Raw Materials, Yields & Waste: Making the Production Line Tell the Truth
Between the raw store and the finished shelf, margin disappears through yields, waste, and unmeasured issues — the batch discipline that makes a Kenyan processor's production line tell the truth.
Ask a miller what their extraction rate was last week and you learn everything about the business. The ones who answer with a number — "78.2%, down half a point, we are checking the conditioning" — run production as a measured system. The ones who answer "good" run it as a hope. The difference is not equipment or scale; it is whether raw materials are issued against recipes and outputs are weighed against inputs, batch by batch.
The batch loop
- Plan: a production order states what will be made and the recipe computes expected material consumption.
- Issue: the store releases materials against the order — weighed, recorded, and nothing more. Open-door stores where production "takes what it needs" cannot be measured.
- Produce: the run happens; downtime, rework, and waste get recorded as they occur, with reasons.
- Receive: finished output is weighed/counted into finished-goods stock; by-products (bran, cake, offcuts) are received as their own items.
- Reconcile: input vs recipe (consumption variance) and output vs input (yield) — two numbers, every batch, reviewed weekly.
Reading the two numbers
| Signal | Likely cause | First move |
|---|---|---|
| Consumption above recipe, yield normal | Over-issuing, spillage, or store leakage | Tighten issuing; weigh at the store door, not the mixer |
| Consumption normal, yield falling | Process problem — moisture, settings, raw quality | Check the input batch quality records; inspect the line |
| Both drifting together | Recipe no longer matches reality | Re-baseline the recipe with a supervised test run |
| Numbers perfect, always | Records are being written backwards from output | Weigh independently; separate the recorder from the operator |
Waste is a transaction
Damaged output, floor sweepings, burnt batches, trimmings — every kilo of waste is recorded with a reason, or it becomes the line's all-purpose explanation. Processors who start recording waste typically discover it was covering for something else: a scale that reads heavy, a recipe that drifted, or product leaving through the side gate.
Raw material quality is a yield input
Agri inputs vary: moisture in maize, fat content in milk, fiber in cane. A batch that yields poorly often traces to a delivery that was accepted generously — which is why receiving records (weight, moisture, grade, supplier) belong in the same system as production. Over time, yield-by-supplier is one of the most valuable reports a processor owns: it converts "we like this supplier" into "this supplier's maize yields 2% better, which is worth KES 1.80 per kilo more than we pay them". Quality holds at receiving enforce the standard; the yield data justifies it.
Committed stock: the promise ledger
A confirmed order for 500 cartons next week is a claim on raw materials today. Systems that show only physical stock let sales promise the same materials twice — production then chooses which customer to disappoint. Available-to-promise = physical − committed − quality-held, and it is the number sales should see. The costing side of the same chain — what those materials truly cost — is the landed cost discipline; together they make the manufacturing margin honest.
What AWRA OpsHub does today
- Landed cost on receipts — freight, duty, clearing and transport captured per purchase order, allocated across the lines by value or by quantity, written onto the batch and carried into the item's weighted average cost.
- Batch, lot and serial tracking — expiry date, supplier and originating purchase order recorded on every batch, with serial numbers where you need them.
- Stock adjustments with mandatory reason codes, attachments and full history — the record a waste transaction would be written as today.
- Quality holds that are enforced, not advisory — quarantined, inspection-pending, damaged, expired and returned stock is excluded from issuing and selling by the allocation query itself. Hold, release and dispose each record the quantity, the reason, a note and the person who did it.
- Supplier performance — on-time delivery rate, quality score and pricing trend per vendor, which is the yield-by-supplier question answered from the buying side rather than the production side.
What it does not do
- There is no bill of materials. No BOM, no recipe, no version, no effective date — the entity does not exist in the database.
- No production order and no issue-against-recipe. Materials leave the store as a stock adjustment or a transfer, not as consumption measured against an expected quantity.
- No yield and no consumption variance. Expected-versus-actual is the entire point of a recipe, and there is nothing here to compute it from.
- No by-product receipt. Bran, cake and offcuts cannot be received as their own items off a run, because there is no run.
- No committed-versus-available. The `reserved` and `allocated` statuses exist as labels, but nothing sets them — a confirmed order does not ring-fence the raw materials behind it.
- No weighbridge or moisture capture at receiving. Verified weight, moisture and grade are not fields — they would be notes, and supplier payment is not computed from them.
This is a discipline every processor should run, and the five steps are the right five. Four of them — plan, issue, produce, reconcile — need a production entity we have not built. What we can genuinely carry today is the receiving and store side: batches, landed costs, holds and adjustments with reasons. If you are choosing a system on yield control, weight our absence here heavily.
Make the line tell the truth
Batch receiving, true landed costs, enforced quality holds and adjustments with reasons — the store side of the batch loop. Production runs, recipes and [yield variance](/glossary/yield-variance) are not built; see the note above.
See the store sideFrequently asked questions
Our production is continuous, not batch-based. Does the loop still apply?
Yes — define a batch as a shift or a day and run the same reconciliation on that window. Continuous processes actually benefit more from the weekly trend line, because drift compounds quietly when there is no natural batch boundary to force a count.
How accurate do recipes need to be before we start?
Start with your best current estimate and let the variance reports correct it: three weeks of real consumption data produces a better recipe than three months of pre-launch measurement. The discipline of issuing against something matters more than the initial precision of the something.
Where does rework go in the numbers?
Rework re-enters as an input to a new run, recorded as such — its materials were already consumed once, so hiding it inflates yield and flatters the week that produced the defect. Visible rework is annoying; invisible rework is a margin lie.
What is a realistic yield improvement from doing this?
Sector-dependent, but the pattern is consistent: the first quarter of measurement typically recovers 2–5% — not because anything was optimized, but because measurement itself closes the deniability that leakage and drift lived in. Optimization comes after, on top.
