Dental Lab Quality Control Checklist and Remake Log With AI

Build one short checklist for each restoration type, tied to the stage where each error can still be fixed cheaply, then log every remake with a single cause code from a fixed list. A monthly AI review of that log shows which causes repeat, which dentists or materials cluster around them, and where to tighten a stage.

The order matters. Most labs start with a long, generic quality form, and it gets ticked without looking. A checklist built from your own remakes is shorter and gets read, because every line exists for a reason you can name. Until the log has a few months of causes in it, treat any checklist as a draft.

Follow me on Instagram@sagnikteaches

Why a downloaded template does not reduce remakes

Search for a dental lab quality control checklist and you will find general forms: "verify margins", "check occlusion", "inspect finish". Nobody disagrees with those lines, and that is the problem. They are true for every case, so they never tell a technician what to look at on the case in front of them, and they never tell the owner what went wrong last month.

Connect on LinkedInSagnik Bhattacharya

A useful system has three parts that feed each other. Stage checks catch errors while the work is still on the bench. The remake log records what slipped through and why. The review reads the log and edits the stage checks. AI helps with the third part only: counting, grouping and summarising free-text notes that nobody has time to read across a month. It does not inspect a restoration, and it should never be the thing that approves one.

Subscribe on YouTube@sagnikteaches

If your case records are still scattered, fix that first with the case tracking template, because the log needs a case number to point at.

Stage checklists, one per restoration type

Keep each list to six to ten lines and attach each to a bench stage, not to the final inspection. By the final inspection a wrong die or a wrong shade is already baked in. Below are starting points; edit them against your own materials and workflow.

Crowns and bridges (zirconia, e.max, PFM)

• Intake: prescription states shade, material, margin type and any stump shade. If any is blank, the case waits for the dentist, because a guess at intake is how many shade remakes start.
• Model: die trimmed, margin visible all round under magnification, no air bubbles within 1 mm of the margin. Bubbles on a margin cannot be fixed later.
• Contacts: proximal contacts checked on the articulated model with thin shim stock or floss, set to your lab's written standard rather than by feel.
• Occlusion: opposing-tooth contact checked in closure and in lateral movement, with notes if the bite registration was poor.
• Shade: checked under the same light as the shade tab photo, with the photo open next to the restoration.
• Finish: surface texture matches adjacent teeth in the photo; margins sealed and smooth under magnification; no pits at the cervical edge.
• Pack: case number on the box, correct teeth numbers, spare die included if requested.

Removable dentures and partials

• Intake: impression type recorded; bite registration present; tooth mould and shade written, with the dentist's note on midline and smile line.
• Try-in stage: wax try-in photographed; midline, tooth position and vertical dimension signed off by the dentist before processing. Skipping the wax try-in to save a day risks a full remake if the set-up turns out to be wrong.
• Processing: flask and cure cycle recorded; no visible porosity once polished.
• Fit surface: no sharp flash or nodules on the tissue side; border thickness checked against the impression outline.
• Clasps and rests (partials): seated on the model without force; wire not distorted by finishing.

Implant restorations

• Intake: implant system, platform size, abutment type and torque value written on the prescription. Missing component details stall this stage, so check them before anything else.
• Model: analogue seating verified against the scan or impression; screw channel position noted.
• Passive fit: framework or crown seats on analogues with no rocking; one-screw test result recorded for multi-unit work.
• Access hole: position and angle checked against the opposing teeth so the screw is reachable.
• Parts check: every screw, abutment and driver in the box matches the prescription, counted by a second person.

Veneers and thin ceramics

• Intake: prepared tooth shades and photographs present; thickness requirements stated.
• Thickness: minimum thickness checked at the cervical and incisal ends against the material's stated limit.
• Translucency and colour: checked on the stump shade, not on a white background, because ceramic changes colour on what is under it.
• Handling: bagged flat and labelled fragile, with the try-in paste or glycerine note.

A short way to test your lists: take your last ten remakes and ask, for each, "which line would have caught this?" If no line would have, the list is missing something. If a line would have caught three, make it bolder and move it earlier.

Remake log fields that make patterns visible

A remake log fails when the cause is a free-text sentence. "Dentist changed mind", "didn't fit", "bad impression" are all different spellings of three causes and an AI will happily keep them separate. Use a drop-down or a fixed code list, and keep free text for one short note.

A starter list of cause codes, each one written so that two people would choose the same one:

1. Prescription incomplete or unclear at intake
2. Prescription changed by the dentist after work began
3. Impression or scan quality (voids, distortion, margin not captured)
4. Model or die error
5. Design error (contact, emergence profile, thickness)
6. Shade or colour mismatch
7. Fit or margin error at build
8. Occlusion error
9. Finish or surface defect
10. Material or equipment fault (batch, furnace, milling bur)
11. Damaged or lost in transit
12. Wrong or missing parts or components

Separate the "remake" from the "adjustment" and the "repair" in a column of its own. A chairside adjustment of a tight contact is a different failure from starting again, and mixing them inflates the rate and hides what is actually going wrong. Write the definition on one line at the top of the log and make everyone use it.

