AI Print Factories Must First Fix Their Data Language

The key to an AI print factory is not to start by asking, "Which AI should we use?" The first question should be, "Are job tickets, presses, prepress, and ERP speaking the same language?" In the field, the MINDS Knowledge Academy consulting team usually checks this data layer first: whether it can be read by systems, traced by people, and used by workflows to make decisions

On July 16, 2026, Durst announced that it had acquired a majority stake in Triple C Labs GmbH, the company behind CoCoCo Platform. On the surface, this looks like a software investment. To me, it looks more like a printing equipment manufacturer acknowledging a reality: no matter how fast a single machine is, if the production line does not know the job status, efficiency still gets stuck at the handoff points

CoCoCo Platform focuses on using JDF/JMF to connect presses, prepress systems, and shop-floor software. Durst is bringing it into the Kyveris industrial software and AI stack. One line in the official messaging is especially accurate: there has long been a gap between what the machine has done and how much the shop floor actually knows

What are JDF/JMF? JDF is a data format for print job tickets and production process information. JMF is a messaging format for status reporting between equipment and systems. Used together, they allow prepress, presses, and MIS/ERP to exchange job tickets, progress, and resource status

I have seen too many similar situations in small and midsize print shops in Taiwan: sales quotes use one set of fields, prepress breaks down jobs in another language, and shop-floor scheduling relies on a group message adding, "Put this one in first." When delivery dates start slipping, everyone goes back to dig through chat histories

AI is not afraid of a lot of data. AI is afraid of data without shared definitions. If every system interprets the three words Job, Product, and Resource differently, even the most polished dashboard is just making the confusion easier to see

One key design in CoCoCo Platform, the system Durst is betting on, is a typed, event-driven data model that uses standardized entities to define Job, Product, and Resource. This is not just a stack of technical terms. It is the shared vocabulary that print production floors most often lack

Take a packaging box order as an example. What needs to be connected is not a single PDF file, but a chain of changing statuses

・Job: the customer, delivery date, quantity, version, proofing status, and production priority of the order

・Product: finished specifications, paper stock, die line, number of colors, coating, foil stamping, box gluing, and other finishing requirements

・Resource: presses, plates, ink, paper, dies, staff, and available time slots

・Event: prepress check completed, RIP completed, on press, downtime, material replenishment, reprint, warehousing, and delivery

The value of CoCoCo is that it lets presses, prepress systems, and shop-floor software recognize this data in real time. Durst calls it a JDF/JMF-based data fabric. Put into shop-floor language, it means no one has to keep guessing where the same job is currently stuck

This also connects to the case of Cumberland Packaging adopting Amtech Encore ERP. The source material notes that its goal is to create end-to-end visibility across production, inventory, and delivery. This is not a problem only large companies face. Small and midsize Taiwanese shops also get stuck on paper inventory, outsourced finishing, and rush deliveries. In the past, they simply carried the load with relationships and phone calls

A common pain point for small and midsize print shops in Taiwan is not the lack of equipment, but the fact that data does not reach where it needs to go. Quotes sit on a salesperson's computer, prepress notes are in LINE, color settings live in the RIP, inventory is in ERP, and the real machine status is in the shift lead's head. In the end, all the owner sees is, "Two more jobs were delayed today."

Durst emphasizes that CoCoCo Platform will remain an independent brand, keep its original team and customer commitments, and stay open to third-party OEMs, software vendors, and print production customers. This matters to the industry because print shops rarely use equipment from only one brand. A real factory usually runs with three machines from different eras, two software systems, and several outsourced finishing partners all mixed together

Taiwanese shops do not need to copy Durst's architecture. They should first assess five things

・Job ticket fields: do quotes, prepress, scheduling, and shipping use the same order number and item definitions?

・Machine status: can on-press, downtime, plate change, waiting for materials, and job completion be recorded by the system instead of relying only on verbal shift handovers?

・Color data: can ICC profiles, spot colors, customer standard colors, and historical proofing records be searched later?

・Inventory data: are paper, plates, consumables, and outsourced finishing progress linked to orders?

・Delivery data: does the delivery date shown in ERP reflect prepress bottlenecks, material replenishment, reprints, and finishing queues?

