Every MES implementation eventually forces a question that looks small in the requirements doc and turns out to be structural: when something goes wrong, how far do you have to reach to contain it? Do you pull a shift’s worth of product, a pallet, a lot, or one specific unit with one specific serial number? That question is really asking you to pick a genealogy model, and it’s one of the few MES design decisions that’s genuinely hard to reverse once production data starts flowing.
Genealogy is the record of what went into a thing and what happened to it — raw materials, sub-assemblies, process parameters, equipment, operators, timestamps. Traceability is the ability to query that record forward (where did this lot go?) and backward (what’s in this unit?). The granularity you choose for genealogy — batch or unit — determines the shape of your database, the cost of your queries, and, when a recall happens, the size of the pile of product you’re pulling off shelves.
Batch genealogy: grouping as the unit of truth
Batch (or lot) genealogy treats a group of material processed together as a single traceable entity. A lot of resin, a mixing tank of API, a heat-treat load of fasteners — all of it inherits one lot ID, and every downstream consumer of that lot inherits the link. In ISA-95 terms, this is typically modeled as material lot relationships: parent lot consumed into child lot, tracked through a genealogy table that’s really just a directed graph of lot-to-lot edges.
The appeal is architectural simplicity. A lot table with parent-child relationships is cheap to store, cheap to index, and fast to query even at large volume, because you’re dealing with thousands or tens of thousands of lot records rather than millions of serialized units. Label printing is simpler — one lot label design covers a whole production run. Data capture is lighter on the shop floor: operators scan a lot barcode a handful of times per shift instead of scanning every individual unit.
The tradeoff is precision, or the lack of it. If a contamination event or a nonconformance is traced to lot 4471, and lot 4471 produced ten thousand units across three shifts, your recall or containment action covers all ten thousand units — even though the actual defect may have originated from a five-minute window on one machine. Batch genealogy can’t tell you that. It wasn’t built to.
Unit-level genealogy: serialization as the unit of truth
Unit-level (serialized) genealogy assigns a unique identifier to each individual item and tracks its specific history — the exact raw material sub-lots consumed, the exact machine and tool that processed it, the exact process parameters at the moment it was made, the exact operator and timestamp. This is the model behind pharmaceutical serialization under DSCSA-style requirements, behind UDI (Unique Device Identification) in FDA regulations and EU MDR for medical devices, and behind the serialized component traceability that automotive OEMs increasingly demand from Tier 1 and Tier 2 suppliers under IATF 16949.
The data model here is fundamentally different, not just bigger. Instead of a lot genealogy graph with thousands of nodes, you have a unit genealogy graph with as many nodes as units produced, each carrying its own edge set back to specific consumed materials and forward to specific downstream assemblies. That’s a different order of magnitude in row counts, a different indexing strategy (you need fast lookups by serial number, not just by lot), and a much heavier data capture burden — every unit gets scanned, printed, or marked at every station where genealogy needs to be recorded.
The payoff is precision in containment. If a defect is traced to a specific unit, or a specific narrow window of process parameters, you can identify exactly which serialized units are affected — potentially a handful, not a lot’s worth. When the FDA or a notified body asks a device manufacturer to trace a specific implant back to its component lots and forward to the specific patient shipment, unit-level genealogy is what makes that query answerable in hours instead of weeks of manual paper-trail reconstruction.
The storage and query tradeoff, made concrete
A batch genealogy schema stays flat and fast almost indefinitely — you’re bounded by number of lots, not number of units, so query performance degrades gracefully as volume grows. A unit-level schema scales with production volume directly, which means you need to think early about partitioning strategy, archiving policy, and whether your MES database (or a dedicated traceability data store alongside it) can sustain the write throughput of scanning every unit at every genealogy-relevant station without becoming the bottleneck on your line.
This is also where OPC UA and MQTT Sparkplug B data flowing off equipment matters: unit-level genealogy typically wants to correlate serialized unit IDs with time-stamped process parameter streams, which means your historian and your MES genealogy tables need a reliable time-and-ID correlation key. Batch genealogy can often get away with coarser, shift-level or lot-level parameter summaries. That correlation architecture is much harder to bolt on after the fact than to design in from day one.
A decision framework tied to regulatory exposure
Don’t pick granularity based on what’s easiest to build. Pick it based on what a regulator, auditor, or customer will actually demand when something goes wrong, and what a recall costs you if your genealogy can’t isolate the problem tightly.
- Regulated medical devices under FDA UDI or EU MDR: unit-level genealogy is effectively mandatory for many device classes. Don’t fight this — design serialized genealogy from the start.
- Pharmaceutical manufacturing: batch genealogy is often the baseline, but serialization requirements at the saleable-unit level are expanding; check whether your product and market require unit-level track-and-trace down to the package.
- Automotive components under IATF 16949 customer-specific requirements: increasingly serialized, especially for safety-critical parts (brakes, airbags, steering) where OEMs want unit-level traceability into their own vehicle build records.
- Food and beverage: largely batch/lot genealogy remains standard practice and is generally sufficient, though high-risk categories face growing pressure for tighter lot definitions and faster traceability response.
- General industrial and consumer goods with low safety exposure: batch genealogy is usually the right call. Unit-level serialization here is often over-engineering that adds cost and line friction without a corresponding reduction in real risk.
Layer the recall-cost math on top of the regulatory answer. Ask: if this product category has a defect event, what’s the realistic size of a batch-level containment action versus a unit-level one? If your lot sizes are small and your process is tightly controlled, the gap between batch and unit granularity may be narrow enough that batch genealogy is defensible even in a moderately regulated environment. If your lots are large and your process has meaningful within-lot variability, that gap is exactly where recall costs balloon — and it’s the strongest argument for paying the upfront cost of serialization.
Why retrofitting is so painful
Moving from batch to unit-level genealogy after go-live isn’t a schema migration — it’s closer to a re-architecture. You need new identification infrastructure (serialization schemes, marking or labeling at every relevant station), new data capture points on the line, a genealogy data model that can handle unit-to-unit and unit-to-lot relationships simultaneously (because raw materials usually stay lot-tracked even when finished goods become serialized), and a rework of every downstream report, recall workflow, and customer-facing traceability query built against the old lot-based model. None of that is impossible, but it’s rarely fast, and it’s almost always a bigger effort than doing it right the first time.
The practical move is to make this decision explicitly, in writing, before configuration starts — driven by where your product sits regulatorily and what a real containment event would cost you in scope, not by which model the MES vendor’s default template happens to ship with.
This article was written with the assistance of artificial intelligence. While we aim for accuracy, the information may be incomplete, out of date, or incorrect, and should be independently verified before you rely on it for any decision. It is provided for general information only and does not constitute professional advice.