A month of data, walked through

An illustration: a six-technician lab that returned 412 cases in a month and opened 19 remakes. The rate is 19 divided by 412, or 4.6 per cent. The owner assumes the problem is spread evenly and "technicians having off days". The log says otherwise.

The shares are rounded, which is why they add up to 101 per cent. Thirteen of the 19 remakes (68 per cent) entered at intake: the six shade cases, the four incomplete prescriptions and the three impression or scan problems. Technician skill is not the main issue; the information arriving with the case is. The owner's change is cheap: a rule that a case with no stump shade or no shade photo is held and the dentist asked, plus a one-line intake check added to the crown list.

Put a price on it as well. Suppose each remake costs about $38 of materials and 1.5 technician hours at $30 an hour. That is $38 plus $45, or $83 a remake, and 19 remakes cost $1,577 in the month before counting delay and goodwill. If the intake rule prevented half of the shade and prescription remakes (5 of 10), the saving would be 5 times $83, or $415 a month. Those figures are illustrative; use your own wage and material costs.

The monthly pattern prompt

Export the month's log as plain text or a CSV with the columns above. Check that no patient names, dates of birth or photos are in it. If your plan or agreement with the AI provider does not cover patient information, the case-number-only log is the safe format to paste. Use a business plan where your provider says content is not used for training by default, or switch off training on a personal plan. Skip the thumbs-up and thumbs-down buttons on these chats too: at least one provider keeps a rated conversation as feedback and may use it for training even on a business plan.

You are helping the owner of a dental laboratory review remakes.
Below is this month's remake log (case numbers only, no patient names).
Columns: case number, date opened, dentist account code, restoration type,
material, stage where error entered, cause code, detected by, cost, note.

Tasks:
1. Count remakes by cause code and by stage where the error entered.
   Show counts and percentages in a table.
2. Report any dentist account, material or restoration type that appears
   in three or more remakes. Do not report patterns on fewer than three cases.
3. List which causes could have been caught by a stage check that does not
   exist yet. Quote the case numbers that support each claim.
4. Suggest at most three changes to the checklists, each tied to the cases
   that justify it.
5. Say clearly where the data is too thin to conclude anything.
Do not guess causes from the notes if the cause code is blank. List those cases.

[paste log here]

An abridged, illustrative reply for the month above (case numbers invented):

By cause: shade 6 (32%), prescription unclear 4 (21%), occlusion 3 (16%),
impression or scan 3 (16%), fit or margin 2 (11%), transit 1 (5%).
By stage: intake 13 of 19 (68%), design 3, build 2, pack 1.

Pattern (3+ cases): shade remakes where the prescription had no stump
shade - cases 2207, 2219, 2231.
Not a pattern: account D14 appears twice (2210, 2240) with different causes.

Suggested change: hold any crown case with no stump shade or shade photo
at intake (supported by 2207, 2219, 2231).
Too thin to conclude: occlusion (3 cases, three accounts, no shared material).

The reply names the shade-at-intake cluster, quotes the three cases with no stump shade and flags that two cases from one dentist account each had a different problem, so no pattern exists there. That last step matters. A model asked for patterns will sometimes invent one from two coincidences, so the "three or more" rule and the "quote the case numbers" instruction are doing real work. Open the quoted cases and confirm each claim before you change a process.

What you would fix in a typical first reply: it may merge "impression or scan quality" with "prescription unclear" because both mention the dentist; it may propose a change that targets a single case; and it may give percentages that do not add to 100 per cent. Ask it to recount, or do the counts in a spreadsheet and give the AI only the summary to interpret. If you are comfortable with formulas, a pivot table on the same log produces the counts reliably and the AI adds the reading.

Sign-off wording that technicians will honour

A sign-off line is a promise that a named person looked at a named thing. Vague sign-offs ("QC passed") invite a quick tick. Make each one specific:

• "Margins inspected under magnification, no open margin or bubble seen. Initials, date."
• "Contacts checked on the articulated model against the lab standard. Initials, date."
• "Shade checked against the dentist's photo. Initials, date."
• "Parts and screws counted against the prescription by a second person. Second initials, date."

Two rules keep this honest. First, the log is for finding where the process leaks, not for assigning blame; if technicians expect a telling-off they will log the cause as "transit" or "dentist change". Say so out loud at the start, and review the "stage owner" field only in one-to-one coaching. Second, a second pair of eyes checks the final box, not the person who built it. In a lab with one technician, make the second check the next morning against the written list.

