Pharma plants are awash in data. There are myriad sources: MES for execution status, LIMS (QC results), process historians (equipment conditions), QMS (quality/deviations) to name just a few, but release decisions still stall when those sources disagree or lack an owner.
To deal with the chaos, eschbach’s visual factory centers daily work on SQDCP boards (the lean tool referring to Safety, Quality, Delivery, Cost, People) that display target-versus-actual, convert gaps into owned actions with due dates. “It’s more than just visualizing those graphs,” emphasizes CEO Andreas Eschbach. “It’s getting the data into the hand of the people, getting the right information to the right people at the right time, and making them work with it.” It can escalate issues from tier-1 shift huddles to tier-5 leadership so problems don’t linger. Tiles show simple trends. Deviations can be turned into actions with an owner and deadline. Notifications can go, for instance, to operators’ mobile devices or a display. From there, QA and maintenance can acknowledge or hand off to the next shift without losing context.
See it in 1 second, act in 60
Eschbach describes the operating rule behind the approach as “1-10-60.” That is: one second to see status, ten to locate the issue, and sixty to decide on countermeasures. If a screen takes longer than that to parse, teams can simplify the layout or reduce KPIs.
“A second to see if you’re winning or losing, 10 seconds to grasp the issue, 60 seconds to decide the measures.” —Andreas Eschbach
Andreas Eschbach
Sites that avoid dashboard sprawl can set clear templates, cap metrics per board, and name a board owner accountable for its clarity and upkeep. Under the hood, the boards can draw on existing systems rather than requiring new data pipes. Process historians, for instance, can supply machine states and downtime codes for OEE and loss accounting; MES adds batch context and release gates; LIMS/QMS provide test results and deviations. Most sites already have these systems running. The challenge isn’t getting new data but connecting what’s already there.
Where IIoT is present, signals may arrive via MQTT or OPC UA, but the historian typically remains the system of record. Time-stamped tiles link back to source for drill-downs, and actions carry owners and timestamps to support auditability in regulated environments. The prerequisite is data hygiene: common equipment IDs and batch keys across systems. Factories that struggle with quality or throughput problems don’t tend to be short on data; they tend to be short on alignment. The systems are often siloed, and people fill the gaps with, say, Excel and SharePoint.
Where AI fits
eschbach’s AI layer, SAMI (Shiftconnector Artificial Manufacturing Intelligence), focuses on unstructured text in shift logs and handovers. The remit is explicit: link real problems to recorded solutions; don’t invent them. That is, SAMI is designed to connect known fixes to current issues to documented mitigations across sites (and languages) to avoid AI hallucinations.
In practice, SAMI does a few things: AI shift summaries posted to boards and delivered as notifications so teams see the day’s exceptions rather than routine noise; a solution finder that connects current issues to past, recorded mitigations; and a mobile chat interface so frontline staff can retrieve context in their language, on the spot. As Eschbach put it, sites “usually see … almost 100 records created in 24 hours,” with AI summarizing that stream down to what actually matters, filtering routine activity and highlighting exceptions.
This matters for tacit knowledge capture. An operator might never write “viscosity” in a log, but they’ll write “it feels like honey” or “it’s like cream.” SAMI surfaces those semantics so anecdotal notes become a usable knowledge base.
Cutting latency of batch release
Batch release is often a race against latency: waiting on tests, deviations, and signatures across LIMS, QMS and MES. Visualizing status across these systems reduces wait time by making blockers explicit and assignable. A control-tower view helps departments see interdependencies and act on them.
As Eschbach put it, pharma clients focus on how they can help with a faster time to patient by optimizing their batch release so that everyone knows what matters at the moment to release a certain batch. The success metric is behavioral: faster handovers, clearer ownership, and fewer ping-pong escalations rather than just more charts.havioral: faster handovers, clearer ownership, and fewer ping-pong escalations rather than just more charts.
Keeping people in the loop
The philosophy is human-in-the-loop: offload retrieval and correlation so people can focus on judgment and GMP-compliant action. Human-in-the-loop will be with pharma for a long time. The aim is to take the robot out of the human.
To keep leadership engaged with the floor, eschbach even includes nudges such as a GEMBA walk randomizer that selects the next topic or area to review, which is often safety, during walks.
Agent-to-agent communication is accelerating. For data transport on the shop floor you’ll see MQTT/OPC UA; for agents, Eschbach notes recent work with the MCP protocol, originally developed by Anthropic. “There is such a powerful world of agent-to-agent communication coming which I would assume is mind-blowing,” Eschbach concluded.
Filed Under: Drug Discovery and Development
