中文Cerm has launched Lexis, a cloud-based AI agent built to automate one of the most error-prone chores in label and packaging printing: turning customer order emails into validated sales orders. The company will stage its public launch at Loupe Americas 2026, taking place 15-17 September at the Donald E Stephens Convention Center in Rosemont, Illinois.
Lexis reads incoming order emails and their attachments, identifies the customer and the order details, matches the content in real time with MIS data, and prepares the information as a sales order in Cerm ready for a staff member to review and confirm. Cerm first introduced Lexis to customers during its User Group Day 2026 in Paris and Dallas and has since gone live with 20 customers. At Loupe, it will be available worldwide as part of the Cerm Extensions portfolio.
The mechanics are designed to remove manual re-keying. Lexis extracts the full content of an email and any attachments and transforms it into structured business data. Using a real-time API connection to the Cerm database, it suggests the most likely customer record; if the sender’s address matches a single customer it confirms automatically, while multiple possible matches prompt a staff member to choose. It then pulls the purchase order number, products, quantities and delivery information from the email and attaches a confidence score to what it found.
Lexis is more than a simple order-creation tool. Cerm describes it as a customer-service support agent that validates prices, creates new products, manages attached artwork, creates new delivery addresses in the MIS, validates PO numbers and performs other tasks that a customer service representative would traditionally do by hand. A human must still review and confirm the data before a sales order is created; nothing enters Cerm without a person signing off. Where a company chooses, Lexis can also send the order acknowledgement back to the customer automatically once the order is entered.
Customers can customise Lexis through a rule-based model based on their own specifications. Rules can account for a specific customer’s typical order format, communication patterns, or additional requirements that need incorporating for a particular account. That configurability is important in label and packaging, where every brand owner seems to have a slightly different PO convention.
Onpack, an early adopter led by managing director Michael Nankervis, has already processed 260 orders through Lexis covering more than 3,500 products, from simple reorders to complex multi-line jobs. Today, 80 per cent of the company’s business flows through Lexis. ‘For our larger orders, what used to take two to three hours to key in by hand now takes under an hour,’ Nankervis said, adding that the constant double-checking inherent in manual entry is no longer part of the process. His view is that Lexis is not designed to replace customer service teams but to remove repetitive administrative work so they can focus on customers.
The launch sits within a wider movement toward agentic AI in print production software. For years, MIS and workflow vendors promised automation, but much of it relied on rigid rules and clean data; order entry from free-form customer emails remained stubbornly manual because email is unstructured, inconsistent and brand-specific. Lexis attacks exactly that gap, using language models to interpret intent and an API link to validate against live MIS data. The human-in-the-loop design is deliberate and, from a risk standpoint, necessary: a misread PO in label printing can trigger the wrong substrate, quantity or delivery address, so confirmation before entry is non-negotiable. What makes the Onpack result compelling is the volume, 260 orders, 3,500 products, 80 per cent of business flowing through, because it demonstrates the system handles complexity, not just simple reorders. The roadmap toward RFQ and ASN processing suggests Cerm sees Lexis as a platform for email-based workflow generally, not a one-trick order tool. For an industry short on customer-service labour and heavy on repetitive data entry, the appeal is obvious. The larger question is trust and data security: connecting an AI agent to a live production database demands rigorous access controls, and conservative print businesses will want assurances before letting software read customer correspondence. Cerm’s staged rollout, with 20 pilots first, is a measured way to build that confidence.
The implications extend beyond order entry. If AI agents can reliably read customer intent from email, the next logical step is end-to-end email workflow automation: RFQs that become quotes, advance shipping notices that update inventory, and service requests that open tickets, all without a human copying data between systems. Cerm’s roadmap signals exactly this expansion. The caution required is real, because the failure mode is not a slow manual process but a confident, automated mistake, a misread PO that triggers production of the wrong job at volume. That is why the human confirmation gate, and the confidence scoring that tells staff where to look, are the responsible design. For label and packaging converters, where a wrong substrate or quantity can mean thousands of metres of scrap, that gate is the difference between helpful automation and expensive chaos. The early Onpack numbers, 80 per cent of business through the agent, suggest the trust can be earned when the system proves accurate. As more MIS vendors add similar agents, the differentiator will be not the AI itself but the guardrails and the quality of the underlying data.
Source: based on reporting from Labels & Labeling (labelsandlabeling.com).
