# Box sizing automation Source: > Box choice that lives in operators' heads costs you DIM weight every shipment. ML recommendations from your own shipping data fix that at the moment of pack. ## The problem Cartonisation rules in shipping platforms are static. They do not learn that an item ships better in a different box. Operators end up over-boxing fragile items, under-boxing efficient items, and paying dimensional weight surcharges that nobody is tracking. Two anchors from recent audits: - On a 2,492-SKU multi-brand catalog, only 25.1% of active SKUs had full dimensions in the system. The missing-dim SKUs concentrated on the highest-AOV bundles — exactly where DIM-weight surprises hurt most. The rate-shop quote was a default; the invoice was the real price. - On a separate audit, 43% of shipping spend went through multi-item orders. Single-SKU dim coverage doesn't fix multi-item picks; only a packing engine that knows what fits together does. ## The solution We train and operate an ML model on your historical shipping data, returns, and damage claims. The model recommends a box per order at the moment of pack, weighted for actual cost (dim weight + base rate by carrier and zone) and for damage risk. The recommendation lives in your existing pack flow. ## The outcome Carrier invoices come in lower. Damage rates drop. Pack stations stop guessing. ## Frequently asked questions **How much shipping history do you need before the model is useful?** Three months of varied shipping data is usually enough to start. The model improves continuously after that. ## Related - [/solutions/rate-shopping](https://www.gotopo.com/solutions/rate-shopping) - [/solutions/packing](https://www.gotopo.com/solutions/packing) - [/solutions/reporting](https://www.gotopo.com/solutions/reporting)