Why a balm that behaves in a test batch falls apart when you scale the run up

Surface area per gram collapses as a batch grows, so bulk cools slowly and grains. Cooling arithmetic, split pours and the process fixes for a large run.

The formula did not change. The percentages are identical, the ingredients came off the same pallet, and the 100 g test batch was smooth, firm and consistent every time you made it. Scaled to 5 kg it grains, sweats, sets soft or separates into layers. Nothing is wrong with the recipe: what changed is the geometry of the vessel it cooled in, and geometry sets the cooling rate that decides fat crystal size.

• Area per gram falls as mass to the power minus one third
• 100 g to 5 kg: 3.7 times slower cool
• 5 kg to 80 C: 45 to 90 minutes
• Danger window: 40 to 25 C
• Rescue in 1 kg splits

The one number that changed: surface area per gram

A melt loses heat through its surface and stores heat in its volume. Volume grows with the cube of a linear dimension and surface area with the square, so as a batch gets bigger the area available per gram of fat falls as the cube root of the mass. Nothing else in the process has changed, but the single variable that controls how fast the batch cools has dropped by a factor of nearly four.

The arithmetic is worth seeing. Treat the melt as a cylinder as tall as it is wide, at a fat density of about 0.9 g/mL, with side, base and open top all exchanging heat.

Two things widen the gap beyond the calculated ratio in practice. A lid removes the top surface and traps a warm, saturated headspace, easily adding an hour to a 5 kg cool. And a lumped time constant assumes the contents are at a uniform temperature, which is only true if you are stirring. In a still 5 kg pot the centre stays hotter than the wall for a long time, so the middle of the batch crystallises under conditions the edge never saw.

What a slow cool does to the fats

Crystallisation rate governs crystal number, and crystal number governs crystal size. Cool fast and you nucleate in a burst, producing many small crystals that build a fine network and feel smooth. Cool slowly and few nuclei form, so the crystals that do appear grow large at the expense of the melt around them. The mechanism is set out under crystallisation and cooling rate. Every fault in the table below comes out of the same slow crawl.

The window that matters for most balms runs from roughly 40 C down to 25 C, which is where the butters do most of their crystallising. Beeswax starts building structure higher, from the low 60s, which is why a wax-heavy formula tolerates scale-up better than a butter-heavy one. If the fault you are seeing is grainy shea butter, balm sweating or separation into layers, those pages carry the detail on each mechanism. What this page adds is the reason all three arrived at once when the batch size went up.

Heating is slower too, and the hold gets long

Scale-up conversations concentrate on cooling, but the heat-up leg fails first and does more damage than makers expect. A domestic bain marie cannot exceed 100 C, so once the fat reaches 80 C the driving temperature difference is only about 20 C and the last few degrees crawl. On a typical setup, 100 g reaches 80 C in 10 to 15 minutes and 5 kg takes 45 to 90 minutes depending on pot wall thickness, water contact area and whether you stir.

That matters for three reasons. The batch spends an hour or more above 60 C, which accelerates oxidation of the unsaturated oils and degrades tocopherol before the product is even poured, a cost explained under rancidity and oxidation. Colour bodies in unrefined butters develop with time at temperature, so a large batch of unrefined shea comes out visibly more yellow than the test batch. And the temptation to move the pot onto direct heat to speed things up creates a hot zone at the base where the fat can locally exceed 120 C while a thermometer in the middle still reads 70 C. The options are compared under melting methods.

Volatiles do not scale linearly

Fragrance is the one component where percentage arithmetic quietly stops being true. Evaporation is a surface process and the mass is a volume, so the fraction lost per gram falls as the batch grows, while the total time spent hot rises. The two effects run in opposite directions and they do not cancel in any predictable way.

