On November 17, Aphena, a U.S. contract pharmaceutical and packaging company, announced that its QAD Adaptive ERP went live. Finance, manufacturing, and supply chain were moved to the cloud, while Redzone's frontline workstations were also rolled out on the shop floor, allowing team leaders and operators to see output, downtime, and quality inspection data on tablets and phones.
In the same week, Germany-based Körber’s lightweight MES for small and medium-sized pharmaceutical companies—PAS-X Neo—completed its first version and will be officially available in November. One line in its official messaging stands out: the goal is to go live within two months.
Put these two news stories together, and you will find that what they answer is not “whose AI is smarter,” but another more plain question: in a regulated industry, on what grounds does one dare to launch a system within two months?
1. First, translate 'compliance' into plain language
Many people think that implementing systems in pharmaceutical factories is difficult because there are many features. That's not it. Many features are just troublesome; compliance is the real barrier.
The original wording of GMP is of course very long, but when it comes down to the system, it is just one sentence: Every judgment about this batch of goods must have a person, a basis, and a time, and at the audit site two years later, you can prove that it has not been altered.
In the industry, this is broken down into nine words, called ALCOA+: attributable, legible, contemporaneous, original, accurate, complete, consistent, enduring, and available. In plain terms, it means—who did it, whether it is written clearly, whether it was recorded at the time, whether it is first-hand, whether the calculation is correct, whether any pages are missing, whether it is consistent throughout, whether it can be preserved, and whether it can be retrieved.
| Compliance Requirements | Things that must exist in the system | Common perfunctory methods |
|---|---|---|
| Attributable | Each record has a unique person, and this person cannot be a shared account | One "inspector" account, used in rotation by three people |
| Sync | The timestamp comes from the system, not a date field filled in after the fact | Allow manual modification of "Inspection Date" |
| Original | Retain the original data and attachments; do not store only a conclusion. | Only enter "Pass"; do not enter the measured value. |
| Durable, accessible | Records cannot be edited; they can only be appended or reversed; they remain searchable after archiving. | Anyone can edit it, and after editing only the “last editor” is retained. |
So "compliance" is not a feature checklist; it is an evidence checklist. When you ask at a vendor selection meeting, "Do you have a quality management module?" the answer is always yes; only when you ask, "Can a completed inspection form be modified?" do you see the vendor's expression change.
II. Two months isn't about fast implementation; it's about a small scope
For Körber's lightweight product line, the official design premises are four: cloud-native SaaS, standardized templates, minimal configuration, and AI assistance. Plus one most critical point—pre-validated components, used to reduce validation effort.
"Pharmaceutical digitalization should not be an all-or-nothing decision." — Pierrick Lebigre, Vice President of Pharma Business Software Products at Körber
That is a very polite way of putting it; the actual meaning is: don't expect to get there in one step. PAS-X Neo is positioned to "get on board first"—a single site, standard processes, and sufficient GMP functionality, replacing paper batch records with electronic batch records first; once production volume, product types, and sites increase, then migrate to the flagship PAS-X MES.
Traditional project
All modules, all processes, all sites — verification documentation swells right along with the scope
Starter Pack
Single site, standard template, pre-validated components, live in two months, with an upgrade path
Two months can be accounted for: templates compress the time for process design, pre-validated components compress the time for documentation writing, and minimal configuration compresses the time for integration testing. The only things that cannot be compressed are two—data cleansing and personnel training; both take time no matter who does them.
Validation here deserves a bit more elaboration. Validation for an MES typically consists of three documents plus one matrix: a user requirements specification, installation qualification, operational qualification, and a traceability matrix mapping each requirement to test cases. The trouble is that the validation workload is disproportionate to the configuration workload—change ten lines of configuration, and the validation documentation may need twenty pages rewritten. So the essence of 'pre-validated components' is to write and sign off the validation documentation for common processes in advance, so that when customers use them, they only need to confirm 'my process is consistent with this template.'
This also explains why such products require customers to accept the standard process. If you don't accept it, pre-validation doesn't hold; if pre-validation doesn't hold, the two months don't hold either. Product and process are one thing here, not two.
The cost is also clear. First, processes must conform to the template: if your special practice is not in the template, you either change the process or wait for an upgrade. Second, exception handling must follow the standard path: how deviations are opened, who approves them, and how they are closed—the product has already defined the framework. Third, in two or three years there is a high probability that you will need to migrate to the flagship edition once, and the migration cost must go into the budget now—don't wait until that day to bring it up.
To judge whether a "lightweight version" is worth adopting, there's a simple method: plot the three growth curves for the next three years—production volume, product variety, and sites—and see which year the template can hold out until. If it can't last two years, that means what you need isn't a starter package, but a proper project done in two phases—go live with a minimal scope first, then expand module by module; don't aim for all modules right from the start.
“Go live in two months” should be broken down into three sentences in the contract: which sites, which varieties, and which processes will go live in two months; what to do if the scope is exceeded; and who signs the validation documents. If these three sentences are not written clearly, two months will become six months.