Two smaller cases that show why the stage field matters

Take a partial denture that comes back from the clinic because a clasp is too tight. The dentist's note says "didn't fit", and a tired technician logs it as a fit error at build. When the case is opened, the master model was fine and the clasp seated without force on it; the impression had a drag that moved the teeth slightly. The honest entry is "impression or scan quality", stage "intake". Counted wrongly, it points the owner at the bench, when the fix is a quick call to the dentist's team about disinfection and tray handling. This is why a short second look at five entries a month is worth the ten minutes.

Now a before and after for one log line. Before: "Crown on 36 came back, colour wrong, redo." After: "Zirconia crown, shade B2 requested, stump shade absent, no photo supplied; try-in showed it too light against neighbours. Cause 6, stage intake, detected by dentist at try-in, cost $83." The second version costs ten seconds more to write, and it lets an AI count it, a manager audit it and a dentist's practice manager read it without argument. If your technicians resist the extra seconds, give each cause a code on a laminated card at the bench so that choosing one takes a glance.

A third habit worth adding is a "near miss" tick on the stage checklist. When a technician catches a problem on the bench that would have been a remake, they tick it and write the code. Near misses show which of your checks are working, something the remake log alone cannot tell you. The prompt above can take them as a second table and report which stage catches the most.

Four ways this goes wrong in practice

A prompt that finds nothing is often a log that records nothing. Common faults:

• The "other" bucket grows. If more than one remake in ten is "other", your codes do not match reality. Read those notes, add a code, and re-run last month.
• Technicians log causes to suit themselves. Spot check five entries a month against the physical case. Any mismatch is a training point.
• The AI over-reads small numbers. A lab with 80 cases a month and four remakes cannot show patterns monthly. Run the prompt quarterly, or accumulate twelve months in a single file.
• Dentists' changes are never logged. If the clinic changed the prescription, you still log the case, with the code "prescription changed after work began". It shows how often you absorb a cost that is not yours, which is a fact for a pricing conversation. See the price list tutorial for how to write a charge for it.

Using the log in a conversation with a dentist

The log also changes difficult conversations. Suppose one account shows four shade remakes in a quarter, each with no stump shade on the prescription. You can write: "Four of this quarter's remakes were shade-related, and in each case the prescription carried no stump shade or shade photo. Could your team add those to the prescription?" Ask an AI to draft the note from the cases, then read and edit it yourself, check tone, and keep it factual. Pair it with the approach in the dentist enquiries tutorial so remake conversations sound consistent.

Setup time and how to know it is working

Expect about half a day to write the stage lists and the cause codes, an hour to build the log as a sheet or inside your case system, and ten minutes a month to export and run the prompt, plus the time to act on what it says. No paid tool is needed. A spreadsheet or the case software you already use is enough.

Judge it after three months on three signs: the share of "other" is under one in ten; the stage where errors enter is moving later or disappearing (a lab that catches errors at the bench is working); and the remake rate, under a fixed definition, is falling or at least the cost per remake is. If the rate is not moving, read the cases rather than the summary, since a rule written but not followed leaves the numbers unchanged.

The checklists are never finished. Once a quarter, delete a line that has caught nothing in a year and add the one your latest remakes call for. A list that grows forever stops being read, and a list that changes with the data stays worth reading.

Questions labs ask about remake logs

What is a normal dental lab remake rate?

There is no single trusted figure, because labs define a remake differently and case mix changes the result. Measure your own for three months using one written definition, then compare month against month. The trend and the cause breakdown matter far more than comparing yourself with another lab.

Should the remake log include patient names?

No. Use your case number, the dentist account code and the restoration type. Anything you paste into an AI chat should carry no patient names, dates of birth or photos showing faces. The case number lets you find the full record in your own system when a pattern needs investigating.

How many remakes do I need before AI can spot a pattern?

Treat anything under three cases with the same cause as a coincidence, not a pattern. For a small lab, run the prompt quarterly instead of monthly so the counts are large enough to mean something, and read the cases behind any claim before acting.

Who should sign off a case before it ships?

The technician who finished the case signs the stage checks they performed, and a second person signs the final check. In a one-technician lab, do the final check the next morning with fresh eyes and tick it against the written checklist rather than from memory.

Further reads

• Writing Pharmacy SOPs With AI: A Step-by-Step Workflow — Another regulated workflow turned into written steps with AI.
• AI for Dental Labs: Turn Handwritten Rx Slips Into Cases — Catch prescription errors before they become remakes.
• Dental Lab Invoice Template: AI-Drafted Statements and Chasers — Keep credits for remakes clear on statements.
• Dental Lab Marketing: AI-Assisted Outreach to Local Dentists — Use a falling remake rate as a plain, honest selling point.
• Dental Equipment Maintenance Log: Service Dates and AI Reminders — Furnace and milling-unit drift is a hidden remake cause.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general dental laboratory quality-management practice; no vendor pricing used.