What that means in practice is that a 5 kg batch dosed at the same percentage as a 100 g test can smell weaker, because it sat above 60 C for an extra hour, or stronger, because the same fragrance was added to a covered pot where condensate on the lid returned to the melt instead of escaping into the room. Neither is a formula error. The fix is to standardise the moment of addition rather than the percentage alone: add fragrance and heat-sensitive actives at a defined temperature, typically 50 to 60 C, with the lid on, and stir for a fixed number of minutes before pouring. That makes the variable controllable, and a sensory comparison against a retained reference sample then measures it. If the result is a run that smells thin, scent fading in balm covers the dosing side.

Mixing, settling and the deep pot problem

At 100 g a silicone spatula moves every part of the melt within a few seconds. At 5 kg in a 19 cm pot, hand stirring produces a turning surface layer over a nearly stationary lower half, and the melt develops a vertical temperature gradient of several degrees. Anything denser than the fat has time to fall. Zinc oxide, iron oxides, clays, starches and undissolved actives all settle when they have three hours of liquid time instead of forty minutes, which is why pigment sinking is largely a scale-up fault. Suspended solids also need a yield stress to hold them, and the coarse network from a slow cool has less of it than the fine one from a fast cool.

The practical answer is a stick blender or an overhead stirrer used through the cooling leg rather than a spatula used at the top. Stirring does double duty here: it keeps the batch thermally uniform so the lumped cooling calculation actually holds, and the shear multiplies nucleation sites, which pushes the crystal population back towards many-and-small. Do not aerate. A stick blender held at the surface whips air into the melt and you get air bubbles through the set balm, so keep the head submerged, the speed low and the vortex shallow.

Diagnose it: log the centre temperature

Every scale-up argument on a forum is settled by one dataset that takes no equipment beyond a probe thermometer. Log the temperature at the centre of the melt every ten minutes, from the start of heating to the point where the poured containers reach room temperature, and overlay it on the same log from the batch size that worked.

1. Probe in the middle, not at the wall. Suspend the probe halfway down and away from the side, and leave it there. An infrared reading off the surface tells you about the surface, which is the part that is behaving.
2. Record the clock time and the temperature every ten minutes. Heating leg, hold, cooling leg, and the poured containers. Note the moment of each addition and each pour.
3. Plot both curves on the same axes. Time on the horizontal, temperature on the vertical, small batch and large batch together.
4. Measure the time in the window. Count the minutes between 40 C and 25 C for each curve. This is the number that predicts grain, and a ratio of three or four to one between the two batches is your diagnosis.
5. Check the hold above 60 C. Sum the time the large batch spent hot. If it is more than about double the small batch, the oxidation and colour differences have their explanation too.
6. File it with the batch record. A cooling curve attached to a batch record turns a one-off rescue into a specification you can repeat, which is the point of keeping batch records at all.

If the two curves are close and the fault persists, the cause is not cooling rate. Look at weighing accuracy, at ingredient lot variation, or at the possibility that your thermometers disagree, which is more common than makers expect and is covered under thermometer readings that disagree.

Rescue the batch in front of you

Remelting is safe for every fault in this family. Grain, stratification, sweating and a soft set are physical, not chemical, and melting past the clear point of the crystals restores the original material. The exception is a batch that already smells oxidised or has been held hot repeatedly, because each remelt adds thermal history to the oils and no amount of recrystallisation fixes that. One rescue melt is reasonable. A second means the process needs redesigning, not repeating.

1. Scrape the whole lot back into one weighing vessel. Rescuing pot by pot guarantees an inconsistent result across the run.
2. Split into 1 kg lots. This is the whole rescue in one instruction. Five 1 kg melts have roughly 1.7 times the area per gram of a single 5 kg melt, and each one can be handled at a pace a person can control.
3. Melt to 75 to 80 C and hold 20 to 30 minutes. Long enough to erase crystal memory in the hard butters. Trust a probe in the fat rather than the moment the last lump disappears.
4. Cool in a water bath with stirring. A sink of cold water to the level of the fat, stirring steadily. The target is to cross 40 C to 25 C in minutes rather than hours.
5. Add heat sensitive materials on the way down. Fragrance, flavour and tocopherol at 50 to 60 C, stirred for a fixed time.
6. Decant into prewarmed containers. Tins straight from a cold store chill the outer millimetre and cause dips and skinning. Warm them to about 30 C first, and pour at the temperature set out under pour temperatures.
7. Leave the filled containers uncovered until fully set, then lid. A lid on a warm tin condenses volatiles onto its underside and slows the surface set, which produces tunnelling and dips.
8. Retain samples from the first, middle and last lot. Assess at ten days, not on the day, because polymorphic change continues in the tin.

