What BOM cost variance actually is
The standard bill of materials says one finished unit of Product A should consume 1.2 kg of raw steel, 80 grams of welding consumable, and 4 minutes of CNC time at a defined cost per minute. The actual issue records over a week say the line consumed 1.27 kg of steel, 92 grams of welding consumable, and 4.6 minutes of CNC time per unit. The gap between the two, multiplied by the units produced, is BOM cost variance.
On a single shift, the gap looks like noise. Across a quarter, it is the difference between a 14% gross margin and an 11% gross margin. Most Indian mid-market plants discover the slippage at the year-end audit, when the stock-take adjustment hits the P&L and the owner asks where the missing 3 points went.
The reason it stays hidden is plumbing. The standard BOM lives in your ERP or in a master sheet maintained by the industrial engineer. Actual consumption sits in stock-issue records on the shop floor and in purchase entries posted to Tally. Scrap and rework numbers sit in the quality log. Joining the four sources by hand every week is a 6 to 10 hour analyst job that nobody owns.
The five sources you have to join
| Source | What it gives | Typical system |
|---|---|---|
| Standard BOM master | Per-unit material and time targets | ERP module or industrial engineer's Excel |
| Production output | Units produced by SKU, line, shift | Shop-floor sheets, ERP production module |
| Stock issue records | Actual material drawn against work orders | ERP inventory module or store register |
| Purchase entries | Landed cost for each raw material lot | Tally Prime purchase ledger |
| Scrap and rework log | Material lost to defects, not finished units | Quality module, often a separate notebook |
Without all five, you can compute a variance number but you cannot diagnose it. A favourable variance that turns out to be unbooked scrap is worse than no number at all - the team stops looking.
What the variance pattern is telling you
The number on its own is not the insight. The shape of the variance, who it shows up against, and how it moves week to week is the diagnostic.
| Pattern | Likely cause | First check |
|---|---|---|
| One material, all SKUs, sudden jump | Vendor price drift on the latest lot | Compare last 3 purchase invoices for the item |
| All materials, one SKU, gradual rise | Yield slippage on that product line | Pull rework and scrap log for that SKU |
| One material, one shift, repeating | Measurement error or pilferage on that shift | Spot-check stock issue against gate register |
| Spike that reverses next week | Stock-issue timing misalignment, not real loss | Match issue dates to production dates |
| Negative variance (used less than standard) | Standard BOM out of date, or under-reported scrap | Re-validate BOM with industrial engineer |
A good variance report does not just flag the number. It flags the pattern and points the production head at the first check to run.
The cost of detection lag
The longer it takes to spot a variance, the more material you have already wasted at the bad rate. Most Indian mid-market plants run the report monthly at best, quarterly for many. By the time the number is flagged, the next set of work orders is already running with the same defective process or the same overpriced lot.
The plants that run BOM variance weekly are not running fancier analytics. They are running tighter loops. The finance head and the production head share the same number on Monday morning, agree on the diagnosis, and act before the week's purchase orders go out.
Item master discipline - the unsexy prerequisite
Every Indian mid-market plant has a quiet item-master problem. The same raw material is called CRC 0.8mm in the ERP, CRC sheet 0.8 in Tally purchase entries, and Sheet 800-grade on the stock issue register. Until those three names point at the same physical material, no variance report is trustworthy.
- One canonical name per material. Picked from the ERP and pushed downstream. Tally and shop-floor systems either match the canonical name or carry a documented alias.
- Unit of measure rationalised. Steel issued in kg in the ERP and in MT in Tally is a 1000x error waiting to land in your variance report. Pick one UOM per item, convert at the boundary.
- Vendor item codes mapped, not embedded. When a supplier's part number changes, the alias updates in one place. The base item identity stays stable so historical variance series do not break.
- Scrap and rework as named events. Not free-text comments. The variance report can only attribute loss if the loss has a recognised category.
The good news: AI does most of the alias matching for you during onboarding. The bad news: the four discipline points above stay yours. If your item master is genuinely chaotic, the first two weeks of any variance project are a cleanup exercise.
How AI handles messy item codes across systems
The KolossusAI onboarding for a manufacturer always starts with item-master reconciliation across the ERP, Tally, and stock-issue records. The AI proposes alias matches based on name similarity, vendor history, and consumption pattern. The industrial engineer reviews and approves the matches in a single pass. From that point on the variance report runs against canonical items, regardless of which system raised the entry.
New items added later are auto-matched against the existing canonical set, and the AI flags the ambiguous ones for human review instead of guessing. This means the variance report does not silently break the week somebody adds a new SKU or onboards a new vendor. AI for Indian manufacturers covers the full multi-system pattern; AI for Tally Prime users is the right entry point if Tally is your primary system of record.
What a weekly BOM variance review looks like
The Monday plant review with a working variance report takes 20 minutes, not 2 hours. The production head opens the report, reads the top three variance flags, and the team agrees on owners and timelines. Each flag carries the drill-down already - the lots affected, the work orders they ran on, and the underlying purchase invoices or stock issues.
The owner does not need to be in the room every week. The variance number is shared on WhatsApp on Monday morning, with a one-line summary the AI generates. The owner reads it during the morning chai, asks one question if the number is unusual, and gets a drill-down answer in plain English before the second cup is done.