What can AI do in a sawmill beyond the scanner?

Beyond the scanner, AI in a sawmill does three kinds of work. The first is more machine-level work: predictive maintenance and drying control, where sensor data feeds a model on one machine. The second is more vision, with cameras on the log deck, the gaps between logs, the safety zones and the yard. The third is work on the mill's records, matching what the machines and cameras see against what the inventory, production and order systems say. The third gets the least attention, and much of it needs data matching more than AI. It is also where a mill's figures can drift furthest from its yard.

What the usual answer says

The leading pages describe the first two kinds. An article by Comact's Francis Clément from February 2025, an equipment maker's own account, lists log feeding and gap control, stem positioning at the merchandiser, deck fill, singulation and safety-zone monitoring. His view of the destination is that "The sawmill of the future will be a fully integrated and connected site, where human intervention will be increasingly rare". The article gives no figures. Other top results lead with predictive maintenance, and those machine-level uses are real, with their own sensor data and their own payback in uptime and degrade.

A hardwood producer's own account adds a warning about the scanner itself. Danzer's company blog, in February 2026, says it "operated a CT scanner for over a decade", and found technical and economic limits to CT scanning in hardwood. For lumber, it says, "existing sawmill equipment lacks the precision to execute AI-driven cutting solutions". Its own optimisation cuts flooring lamellae, a secondary product, to formats "aligned with the current order book and optimized for value".

How far vision has come

Counting is a fair test of where vision stands outside the line. A 2025 study in Forests by Mazzochin and colleagues built a public dataset of 466 images holding about 13,048 eucalyptus logs, used for both training and validation. On that dataset their method counted logs with 92.3% accuracy, at about 0.713 seconds an image on a single graphics card.

In our judgement, that is enough to track a pile's size from week to week, and not enough to settle a count against a supplier. And a count says nothing about which records the pile belongs to.

Where the records go wrong

From Quarri's own work with a sawmill: 22,000+ inventory tags were still marked on hand, some more than 20 years old. A camera sees what is in the yard. A problem like that sits in what the record says is there.

Much of the work of finding it is not AI. Sorting tags by the date they last moved is a query. AI earns a place where the matching is fuzzy. Examples are pairing a handwritten tally with the production system when identifiers don't agree, reading paper records that were never keyed, or linking an order's promised grade to what the optimiser actually cut when the codes differ. None of these needs new hardware. Each needs two records the mill already keeps, read together.

A check before buying more vision

Take the inventory record and sort it by the date each tag last moved. Count the tags that have not moved in a year, then in five. If those counts are small and the yard agrees, the records are sound, and more vision on the line may be the right next step. If they run to thousands, any stock figure built on that record is suspect until the yard is walked, and adding cameras will not change that.

When it doesn't apply

A mill that does not tag or track inventory by piece has no tag record to age, and its record questions sit elsewhere, in tallies and orders. A mill whose yard turns over completely every few weeks also gives drift little time to build up. Vision used for safety justifies itself without any link to records, and so do maintenance models judged on uptime. And the counting study's accuracy is for one species and one dataset. Any counting system should be tested on the mill's own yard.

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.