Ingredient safety conclusions, use levels and conditions, toxicology endpoints — organized by ingredient and question; old and new versions are kept apart.
R&D know-how loop · Gets smarter the more you use it
Don't just record it.
Pass the know-how on.
Stability tests, pilot-batch QC, complaints, client feedback and every failure are linked back to the formula they came from and distilled into know-how. Next time, that know-how reminds the formulator at exactly the right step, while the system guards the critical checkpoints. The longer you use it, the sharper the reminders — and the fewer repeat mistakes.
Gets smarter the more you use it
Recording is only level one. The system climbs three more.
Most R&D systems stop at recording: the data is there, but the know-how still lives in senior formulators' heads. INCILABS takes records up to Share, Remind and Control — every round you use it makes the next one easier.
Formulas, experiments, tests, complaints and clients' own words, all attached to one formula record.
Where most systems stopRecords become know-how: which formula types tend to fail how, which usage levels are involved, how it was fixed last time. New hires open a formula and see the pitfalls veterans already hit.
Know-how stays when people leaveKnow-how appears at the right step: while formulating, before sampling, at weigh-in, and when concluding. It never blocks — the formulator follows it or logs a reason.
The right word at the right timeHard limits are enforced by the system: banned or over-limit ingredients, client blacklists, unapproved materials, data permissions and change logs. Know-how reminds; red lines are controlled.
What must be stopped, is stoppedIllustrative. Hit rate comes from automatic “reminder vs. result” checks: each conclusion logs the outcome, accurate rules gain weight, inaccurate ones are downgraded or retired.
Where the smarts come from
QC results & test data analysis
First see which formula types fail,
then which usage levels are involved.
Test methods and pass/fail calls are mapped to one shared set of issue dimensions and counted separately per sub-category. A pattern only counts if it holds within the sub-category; cells with too few samples simply say “Insufficient evidence”.
- ✓One set of issue dimensionsSeparation, syneresis, thinning, discoloration, odor change, microbes… different testers' wording maps to the same term.
- ✓Dose vs. issue rateEach material is binned by usage level, with the issue rate per bin — numerator and denominator both shown.
- ✓Recording vs. formula differencesChange the tester or the form and results shift too; that effect is separated out before blaming the formula.
- ✓Lab vs. pilot batchR&D tests and pilot-batch QC for the same formula sit side by side, showing what changed at scale-up.
| Sub-category | Separation | Thinning | Color | Odor | Microbial |
|---|---|---|---|---|---|
| Repair cream | 18% | 27% | 6% | 5% | 1% |
| Active serum | 9% | 11% | 31% | 8% | 2% |
| Mask essence | 3% | 16% | 7% | 6% | 12% |
| Mist / toner | — | — | 9% | 14% | 8% |
| Cleanser | 15% | Too few | Too few | 6% | Too few |
Mechanism note: electrolytes weaken carbomer thickening systems and some emulsifiers, showing up as thinning and syneresis on storage — a common fix is switching to an electrolyte-tolerant thickener / emulsifier system.
This is correlation, not causation: the conclusion only states the historical issue rate at a given usage level; the next step is lab validation.
Sample feedback analysis
Nine in ten samples with feedback need another round — and the reasons are in the client's own words.
The comments clients send back after every sample are the most direct failure record there is. Group them by issue and split by sub-category, and you can see where each product type usually gets stuck — and avoid it before the next project starts.
One response can raise several issues. Source: two years of verbatim sample feedback from a functional-skincare contract manufacturer, grouped by keyword; client, brand and formula details removed.
| Sub-category | Fragrance | Instant feel | Pack / sheet | Texture | Irritation |
|---|---|---|---|---|---|
| Cream / lotion | 23% | 14% | 7% | 11% | 5% |
| Serum / ampoule | 14% | 21% | 8% | 10% | 8% |
| Sheet mask | 21% | 18% | 31% | 15% | — |
| Toner / mist | 39% | 18% | 28% | 12% | 7% |
Skin feel & efficacy analysis
“Light”, “absorbs fast”, “no pilling” — turned into metrics you can calculate from the formula.
Clients describe feelings; formulas are ratios. By mechanism, we map each skin-feel and efficacy need to metrics calculated directly from the formula composition, then compare them with your own sample feedback, public market practice and consumer reviews.
The red line on the meter = where the current formula sits among your similar formulas (illustrative). Mechanisms summarize public industry knowledge; the starting point is “where to start testing” — Not validated — not a recommended level.
Reminders & controls during experiments
Don't wait for test results —
see the risks before you sample.
When a formula is ready to sample, the system finds the most similar tested formulas and puts their past issues, usage notes for the materials involved and the client's comments on the previous version into one panel.
- ✓Close relatives are the main evidence“What went wrong with the most similar tested formulas” beats generic rules.
- ✓Three levels: Watch / Consider / InfoOnly patterns that hold within the sub-category trigger reminders; rules that hit everyone don't nag on every formula.
- ✓Reminds, never blocks — reasons loggedFormulators can follow the advice or record why it doesn't apply; the experiment goes ahead either way.
- ✓Reminder vs. result, checked automaticallyResults are logged at conclusion, so reminder accuracy is plain to see and rules are corrected.
Tacky / slow to absorb Watch
Of the 5 most similar repair creams already sampled, 3 got client feedback of “tacky, sits on the skin”; this formula's combined polyols and gums fall in the historically high-risk range.
Material usage note Consider
An electrolyte-type repair active: weakens carbomer thickening; the supplier recommends adding it below 45 °C, after thickening is complete. Usage here is on the high side, so run high-temperature stability too. The tablet will remind you again at this step.
Client's words on last version Info
“The repair feel is fine, but it sits on the skin after spreading — doesn't absorb as well as the reference.” — Version B. This round's changes are linked to this feedback.
