Where does AI fit in a box plant's order book?

AI fits a box plant's order book at three points. First, getting orders in: reading purchase orders, emails and specification sheets into structured lines with board grade, flute, dimensions, print, quantity and date. Second, combining orders on the corrugator: choosing board and widths to cut trim, where optimisation beats expert planners in published tests. Third, promising dates: setting delivery dates from the actual schedule and re-promising when it changes. Combining has the most evidence, but plants have run optimisers for it for decades, so the newer gain is at intake and date-promising. The optimiser also depends on intake: it can only plan the orders it is given.

What the usual answer says

The usual answer lists AI features: automated order entry, demand forecasting, corrugator scheduling and trim optimisation, delivery date prediction. All are available in some form. A ManufacturingTomorrow piece from August 2026 makes the useful caution: much of what is sold as plant-floor AI is "optimization logic, scheduling rules, and automated decisioning that manufacturing execution systems and industrial control platforms have used for three or four decades". It suggests buyers ask "what specifically is new about the underlying method".

Combining and trim: the evidence

Iori Minakawa and colleagues, in a 2025 study in IEEE Access that we read as its abstract, used evolutionary optimisation to choose paper type and width for each order. They tested it on "simulated order lists generated based on real factory data" and concluded: "In all cases, planning based on evolutionary optimization produced better schedules than those created by human experts". The comparison is with people, not with the scheduling software many plants already run. It shows the value of optimising over hand planning, not the gain from a new tool.

How sensitive trim is to such choices shows in a worked example by Constantine Goulimis and Gastón Simone, who work for Greycon, a vendor of planning and trim software. In a 2020 paper, a five-order example shows that changing one stocked roll width by 75 millimetres moves corrugator waste on the same orders from 3.071% to 2.584%. Roll widths are a stocking policy, set rarely, which is why they are worth getting right.

Promising dates: plan by order

Dates are the order book's other half. A study in sawmilling, by Francisco Vergara and colleagues and published in 2024, compared planning by product totals per period with sequencing whole orders. Product-level planning left 17% of orders late, against 0.5% when orders were sequenced by due date. Its model placed no penalty on backlogs. Our inference is that the same risk applies when a box plant plans by run and board: a plan can meet every run target while individual orders slip.

Intake: the quiet precondition

An optimiser needs every open order as clean data: grade, flute, blank size, score lines, print, quantity, due date, delivery location. Where orders arrive as emails, PDFs and phone calls and are keyed by hand, errors flow straight into the plan. A wrong flute, a transposed dimension or a date keyed as the order date all reach the corrugator. The optimiser then produces a very good schedule for the wrong orders.

AI reads those inputs well when checked against rules: the customer's item master, the specification last used for that item, feasible board and flute combinations. Orders that don't match go to a person. Measure it by how many orders need correction after entry, before and after.

What changes for the people

Customer service moves from keying to checking the orders the system could not match. The planner moves from building the corrugator plan to reviewing the optimiser's and overriding it where they know better. Those overrides are worth recording as data. Over time they show where the optimiser's rules miss something, and they keep the planner's knowledge in the business when the planner moves on.

When it doesn't apply

Sheet plants that buy all their board from one supplier's corrugator have no combining problem of their own, though intake and promising still apply. Plants with a few large repeat customers on EDI already have clean intake. And the 2025 study used simulated order lists from one factory's data, so it shows that optimisation can beat experts, not by how much at a given plant.

Quarri for pulp, paper and packaging is built around how a fibre operation runs, from furnish to converted 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.