Prevent it: design the process for the pot

Scaling a formula is arithmetic and takes a minute with the batch calculator. Scaling a process is engineering, and the levers are all about heat exchange area.

• Geometry

  Wide and shallow beats tall and deep

  A 5 kg batch in a wide, low pan has far more surface per gram than the same mass in a stockpot. If the melt must live in a tall vessel, decant it into shallow trays or straight into containers as early as the pour temperature allows.

• Cooling

  Water bath, not still air

  Water carries heat away roughly twenty times faster than still air at the same temperature difference. A sink of cold water at the level of the fat is the cheapest single upgrade to a scaled process.

• Agitation

  Stir through the window

  Steady stirring from the melt down to visible thickening keeps the batch uniform and multiplies nucleation sites. This is the difference between many small crystals and a few large ones at the same cooling rate.

• Splitting

  Batch the melt, not the recipe

  Weigh one 5 kg mix for consistency, then melt and cool it in 1 kg lots. You keep the compositional uniformity of a single weigh-up and the thermal behaviour of a small batch.

Weighing changes at scale as well, and in a direction that catches people out. A jeweller's scale reading to 0.01 g is useless above 500 g, while a 6 kg bench scale reads to 1 g, so a minor ingredient at 0.1% of 5 kg is 5 g and is fine, but an active at 0.05% is 2.5 g and now has a 20% weighing tolerance on a coarse scale. Weigh majors on the bench scale and minors on the precision one, and calibrate both, as set out under weighing and calibration. The equipment thresholds where hand methods stop working, and what replaces them, are covered in scaling up production and workshop equipment.

Filling time is part of the formula

A 5 kg batch is around 1000 lip balm tubes or 330 fifteen gram tins. Hand filling at a brisk pace takes 40 to 90 minutes, and the melt does not wait. The last container is filled from a bulk that has dropped 10 or 15 degrees, or has begun to set at the edges, which shows up as a difference in gloss, hardness and shrinkage across the run. That is a major contributor to batch to batch inconsistency, and it is invisible at 100 g because a small batch is emptied in three minutes.

Three real answers exist. Hold the bulk at pour temperature in a warm water bath and draw from it as you fill, which is what a jacketed pouring pot does. Split the batch and reheat lots one at a time. Or fill faster, which is where a filling head earns its cost, discussed under filling equipment.

What does not scale, and where to stop

The honest limit is that a formula proved at 100 g has not been proved at all for production. It has been proved for a cooling regime that a 5 kg pot cannot reproduce, and no percentage adjustment substitutes for that. The decision rule that follows is a step ladder rather than a jump: prove the process at 500 g, then 1 kg, then the target size, logging the cooling curve at each step and holding retained samples for ten days before scaling again. If a step introduces a fault, the previous step is your production size until the equipment changes.

Some things genuinely do scale linearly: the ingredient percentages, the label ingredient list, the allergen declaration and the safety assessment, provided the formula is unchanged. What does not scale is anything with a time or a temperature attached, meaning melt time, hold time, cooling time and filling time, along with every consequence those have for crystal size, colour, fragrance and oxidation. Write those into the batch record as targets with tolerances, and the sixth 5 kg run will behave like the first, which is the only definition of scale-up that matters. The formal framework, if you sell, is process consistency under batch scaling.

Reviewed and updated 6 September 2026. Spotted an error? Tell us and we will fix and log it.