change a dimension |
| Prices and parts reach accounting without re-keying | Double entry costs time and breeds mismatches | Import one quote's lines into your invoicing tool |
| Everything is exportable in one go | It is your exit route (see ownership, below) | Find the export button and run it on the trial job |
Here is a before and after from a hypothetical six-person workshop. Before: a 46-line quote is typed into a costing spreadsheet, then re-typed into the invoice, about 40 minutes of keying. After: the same quote exports to a CSV file that the invoicing tool imports in minutes. The time saved is not the point. The point is that the second path has one entry of each number, so there is one place for a typing error to occur, not two.
Drawing accuracy: the hand check that catches most errors
Choose a small job you know completely and work it out on paper before the app sees it. Here is one to copy: a run of four full-height cupboards, each 600 mm wide, 2,100 mm high and 560 mm deep, built in 18 mm board, with one full-height door each. Sheets are 2,440 by 1,220 mm (8 by 4 feet), with the grain running along the long side and a 3 mm saw cut. Backs, plinths and hardware are left out to keep the sum short. The parts, worked by hand:
| Part | Size in mm | Quantity | Total area in m² |
|---|---|---|---|
| Sides | 2,100 x 560 | 8 | 9.41 |
| Top and bottom panels | 564 x 560 | 8 | 2.53 |
| Shelves (three per cupboard) | 564 x 540 | 12 | 3.65 |
| Doors | 2,096 x 596 | 4 | 5.00 |
| Total | 32 | 20.59 |
The width of the horizontal panels is the width between the sides: 600 minus two 18 mm boards gives 564 mm. That is a classic place for a plausible error. An app that copies the cupboard width onto the tops and bottoms prints 600 mm, and the cut list still looks tidy. The total area barely moves (about 0.16 m² more here), so a sheet count would not reveal it. Only a part-by-part check will.
Now the sheet count. One sheet is 2.44 x 1.22 = 2.98 m², and 20.59 divided by 2.98 is 6.9, so the area floor is 7 sheets. Nobody can do better than that, and nobody reaches it, because the parts are the wrong shape. Twelve of them (eight sides and four doors) are over 2,000 mm long and at least 560 mm wide, so only two fit across a 1,220 mm sheet. Twelve long parts need six sheets, and each leaves a strip about 340 mm wide. Every remaining part is at least 540 mm in both directions, so none fit in the strips. The other 20 parts fit eight to a sheet (four along the length, two across), which takes three more sheets. That gives 6 + 3 = 9 sheets. Mixing does not help: a sheet carrying one long part has room for only four short parts beside it, and any mix of two-long, one-long-plus-four-short and eight-short sheets still comes to nine. The yield is 20.59 divided by 26.79 m² of board, about 77%.
Edging gives a second figure to compare. Taking all four edges of each door and the front edge of each side, top, bottom and shelf, the run is 21.54 m for the doors, 16.80 m for the sides, 4.51 m for the tops and bottoms and 6.77 m for the shelves: 49.62 m in total.
| What the app reports | Hand answer | How to read a mismatch |
|---|---|---|
| Number of parts | 32 | Any other number means a part is missing or duplicated |
| Total part area | 20.59 m² | A bigger figure usually means a wrong panel width or an allowance added |
| Sheets | 9 | 7 or 8 is impossible for this shape. 10 or more means waste, so ask to see the layout and check the grain and rotation settings |
| Edging | 49.62 m | A lower figure usually means an edge was left out |
| Top and bottom size | 564 x 560 | 600 x 560 means the thickness of the sides was ignored |
Also look for settings the app should let you control: saw kerf, the trim taken off each factory edge, and whether parts may be rotated. An app that hides them is answering a different question from the one your saw asks. For shop drawings, apply the same discipline: the dimension strings must add up to the overall size, the door must sit in its opening with your usual reveals, and the thickness shown must match the board you use. The shop drawings tutorial lists what AI drafts well and what you must check, and building a cutting list and material order from a drawing walks through that routine step by step.
The trial job: how to run it and mark it
Run the trial like an order, not a demo. Take the finished job, start a stopwatch, and do the real sequence: load the files, produce the quote, produce the cut list, produce a drawing, export everything. Stop only when you would be willing to send the outputs to a customer or a saw operator. The stopwatch matters, because the useful number is not how fast the app is. It is how long the whole job takes once you have checked its work.