III. What Role Does AI Play in Batch Records?
There is a detail worth noting in the Aphena news: the press release mentions ChampionAI's 'potential' — the CFO said he saw the possibility of it helping the team make decisions faster, and QAD|Redzone's CEO said it could give shop-floor teams advice and reduce manual work. But throughout, it does not say which capabilities were enabled, nor is there a separate go-live time; even the switchover date and site cadence were not disclosed.
This level is right. In regulated settings, AI's disclosure language and its actual permissions must both be kept restrained. The reason is simple: once AI's recommendation enters the batch record, it becomes part of the validation scope. Auditors will ask three questions—
- What is the input for this suggestion, and which table and version does it come from;
- After the model is switched to a new version, do the historical conclusions still hold;
- Who adopted it, and where is the reason for adoption written?
If it cannot answer all three questions completely, AI can only stay in a "reference" role and cannot enter a "decision" role. This is also why the order in which pharmaceutical companies implement AI usually starts with areas such as documentation, training, and production scheduling—where "if it's wrong, it can still be corrected"—rather than with batch release and decision-making—where "if it's wrong, it leads to a recall."
One lesson: in the GMP context, “what the AI says” is not evidence; “a person signed to adopt the AI’s recommendation” is evidence. So what the system must retain is not the model’s output, but the moment that person pressed the confirmation key.
There is another category of records that auditors love to dig through on site but that is often missing from the system: deviations and changes. When a deviation occurs, how it was opened, who assessed the impact, what measures were taken, and whether it made it back into the batch record. If the system can only record "pass/fail," deviations can only live in Excel and email, and the validation package breaks down at this point.
IV. Back to Odoo: Six Anchor Points
If the client is not a pharmaceutical company but instead a semiconductor, medical device, or automotive parts manufacturer, the same logic still holds—only the names of the regulations change to IATF 16949, ISO 13485, and customer audits. When delivering to such clients with Odoo, there are six areas worth designing in advance.
| Requirements | What does it correspond to in Odoo? | Places that are easy to miss |
|---|---|---|
| Every judgment must have someone | Inspector, measured values, and attached photos of the quality check sheet (quality.check) | Only enter 'Pass/Fail'; do not enter actual measured values; shared account |
| Changes must leave a trace | mail.thread field tracking, who changed the batch status and when | Key fields don't have tracking enabled, so changes leave no record in the system |
| Batch Genealogy | Upstream and downstream chain formed by batch/serial numbers (stock.lot) and inventory move line details | Only forward traceability was implemented; when a customer complains and you need to trace raw materials backward, you're at a loss. |
| Signature and Time | Electronic signature module, approval workflow; time is subject to the server | Use "write 'reviewed' in the remarks" instead of a signature. |
| Records cannot be changed | Completed inspections are locked and can only be appended or reversed; accounting vouchers have a hash chain | Production and quality inspection records have no hash chain and by default do not prevent editing. |
| Permissions and Environment | Record rules plus user groups; keep on file the configuration differences between the validation environment and the production environment. | To meet the deadline, the implementation account was given full permissions, which were not revoked after go-live. |
The fifth line is where things are most likely to go wrong. Odoo's accounting module has entry hashing and a lock date, so it can achieve "cannot be changed"; production and quality inspection records do not have this mechanism, and by default, users with permissions can still edit a completed inspection sheet. To meet audit requirements, you can only add constraints to the model yourself: once the status reaches "Completed", writing is prohibited; if a change is needed, use "reversal + new creation", and write this rule into the implementation document.
This is not a defect of Odoo, but a difference in product positioning—it first addresses 'keeping business operations running smoothly,' and then lets people add constraints according to industry requirements. The problem is that many projects simply never take this step, and only remember when they are at the audit site—by then the data is already dirty.
V. Three Sentences for Two Kinds of People
For the client making the selection. Add a column "verification cost" to the scoring sheet, with a weight no lower than feature coverage. Ask three questions: how many pre-validated components there are, who signed off, and whether the validation documentation can be exported for audit review. Vendors that can answer these clearly will most likely also be able to answer clearly about their go-live timeline.
For the implementing party responsible for delivery. Write the “verification package” into the contract as a formal deliverable: requirements traceability matrix, verification plan template, change control records, and training records. If these four things are not done, even if the project passes acceptance, it will still fail the customer audit, and ultimately they will come back to you.
A sentence that applies to both sides. Before going live, run a mock audit: pick five batches at random, trace from the finished product batch number back to the raw material batch number, and then back to the supplier and inspection records. Only if it can be completed within half an hour and every step has a person assigned can the system be considered truly established.
A system going live only shows that it 'works'; only passing an audit shows that it is 'trustworthy.' What separates these two things is not functionality, but evidence.
Sources: QAD | Redzone press release (2026-11-17, Aphena Pharma Solutions go-live announcement), Körber press release (2026-05-18 announcing PAS-X Neo, first version available in November 2026). Vendor press releases are self-reported and not accompanied by third-party audits; Odoo-related descriptions are based on general implementation experience, and actual behavior should be subject to testing with the version in use.
Zhongdi Information · Enterprise Management Digitalization Observations · Cases in this article are cited from public reports