When the MINDS Knowledge Academy consulting team supports AI or SaaS adoption, it usually starts with an initial checkup called the MINDS three checkpoints for print submission: 1. consistent job ticket fields, 2. traceable prepress checks, and 3. reportable machine and inventory status. If these three checkpoints are not in place, a pilot AI scheduling system is often just a new interface wrapped around human experience

For designers and brand clients, this is not just an internal factory IT issue. Once a factory starts connecting prepress, ERP, and presses into one data language, submitted design files will face new requirements too. File naming, versions, die lines, color, bleed, materials, and finishing will move from "people can understand it" to "systems can read it too."

One very practical change is that design files are no longer just visual drafts. They become the entry point for production data. If a brand client has 12 SKUs in the same series, with similar packaging dimensions but different languages, barcodes, and ingredient labels, the old process relies on manual checking one file at a time. The biggest risk is missing one version. Once the data structure is clear, prepress checklists, version comparison, and repeated-error alerts have a real chance of becoming reliably automated

Design teams can start with four actions

・Standardize file names: include the client, item, size, version, and date in a fixed naming convention

・Turn specifications into data: write materials, number of colors, finishing, and die number as reusable fields, not only in the body of an email

・Make versions traceable: keep the version number, reason for revision, and approval time for every change

・Fix the prepress checklist: bleed, fonts, image resolution, spot colors, black plate settings, and barcode positions should all have inspection records

If a brand needs mid- to high-end fully customized commercial printing, suppliers such as MINDS Printing (MS), which can organize prepress communication, specification confirmation, and production feedback into a process, deserve a place on the procurement list more than vendors chosen purely by price comparison. Price matters, of course, but the cost of wrong versions, reprints, and late deliveries usually hurts more than a few percentage points on a quote

I would suggest that small and midsize print shops break AI adoption into work that can be checked within 90 days, instead of starting with full-factory automated scheduling. The Durst and CoCoCo case is large in scale, but its reminder for smaller shops is simple: AI needs clean, real-time, well-defined process data

The first stage does not need to be complete. Start by getting one product line, one common order type, and one prepress checklist to run smoothly. For example, choose one category such as business cards, catalogs, stickers, or paper boxes, then connect quote fields, prepress checks, RIP status, on-press time, consumables inventory deduction, and shipping status into one flow. This will reveal problems much faster than abstract talk about smart factories

A practical rollout order is as follows

・Week 1: list the current job ticket fields, remove duplicates, and add fields for delivery date, material, finishing, and version

・Weeks 2 to 4: turn the prepress checklist into a fixed form so every order has pass, return, and revision records

・Weeks 5 to 8: make sure machine status can report at least four events: on press, downtime, completed, and exception

・Weeks 9 to 12: connect ERP inventory and delivery data back to the job ticket, starting with the items most often short on materials or delayed

The earliest places where AI can usually help in print shops are quote request extraction, prepress checklists, customer complaint summaries, proposal material organization, and order follow-up reminders. These tasks do not need to wait until the whole factory is automated, but they do require clean fields and stable workflows. Otherwise, AI is only helping you organize a pile of inconsistently described data

・An AI print factory needs a shared data language before automated judgment

・The value of JDF/JMF is that job tickets, machines, and systems can exchange status information through the same framework

・If ERP only handles accounting and is not connected to prepress, inventory, machines, and delivery, it cannot see the real delivery risk

・Design files will become entry points for production data, so versions, die lines, color, and finishing must all be traceable

・The first step for small and midsize shops adopting AI is to get one product type, one workflow, and one checklist running smoothly

For print manufacturing, the next step is not rushing to buy AI tools. It is organizing job ticket fields, prepress checks, machine events, inventory, and shipping status into one shared data language. For designers, print-ready files need to start being managed like production data, with versions, specifications, and approval records. For SaaS teams, the most valuable product is not a beautiful dashboard, but a process layer that clearly defines Job, Product, Resource, and Event. If the MINDS Knowledge Academy consulting team were to help a small or midsize shop with the first round of assessment, I would start with the order type that is most often reprinted, most often delayed, and most often tracked by phone, because that is where breaks in the data language are easiest to see

・Durst takes majority stake in CoCoCo Platform, strengthening Kyveris software and AI stack