That gives a simple sum. A kitchen quote that takes you three hours today might take 40 minutes to run in the app and 50 to check: 1.5 hours, a saving of 1.5 hours per quote. At six quotes a month that is nine hours, and at a stand-in rate of $45 an hour it is worth $405 a month. If the app costs $240 a month in full, the net gain is $165. If checking takes two hours instead of 50 minutes, the saving falls to 20 minutes a quote, worth $90 a month at six quotes, and the app loses money.
Record the result in a table you can show to a partner or an accountant. A filled-in example (illustrative) for two apps on the same job:
| Measure | App A | App B |
|---|---|---|
| Files opened without conversion | All but one DXF | All |
| Parts matching hand sums | 32 of 32 | 24 of 32 (tops and bottoms sized 600) |
| Sheets reported | 9 | 8 |
| Edging reported | 49.6 m | 44.0 m |
| Quote total against what you charged | 3% under | 11% under |
| Total time including checking | 1 h 40 min | 2 h 55 min |
App B may look quicker on its sales page, but it fails the part sizes, reports an impossible eight sheets, and costs more of your time once you correct it. App A needs one conversion step but passes. That is the whole value of the exercise: it replaces opinion with a table.
Monthly cost: count every line
The headline price is rarely the bill. Ask for each of these in writing before you decide.
- Who counts as a paid user: everyone with a login, or only office users?
- Storage included on your plan, against the size of your drawing archive.
- Any usage allowance for AI features, and the price beyond it.
- Modules sold separately: cut lists, shop drawings, CNC export, reports.
- Whether exports or integrations need a higher plan.
- Set-up or onboarding fees, and whether training is included.
- Minimum term, notice to cancel and how renewal prices are set.
- Tax, which is added to many published prices.
To see how far the headline can mislead, take two hypothetical plans for a six-person workshop with two people in the office and four in the workshop. The prices are invented for the sum.
| Line | Plan A | Plan B |
|---|---|---|
| Headline price | $60 per office user | $45 per registered user |
| Who is charged | Two office users | All six people |
| Users | $120 | $270 |
| Storage | 10 GB included (enough) | 5 GB included; a 14 GB archive needs another 10 GB at $25 |
| Export and integrations | Included | Higher tier, $40 extra |
| Monthly total | $120 | $335 |
| One-off set-up | None | $300, or $25 a month over a year |
Plan B's headline is $15 a user cheaper and its real bill is nearly three times higher. The sum also shows why storage deserves its own line. Drawings and photos fill space faster than quotes do.
File ownership and leaving: your drawings, your data
This is the check people skip, because it feels unlikely to matter until it does. Read the terms and the privacy policy, not the sales page, and look for these items.
- Who owns what you upload and what the app produces. A plain statement that you own your content and outputs. Be aware that OpenAI's and Anthropic's terms, for example, assign you whatever rights they hold in outputs, "if any", which is not a guarantee that no one else has a claim. Keep your own edits and your sign-off on anything that leaves the workshop.
- Training. Whether your files, or a client's drawings, are used to train the vendor's models or its AI provider's. Business terms often differ from consumer terms, so check the plan you would buy.
- Self-service export. A button, in open formats, covering quotes, cut lists, drawings and price lists. "Export on request" means you are waiting on the vendor at the worst moment.
- After you cancel. How long your data is kept, how it is deleted and whether you can get a final copy.
- If the vendor closes. Clockwise, an AI calendar app, deleted its users' data when it shut down on 27 March 2026. Any supplier can go the same way.
The practical habit is a monthly export saved to your own storage, whatever the vendor promises. Also send one short email before you upload a client's drawings, and keep the answer.
Subject: Questions before we start a trial
1. Who owns the quotes, cut lists and drawings we create in the app?
2. Are our files used to train your models or your AI provider's?
3. Is there a self-service export of everything, and in which formats?
4. How long do you keep our data after we cancel, and how is it deleted?
5. What is the full monthly cost for [number] office users and [number] workshop users?
Thanks, [your name]A vendor that answers these five questions plainly is easier to trust than one that points to a long terms page. The wider habit of asking before you buy is covered in what to do before you buy any AI tool, and how to tell whether an AI tool is a thin wrapper helps you judge how much of the product is the vendor's own work.