- Issues seen in close-relative formulas; links between material levels and issues
- Material usage notes; client comments on the last version
- Follow it or log a reason, the experiment never stops; reasons feed back into the rules
- Banned ingredients, regulatory limits, client blacklists and caps — checked formula by formula
- Unapproved materials can't enter released formulas; approval records which check blocked them
- Who sees which formulas and who edits which data is permission-based, with every change logged
Learning from failure
Failure isn't delete-and-restart — it's a version chain with reasons.
At each round's conclusion, log the measurements, a two-level failure reason and what changes next round. Client feedback is split by issue with clear attribution. Reviews no longer mean digging through chat history — the whole chain is visible at a glance.
Client: tacky, slow to absorb
- Result
- Repair feel approved
- Cause
- Feel › polyols & gums high
- Next
- Polyols −3%, lighter oil
Accelerated stability failed
- Result
- Clearly thinner after 2 wks at 45 °C
- Cause
- Stability › electrolyte vs. thickener
- Next
- Electrolyte-tolerant thickener
Client approved
- Result
- Stability passed · feel confirmed
- Action
- Released formula
- Then
- Full stability started
Know-how saved
- Noted
- Tacky → polyols + gums total
- Noted
- Thinning → electrolyte vs. thickener
- Used for
- Reminders on the next similar formula
Example data. Failure reasons use two levels (e.g. “Feel › polyols & gums high”); delays are attributed to us, the client, external parties or materials, and review charts are computed live from these records.
Closed-loop complaints
A complaint isn't done when the form is filled — it needs an owner, a deadline and a closed loop.
In many plants, the root-cause, countermeasure and outcome columns of the complaint log stay empty for years — the problem was logged, but nobody knows what happened next. In INCILABS, each complaint runs a full 8D loop, and analysis fields must be complete before it can close.
- ✓Named owner, deadline in daysAn owner is assigned at registration, a first reply to the client is due within days, and overdue items turn red.
- ✓Actions land in each department's to-dosManufacturing, purchasing and R&D each see their assigned actions; only completed ones move to verification.
- ✓Recurrence detected automaticallySame client, same issue within a set window? It links to the previous case, so you can see why the last fix didn't hold.
- ✓Formula changes need a revision record“What changed after the complaint, and did it work” is now on record.
Root cause & actions
Root cause type: Formula / material
Action: R&D · adjust acid-active gradient, raise pH → revision F0-B linked
Action: QA · spot-check in-market lots of the same line
Required before closing
Example screen with illustrative content. With any item missing, it can't be submitted for closing and approvers can't approve; the approver can't be the owner.
Scientific literature library
Your records say “these often go together” — the literature tells you why.
In-house know-how shows correlation; mechanism and safety evidence must go back to authoritative assessments and original papers. We break safety assessment reports and journal articles into searchable claims, each traceable to the source's version, page and exact sentence.
Each paper is tagged by finish and issue, so “why does it fade?” lands straight on the relevant few.
Barrier repair, hydration, soothing, anti-aging, brightening, acne & oil control, formulation & stability — mechanisms, test methods and root causes that map to the issue dimensions in your sample feedback.
Use-level conclusion for leave-on products under the assessed conditions (example), with its preconditions listed
Safety assessment report (example) · version year · p. 12
The report's sentence, quoted verbatim — click to jump to that page of the source
This formula's level exceeds that conclusion; a newer assessment exists for this ingredient — use the latest version
Example screen with illustrative content, not a real citation. If no source supports an answer, it says so — it never substitutes a conclusion for a similar ingredient.
Four principles we hold to
Know-how must be trustworthy
to be worth a reminder.
We only say “at this usage level, this many issues occurred historically”, with numerator and denominator, and leave proof to experiments.
Sub-categories with too little data aren't rated. Better a blank than a forced conclusion.
The formulator decides. If it doesn't apply, they log why — and that reason becomes part of the review.
Formula, lab, QC and complaint records are analyzed and learned from on your own servers, never sent out.
FAQ
Putting the know-how loop to work
Start by organizing the records you already have — no need to change how you test or log first.
Our test records are messy. Can we still use them?
Yes. Import each department's spreadsheets as they are, then standardize test methods, pass/fail terms and issue dimensions. Differences caused by how things were recorded are flagged separately, not mistaken for formula differences.
How much data do we need for conclusions?
Each sub-category and issue dimension is looked at separately. Cells with enough samples show numerator, denominator and evidence level; those without say “Insufficient evidence” and fill in as records accumulate.
Most complaints aren't formula problems. What then?
Complaints are first sorted into formula, packaging, process and usage. Only formula-related ones enter the formula know-how base; the rest go to the right department, so packaging problems aren't blamed on the formula.
Won't reminders get in formulators' way?
Rules that hit every formula don't fire one by one; only patterns that hold within the sub-category and relate to this formula appear, in three levels: Watch, Consider and Info. Reminders never block an experiment.
How does this relate to Machine Learning and Lab Mentor?
Lab Mentor makes sure experiments are done right and fully recorded; the know-how loop links tests, feedback and complaints back to formulas; machine learning learns from those records to suggest the next formula's starting point. All three share the same formula record.
How is the literature library used? Only by R&D?
Formulators see the relevant claims and source text in pre-sampling reminders, know-how cards and ingredient pages; regulatory and safety colleagues can look up safety conclusions and conditions by ingredient. Every claim carries the source's version, page and sentence — and if nothing is found, it says so.
Does any data leave our company?
No. Formula, lab, QC and complaint records are analyzed on your own servers. Public market data is used only for comparison.
Bring know-how scattered across departments back to the formula.
Start with one product type and a year of stability records: see where the problems are first, then let them remind you in the next experiment.