Pharma and manufacturing
Production, laboratory and quality data in one place. The batch decision stays with the Qualified Person.
Context
Manufacturers of medicines, supplements and cosmetics work under several regimes at once: pharmaceutical GMP, food law and the cosmetics regulation, often within a single site. Batch data sits in ERP, MES, LIMS and QMS, and some records are still on paper. Each of these systems is validated separately, so combining data across them is done by hand.
We start with a data layer that reads source systems in read-only mode and does not disturb their validated state. On it we build reporting, applications for production and quality, and AI solutions. Validation documentation follows GAMP 5, and in GMP-critical use cases we apply static, deterministic models, in line with the direction of the draft Annex 22.
Challenges in the industry
Manual documentation review before batch release
Before certification, the Qualified Person confirms the batch record, laboratory results, the status of deviations and changes, and the integrity of the audit trail. The data sits in MES, LIMS and QMS, and partly in paper documents, so the summary is compiled by hand while the batch waits in quarantine.
OEE decline with no clear cause
The shift report shows that OEE is falling but loses the cause: micro-stoppages, changeovers, cleaning between batches, waiting for raw material release or in-process control. Without that breakdown it is unclear whether time is lost at the machine, in the procedure or in quality.
Deviations and CAPA handled one by one
Deviations, OOS results and CAPA actions are processed ticket by ticket. Similar problems return on other lines and at other sites because root causes are not compared over time, and the effectiveness check on a CAPA is often a formality at task closure.
Planning with a long manufacturing cycle
Manufacturing, testing and release of a batch take weeks, while demand moves with the infection season, promotions and wholesaler decisions. Too little stock means shortages in pharmacies, too much means products with a short shelf life.
Where we deploy software and AI
Support for batch record review
Validated rules compare data from MES, LIMS and QMS against the specification and the batch record. They flag missing signatures, open deviations, results close to limits and audit trail entries that need explanation. The Qualified Person receives a summary with links to source records and makes the decision.
OEE and stoppage cause analysis
We combine data from PLC, SCADA and the historian with MES and shift reports. The report separates availability, performance and quality losses by product, shift, changeover type and cleaning between batches.
Early detection of failure risk
A model built on sensor data and service history points to machines with rising failure risk and states the evidence: vibration, temperature, power draw. The alert goes to maintenance. The calibration and inspection schedule for qualified equipment changes only through change control with QA involved.
Demand forecast for batch planning
The forecast accounts for seasonality, promotions, sales data and illness periods. It enters the S&OP process as a proposal with visible assumptions and an uncertainty range, which the planner adjusts before batches are scheduled.
Deviation trends and root causes
Text analysis of deviation descriptions finds reports with a similar course across different lines, raw material suppliers and shifts. This is a non-critical use case: the model proposes links, and the quality department decides on a common cause and on CAPA.
A shared data layer for ERP, MES, LIMS and QMS
Every value in the data layer carries information on which system, when and which record it came from. A report leads to the source record, and the originals remain unchanged in the systems where they were created.
Regulatory requirements are taken into account from the analysis stage
Annex 11 and its revision
Annex 11 from 2011 is in force. The draft of the new version, consulted on in 2025, tightens requirements on system suppliers, account management and audit trail review before batch release. We already build new systems to the draft: unique accounts, a non-editable audit trail, and a contract and documentation ready for a supplier audit.
Data integrity: ALCOA+
Data must be attributable, legible, contemporaneous, original and accurate, as well as complete, consistent, enduring and available. The audit trail records who changed what, when and why. The analytical layer does not alter source data.
Validation according to GAMP 5
We base validation on the second edition of GAMP 5: software categorisation, risk assessment with the client's QA, and critical thinking instead of testing for its own sake. We deliver specifications, tests and reports, and the client's quality department approves the validation.
AI in GMP: Annex 22 and the AI Act
The draft Annex 22 allows static models with deterministic output in critical GMP use cases, with independent test data and drift monitoring. Generative models stay out of quality decisions. We classify each use case separately under the AI Act: most tools in a plant are not high-risk systems.
How we start
Workshop with production and quality
We meet with QA, the Qualified Person, IT and OT, maintenance and supply chain. We map processes and systems and divide areas into GxP and non-GxP.
Impact assessment and requirements
For each use case we prepare a GxP impact assessment, a preliminary user requirements specification and a GAMP 5 category. This sets the scope of validation and human oversight.
Pilot outside the GxP process
The first project runs in an area with no impact on product quality, for example stoppage analysis or demand forecasting. We use it to test data quality and cooperation with QA before moving into critical areas.
Validation, rollout and maintenance
Solutions that affect quality are validated according to the client's procedures. We release subsequent versions under change control, with the option to test before deployment, and models are covered by drift monitoring and periodic review.
Related services
Frequently asked questions
No. A batch is certified by the Qualified Person (QP), who is responsible for confirming that it was manufactured and controlled in accordance with requirements. The system prepares a documentation summary and points out discrepancies, but the decision and the responsibility remain with a person.