The Path to Packaging Intelligence

There are three recurring bottlenecks in the packaging design-to-print process (see Figure 1):

• Briefing
  The process of defining packaging requirements—brand, product, regulatory, and print specifications—to guide design and production teams.
• Artwork Creation & Approval
  The stage where packaging visuals are designed, reviewed, and approved across stakeholders to ensure brand, legal, and print compliance.
• Prepress
  The technical preparation of approved artwork for print, including color separation, trapping, proofing, and file optimization for production accuracy.

Among these, artwork creation & approval stage is the most time-consuming in the packaging lifecycle.

This phase demands close coordination between design, marketing, regulatory, quality, and supply chain teams. A package update that should take 5 days can sometimes take months. A single missed allergen declaration can trigger a recall costing millions. Even brands with mature processes and agency support report an average of 31 days in this stage alone.

Most packaging delays are driven not by design work itself, but by coordination gaps and repeated rework.

In a market where speed to launch is a critical competitive advantage, such delays translate into substantial opportunity costs.

The risks extend beyond delays. Packaging errors that reach the market can lead to recalls, regulatory action, and lasting brand damage. FDA data shows that the global food and cosmetics sectors experience roughly 1300 product recalls annually, with 31% linked directly to artwork and labeling errors. The situation is further compounded by an ever-evolving regulatory landscape, increasing the burden on packaging and compliance teams.

Most brands already invest heavily in artwork checks. The issue is not effort, but efficiency and consistency.

With Comply, teams typically reduce manual inspection effort by 60–70% and improve first-time-right accuracy by more than 60%. Reviews move faster, and errors surface earlier.

Comply is AI-powered quality assistant by Esko purpose-built for packaging artwork. Unlike generic document AI, it understands packaging as it actually exists: layered vector files, embedded fonts, structured tables, barcodes, GS1 data, claims, and regulatory statements. These elements are evaluated against brand-defined rules, with accuracy that general-purpose LLMs cannot achieve.

In head-to-head tests, generic models such as Azure, Claude, ChatGPT, and Gemini routinely miss layout-dependent issues that Comply catches instantly.

At its heart, Comply is designed as a human-in-the-loop system. AI performs exhaustive checks, but users remain in full control of approvals.

Teams define once what “good” looks like—fonts, claims, tolerances, code rules—and Comply continuously evaluates every new artwork against those standards. Only exceptions are surfaced, cutting review cycles while strengthening confidence in compliance.

For design and marketing teams, this means faster approvals. For quality and regulatory teams, it means stronger oversight. And for brand owners, it means reduced risk, fewer recalls, and faster speed-to-market, without limiting creativity or introducing workflow friction.

Comply fits seamlessly into the Esko ecosystem. Its integration with WebCenter attributes and Copy Management (CMP) enables dynamic field extraction and fully automated validation, ensuring accuracy at scale across global packaging portfolios.

To demonstrate why Comply works better for packaging documents, Esko conducted a controlled experiment using the same packaging artwork across various LLMs and other AI models.

A detailed copy sheet outlining review requirements was configured directly within Comply. That same checklist was also converted into a text-based prompt and tested against several LLMs. The results, shown in Figure 4, highlights clear differences in accuracy and reliability across approaches.

Prompt used for the comparison:

The artwork included complex elements such as nutrition tables, ingredient panels, symbols, and structured layouts.

The comparison clearly demonstrates that while generic LLMs can interpret portions of text, they frequently miss layout-dependent, structural, and compliance-critical details. Comply, by contrast, evaluates the artwork in its full packaging context—structure, placement, formatting, and intent—not just raw text.

Esko brings decades of expertise in packaging and print innovation, which forms the foundation of its Packaging Intelligence layer. This deep industry knowledge enables Esko to interpret packaging artwork with unmatched precision. By integrating capabilities across packaging structure, graphics, collaboration, and content, Esko is uniquely equipped to identify and inspect the intricate details embedded within artwork files, ensuring accuracy, consistency, and compliance at scale.

