Can AI schedule a sawmill against open orders?

Yes, but the bar is higher than it sounds, because the simple rule many mills already use does well. In a 2024 study, sequencing whole orders by earliest due date left 0.5% of orders late, while a textbook optimisation model that planned by product total left 17% late. So the question for an AI scheduler is what it adds beyond the due-date rule. The study doesn't test that. Our answer is re-planning when logs, machines or orders change, handling uncertain yields, and reading orders that arrive as documents.

What the usual answer says

Descriptions of AI scheduling for sawmills say it uses machine learning and optimisation to plan cutting patterns and runs against orders, log supply and capacity, improving on-time delivery and yield. That is the promise. It rarely says what the scheduler is compared with, and the comparison decides whether it is worth having.

What the research found

Francisco Vergara, Cristian Palma and John Nelson, in a study published in April 2024, compared ways of planning the same sawmill orders under the same constraints. One was an optimisation model "where orders were split up into products demand by period". The others sequenced whole orders by simple rules. The costs were similar. The lateness was not: "0,5 % of orders were delayed using PS-E, and 17 % of orders were delayed using PL", where PS-E is sequencing by earliest due date and PL the product-level model.

They also note how mills schedule in practice: "in most cases static heuristics, such as earliest due date (E), longest processing time (L), and shortest processing time (S), are used because of their simplicity". For a mill cutting to order at 300,000 cubic metres a year, they estimate that adopting the order-based approach, under a conservative backlog cost, would cut backlogged orders by 51,000 cubic metres a year, compared with product-level planning.

A related 2024 study by Palma and colleagues, in Mathematics, which we read as its abstract, found that modelling orders explicitly in a planning model cut costs by about 6%, by allowing early order completion.

The lesson from both is to keep orders whole. The due-date rule does that, and so should any scheduler that replaces it.

What the study can't show

The test orders were fixed and the logs known. In a running mill, logs arrive with a different size mix than planned, machines go down, and customers add orders or move dates. If rebuilding a schedule takes a day, the mill follows the old one after it has stopped being right, or falls back to the due-date rule without knowing what that costs. The study doesn't measure any of that.

That is where solving speed matters. From Quarri's own work with a sawmill: a production scheduler re-solves an order book in about a second, and lands within 1 to 1.6% of the best possible schedule found by exhaustive search. That is our own evidence, and it shows speed and closeness to the best schedule, not fewer late orders than the due-date rule. That comparison is the one to ask any vendor for.

Where AI fits around the solver

Optimisation does the scheduling. AI helps with what feeds it: reading orders that arrive by email or PDF into structured lines, forecasting the log mix from the deck and the week's deliveries, and flagging orders whose promised dates can no longer be met. It can also explain in plain language why a schedule changed.

What the scheduler needs to know

It needs open orders with products, quantities and dates; the log inventory by class; what each class yields in each product under each cutting pattern; and machine capacities and planned downtime. Check where the yield table came from and when. If it predates the current scanner and tally data, the mill's own data can supply a current one.

How to judge a scheduler

Measure it on late orders against the due-date rule, not on product totals. Take a past month's orders, logs and downtime, run both as if live, and count late orders for each. Then change something mid-month, such as a breakdown or a rush order, and compare again. If the scheduler doesn't beat the rule after disruptions, the rule is the better buy.

When it doesn't apply

Mills producing only to stock, for sale on the open market, have no order due dates to protect, and product-level planning suits them. Very small mills with a few orders a week can sequence them by hand. And the 2024 figures come from one set of test orders, so they show the size of the gap, not what any given mill will see.

Quarri for sawmills is built around how a sawmill runs, from log intake to shipped order.

Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.