Reading a vendor's claims without being fooled
Four claims turn up often enough to be worth a sentence each.
"Hours saved." Hours-saved figures are usually estimates: tasks counted, multiplied by a minutes-per-task the vendor chose. Ask what the minutes figure was, and replace it with your own timing from the trial.
"Accuracy." A claim of accuracy on the vendor's test drawings says nothing about yours. Give it the hand-check job and report your own figure.
"AI draft, you approve." This is a good sign, not a limitation. It means the tool flags what it is unsure about and leaves every price to you. Be wary of one that presents suggestions as finished.
"Works with your catalogue." Setting up a catalogue of your products, parts and labour is real work, usually done by you. Ask how long it takes and whether the vendor does it, and rebuild one of your past quotes from the catalogue as a test. If the totals do not match yours, the set-up is not finished.
A one-page scorecard for the joinery app trial
Print this and fill it in during the trial. Items marked "must pass" decide the purchase; the rest break ties.
| Check | Must pass? | Pass mark |
|---|---|---|
| Opens the files from your last three jobs | Yes | All, or conversion under 10 minutes a job |
| Part sizes on the hand-check job | Yes | Every part correct |
| Sheet count and edging | Yes | At or above the true minimum, and within one sheet of it |
| Quote total on a finished job | Yes | Within about 5% of what you charged (my pass mark), every line traceable |
| Outputs leave in editable formats | Yes | Nothing re-keyed |
| Full monthly cost in writing | Yes | Covers users, storage, usage, modules and tax |
| Self-service export and ownership terms | Yes | Both stated plainly in the terms |
| Time including checking beats your current time | Yes | At least one hour saved per quote (my pass mark), or a clear net gain at your volume |
| Settings for kerf, trim, grain and rotation | No | Visible and adjustable |
| Support answers a real question within a day | No | Test it with one |
If any must-pass line fails, ask the vendor to fix it in the trial or walk away. If everything passes but you have only tested one job, run a second, awkward one before you pay. A wardrobe with a sloping ceiling will tell you more than another straight run. The cabinet shop software tutorial and cutting list software against AI are useful companions when you compare categories rather than single apps.
Joinery AI app questions before you pay
Can ChatGPT or Claude replace a joinery AI app?
For drafting a quote or reading a door schedule, a general assistant is a fair baseline to test first. For exact part sizes, sheet counts and edging, check every figure by hand, because long tables can lose rows and arithmetic can slip. A specialist app has to beat that baseline on your own files to earn a subscription.
How long should a trial of a joinery AI app be?
Long enough to run one full finished job through quoting, cut list, drawing and export, and then a second, awkward job. If a trial is shorter than a week, ask for an extension in writing before you start, so the clock does not run out while you are still waiting on a supplier price list.
What file types should a joinery AI app accept?
Whichever you actually receive: PDF drawings and schedules, supplier price lists in Excel or PDF, photos and the CAD or model files your designers send. Test with your messiest real files rather than the vendor's samples, and check any page or size limit, because a tool that fails on your largest drawing set fails when it matters.
Who owns the drawings an AI joinery app produces?
Read the terms, not the sales page. Look for a plain statement that you own your content and outputs, a self-service export in open formats, a stated retention period after you cancel, and a clear line on whether your files train the vendor's or its AI provider's models. If any of those is missing, ask in writing before you upload a client's drawings.
Further reads
- AI Tools for Shop Drawings: Checking Fabrication Drawings Against the Spec — Checking fabrication drawings against the specification.
- Best AI App for Carpenters: Measuring, Quoting and Invoicing Compared — Measuring, quoting and invoicing apps compared.
- Cabinet Design Software: What AI Adds for a Small Cabinet Shop — Compare cabinet design software by price band, CNC output and cut-list accuracy, and see what AI renders and drafting really add for a small shop.
- Countertop Fabricator Software: Templating, Slabs and Where AI Helps — A vendor question list for templating, slab layout, quotes and cut files, with a scorecard, a yield sum and where ChatGPT helps with quote wording.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Anthropic's help page on uploading files to Claude and its consumer terms (read 29 October 2026); OpenAI's terms on output ownership; Apple's Measure support page; the hand sums in this tutorial are worked from the stated part sizes. No vendor prices are quoted.