Figure 10 shows how Comply understands and extracts every component of a packaging artwork. This proprietary foundation enables significantly faster reviews without compromising accuracy.

While the Comply user experience is intentionally simple and intuitive, the system is powered by a sophisticated multi-model AI architecture under the hood. Each model is trained on packaging-specific data such as variations of labelled elements from artwork files, GS1 barcodes, and multilingual text samples. Together, they form a connected AI ecosystem built to understand packaging the way experts do.

Here’s a closer look at the intelligence stack that drives Comply:

• Text & Font Models
  These models extract text directly from layered PDFs and recognize both standard and custom fonts (if base font is available) even when they’ve been converted to outlines. This ensures brand and regulatory fonts are used correctly across all variants.
• Barcode, Table, and GS1 Models
  Print-aware models detect barcodes, decode content, and validate GTIN, expiry, and lot codes against GS1 schema. A companion segmentation model reads complex nutrition tables and ingredient panels, even under distortions or background textures.
• Trim-Box & Geometry Models
  CAD-aware models calculate die line geometry, bleed, and safety margins, ensuring artwork remains structurally aligned with packaging specifications and is truly print-ready.
• Image and Logo segmentation models
  These models identify logos, symbols, and images within artwork files and verify that the correct versions are being used.
• Semantic Text Aggregation
  Beyond simple extraction, semantic reasoning models group and compare text by meaning, detecting subtle differences between approved copy and actual artwork, such as “low sugar” vs. “reduced sugar” or “SPF 30” vs. “30 SPF.” They also verify the actual meaning of sections. For instance, confirming whether sugar appears as an ingredient versus a standalone declaration. These nuances can make or break regulatory approval.
• Ontology & Rule Engine
  The results from all models flow into a knowledge layer that understands packaging context and maps extracted entities to guidelines, phrases, and country-specific rules. This is where Comply translates AI output into actionable insights for teams.

The entire system runs on the S2 Cloud architecture from Esko, leveraging accelerated inference for speed and scalability, with full auditability and explainability baked in.

The Packaging Intelligence layer doesn’t merely apply foundation models but enhances them for packaging-specific use cases. While generic AI models can interpret standard documents such as invoices, purchase orders, or text heavy files reasonably well, packaging specific AI is built to understand the intricacies of artwork structure and packaging specific compliance guidelines.

Packaging files are a composite of layered vectors, embedded fonts, die shapes, overprints, curved text, and tightly packed content. Text elements are often positioned closely together, sometimes representing entirely different attributes. Graphic components are interwoven with regulatory information and brand messaging to maximize limited space. Traditional document AI was never designed to handle this level of structural and semantic complexity.

This is why Esko has created a differentiated offering, where highly variable, layout-dependent packaging data is accurately recognized by AI models, with minimal to zero customer-specific training.

While generic AI may provide a baseline understanding, the Packaging Intelligence layer developed by Esko is trained on extensive, real-world packaging datasets to deliver far superior accuracy. Its deep domain expertise enables it to interpret the nuanced structure of packaging artwork far more effectively than any off-the-shelf AI solution.

Comply marks an important step in the broader packaging intelligence vision being built at Esko. As packaging ecosystems grow more complex, AI will increasingly act as the connective layer between design, production, compliance, and the complete design-to-print process.

With Comply, we’re taking a decisive first step toward that future. It’s a product of years of domain expertise, AI research, and understanding of customer challenges. It reflects our belief that AI is most valuable when it solves real problems, in real workflows, with real accountability, driving real business outcomes for brands.

To move forward, the packaging intelligence layer will not only drive validation through Comply but also move towards becoming the de-facto co-pilot for the entire packaging lifecycle from brief, creation, update/edits, and validation to distribution.

If you’re ready to lead your organization into the next generation of packaging, now is the moment to act. Connect with Esko to experience Comply and see how Packaging Intelligence can transform your design‑to‑print operations—one workflow, one artwork, and one breakthrough at a time.