AI GMP WorkshopWho Decides, Who Stops?
The AI GMP workshop is a working format for teams from production, QC, QA, validation and IT. Working from a GMP case in which an AI model takes part in the decision, they define which control takes effect before the decision, which role owns that control and which evidence is shown to an inspector. The format premiered at Pharma Smart Manufacturing 2026 in Copenhagen.
Four results at the end of the workshop.
Every table leaves behind its flipchart. We also evaluate the fields that stayed empty.
- 1Control overview per process
For every process, one sheet records the controls around the model's output, the roles that own them and the state of the evidence.
- 2Gap list of unassigned responsibilities
Every control without an owner and every commitment without a record goes on the list as an open task.
- 3Criticality classified
How critical a given use is depends on where it sits in the chain from data to product to patient.
- 4A decision paper for your management
It records which role the teams propose for the model lifecycle and which gaps to close first. The one-day version adds photo documentation and a transcript per flipchart.
30,000 vials pass the camera. A human looks at 315.
An inspection machine with an AI camera checks a batch for particles and cracks. Qualified visual inspectors draw the sample under ISO 2859-1, 315 vials at this lot size. Who is responsible for the model was left open at all five tables in Copenhagen.
In every workshop case, the output of a model determines what a human gets to see. Which role owns the model lifecycle and who stops the model is for your company to define. The four prepared cases are fictional.
The release of a batch of 30,000 vials rests on the AI camera and a sample of 315.
Only the chromatograms flagged by a supplier's model go to a second analyst.
QA has two days to review a PQR that a generative model drafted from MES, LIMS and QMS.
A model generates the test cases for an MES update and evaluates the results itself.
In the one-day version your own processes replace the four cases. Candidates include visual inspection, chromatogram triage, PQR drafting, test generation, deviation classification and trend analysis.
How does the AI GMP workshop run?
The AI GMP workshop has three stages: an introduction to Chapter 4, Annex 11 and Annex 22, table work on one case with an unannounced disruption and a debrief along the chain from data to patient. The short version takes 90 minutes, and the one-day version takes a full day with your own processes. Participants need no GAMP background.
- 1
Orientation
We explain the three drafts published on 7 July 2025. As an analogy we use a delivery from our keynote in Shanghai in 2026. Cargo, truck and autopilot stand for the three rule books.
- 2
Table work
Each table works on a case from its own discipline. The teams record what is checked before the decision, who is responsible for that check and what evidence exists for it. Partway through, we introduce an unannounced disruption from day-to-day operations at each table.
- 3
Debrief
We rank every answer by how close it sits to the patient. This produces the control overview, the gap list and the decision paper for your management.
In the afternoon of the one-day version, we transfer the results into the control map of the readiness assessment. Flipcharts, sticky notes and easels are the core of the format, which is why we only work on site.
Four questions for your plant.
Answering no to any of them shows where your first case begins.
What five tables found.
On 14 October 2026, around 80 participants worked at five tables at Pharma Smart Manufacturing 2026 in Copenhagen on cases from production, QA and validation. Three findings recurred at almost every table.
The organizer included two results of the workshop in the conference synthesis: the sentence “Quality is not a system property. It is an architectural one.” and the rule “Human-in-the-loop is a design decision” (Fleming, Pharma Smart Manufacturing 2026, closing plenary, 14 October).
All five tables rated their case as critical. For the camera, that is correct. For the test generator, the groups transferred the criticality of the MES process under test to the tool that generates the test cases.
Who runs the workshop.
QFINITY developed the format and leads the workshop. A consultant from our team guides the table work. They know the authorities' drafts from their work in the GAMP committees. They judge what inspectors expect from first-hand audit practice and from years of hands-on work at and with our clients.
Case cards, disruption cards and the evaluation grid are QFINITY working material and remain so. Your teams take the results home. You can read how the premiere in Copenhagen went in our report on Pharma Smart Manufacturing 2026.
Frequently asked questions about the workshop.
For the 90-minute short version the four prepared cases from production, QC, QA and validation are enough. In the one-day version the tables work with two to four case cards that we prepare for your processes beforehand, each based on an interview with the process owner.
Yes. The tables then work with a planned use case or with one of the four cases. The questions about data criticality, owners and records apply even before the first model is in operation.
No, and it does not promise that. It makes the gaps visible and names where ownership is missing. Teams that need evidence by a deadline start with the readiness assessment. We recommend the workshop when production, QA and IT see responsibility for the model differently: it delivers your teams' answers, while the readiness assessment provides evidence of where you stand.
The effort depends on the version, the number of case cards and the number of participants. Preparation takes half a day per case card, plus the workshop day and the debrief. After the intro call you receive a proposal with the estimated effort and possible dates.
The workshop runs in English or German, and the slides are available in both languages. At your site, 12 to 40 people work at three to eight tables. With 20 or more people, pairs of tables work on the same case and compare their answers.
From the workshop to the evidence.
Which process will be your first case?
In a free intro call (about 30 minutes) we clarify which of your processes make suitable cases and which version suits you. Within two working days we send you a scope estimate and possible dates.
Discuss the workshop now
