My Candle Business Is Scaling and Random Quality Problems Are Increasing — How Do I Determine Whether the Problem Is Raw-Material Variation or Production Inconsistency?

Counts mislead; rates by group don't

When a business grows from one maker and a few batches to several people, several scents and several bags open at once, problems start to feel random. They aren't random. They are spread across more combinations than anyone can hold in their head. The usual reaction is to suspect whatever changed most recently, often a new bag of wax, and to change it. That tests one guess and throws away the history.

A cross-tab replaces the guess with a count, but only if the count is a rate. In the worked example, the Citrus Lemon SKU has the most defects (19), simply because it is the biggest line (500 of 1,200 candles). Its rate, 3.8%, is the same as the others'. Operator C has fewer candles than A (300 against 500) but four times as many defects (32 against 8), and that only jumps out when you divide. So the tally needs two sheets: one for defects, and one for production counts by the same keys.

The defect tally: what to log for every fault

Most of what the tally needs is already on a good batch sheet (the traceable batch-sheet template), which is why the defect log can stay short: it points at the batch code and copies the keys. Log defects found anywhere: at demoulding, at QC burn, on the shelf, from customers. Tag where each was found, because a customer-only defect can be a use problem rather than a making one.

Decide in advance what counts as each defect, write it down and show everyone a photograph of the borderline case. "Rough top" must mean the same thing on Monday and Saturday, and to every inspector. Otherwise the tally measures the inspector. The production-count sheet is simpler: for each batch, the number of candles made, with the same batch code, lots, operator, day and SKU.

Four cuts of the same 48 defects

Worked example: six weeks of production, 1,200 candles in three scents, all at 8% of total weight in Luxury Soy Wax Chunks (a 150 g candle is 12 g of oil and 138 g of wax). Three bags of wax (lots L1–L3), three operators, six production days a week, 48 defects in total: 4.0% overall. The counts are illustrative; the method is the point.

Read it cut by cut. Lots: the three wax bags sit within 0.54 percentage points of each other, about the overall rate: no lot cluster. Operators: C's 10.7% is six times A and B's combined 1.8%: a clear cluster, with 32 defects from 300 candles. Days: Wednesday and Saturday are high, the rest low. SKUs: flat. So this is a process problem, not a materials one. Operator and day both look high, and they could be the same signal seen twice. The two-way table below settles that.

Defect types add a second clue. In the worked example C's defects are mostly rough tops and sinkholes, which sit on the pour-and-cooling path. That is where a shadow run should look first (the shadow-run post). A wax-lot problem would more often show as something that follows the bag into everyone's candles: a change in how the wax sets, or sweating across all scents. Those are tendencies, not rules.

The decision rule: where the cluster sits

The table is our reading of each pattern, ranked most common first. That ranking is editorial judgement, not CSI data. The last column is what to do before buying anything.

Two cuts at once: the two-way table and the swap

In the worked example C works Wednesdays and Saturdays, so the day panel and the operator panel can't be read separately. A two-way table puts operator against day. Each cell is the rate for that person on that day.

Wednesday is the useful cell. A and C both worked it, with the same lots and the same room. A's candles came out at 2.0%, C's at 11.0%. So Wednesday itself is fine, and the high day rate belongs to C. The table can't tell you why. It could be C's steps, or C's station (a different scale, probe or bench), and that is what the swap and the shadow run are for.

What it's worth, in materials alone: at A and B's combined rate, C's 300 candles would have produced about 5 defects instead of 32, so 27 are excess. At the Citrus Lemon candle cost (₹217.07: oil ₹33.98, Chunks ₹82.79, jar, wick), that is about ₹5,860.90, out of ₹10,419.37 across all 48 defects. The labour, rework and customer goodwill cost more than the wax. Costs exclude labour, freight, packaging and GST.

When nothing clusters: measurement, definition or ambient

Sometimes every panel is flat: lots, operators, days and SKUs all sit near the overall rate. That rules out the two things the question asked about. No lot is worse, so it isn't raw-material variation between lots, and no person or day is worse, so it isn't one source of production inconsistency. What remains is anything that touches every candle equally.

Measurement and definition. Is the defect being counted consistently, or has "rough top" crept wider as volumes grew? Has the inspector changed? Are all scales and probes reading true (ice water for the probe, a shared reference item for scales)? Ambient. A hot season, monsoon humidity or a workshop move can lift every group at once (the seasonal post, the AC-room post). The process's baseline. Some defect rate exists in any hand-poured process. If it is steady and even, reducing it is a process-improvement job (slower cooling, pre-warmed jars, better centring) rather than a hunt for a cause. For planning controls as production grows further, the 100-to-1,000-unit control plan is the next read.

Wax, mixer, probe and two of the SKUs

Five items: the wax and two of the scents in the worked example, and two tools that matter only if the tally points at process. They are chosen on page statements, price and stock on 26 September 2026. CSI has no defect-rate or lot-variation data, so nothing is ranked on consistency. One production day of 200 candles at 8% uses 27,600 g of wax, which is why lot rows change quickly at scale.

Frequently asked questions