Lab automation is the use of robotics, instruments and software to carry out laboratory tasks that people would otherwise perform by hand. It covers liquid handling, sample and plate movement, instrument control, scheduling and data transfer. Lab automation software coordinates those tasks across connected instruments and passes results between systems.
What lab automation software and hardware cover today
Lab automation is not one product category. It is a set of lab automation technologies that solve different problems, bought at different times, usually by different people. Most labs run several of them side by side without ever having planned a single system.
Liquid handling
The largest and oldest category. Automated liquid handlers pipette, dilute, aliquot, add reagents and replicate plates, from single-channel benchtop units to 96 and 384-channel heads on a full deck. The purpose is not only speed. A liquid handler removes pipetting variance between operators and between days, which is why it is often the first piece of lab automation equipment a regulated lab buys.
Plate and sample handling
Robotic arms, plate hotels, stackers and carousels move consumables between positions. Around them sit decappers, cappers, sealers, peelers, tube sorters and centrifuges with automated access. This layer is what turns a set of individual instruments into something that can run unattended past the end of a shift.
Scheduling and orchestration
This is where most lab automation software sits. A scheduler allocates instrument time, sequences steps across shared hardware, resolves collisions when two runs need the same resource and handles error recovery when a step fails mid-run. Orchestration software is also what allows a workcell to be reconfigured for a new assay without rebuilding the physical setup.
Instrument integration and data movement
Automated laboratory technologies are only as useful as their ability to talk to each other. Integration runs on instrument drivers and, increasingly, on open standards: SiLA 2 for instrument control, OPC UA LADS for device interfaces and Allotrope ADF or ASM for analytical data. A lab automation software system typically also connects upward to a LIMS or ELN so that results, sample identifiers and run parameters move without manual re-entry.
Task-level automation versus total lab automation
The distinction matters more than any vendor comparison. Task-level automation, sometimes called an island of automation, automates one step or one assay: a liquid handler, a plate reader, a single workcell. Total lab automation connects the majority of the sample path, often on a physical track, so that a sample moves from receipt to result with limited human handling. Total lab automation is common in high-volume clinical chemistry and pathology, where sample types are consistent and volumes justify the capital. Most other labs, including the majority of regulated labs running varied protocols, operate islands.
That last point is the practical reality of automation in the lab. A regulated lab is rarely fully automated or fully manual. It is automated steps with people in between, and that shape determines where the documentation risk sits.
Where laboratory workflow automation stops
Automating an action does not create evidence that the action happened correctly.
An automated step produces instrument data: parameters, timestamps, error codes, results. That is a record of what the instrument did, not of the workflow around it.
Between and around every automated run there are manual steps that no one automates because they are too varied to justify it. Reagent preparation and lot verification. Deck layout and labware placement. Sample receipt, accessioning and visual condition checks. Labeling. Transfers between islands of automation. Judgment calls when a run flags an exception, and the decision about whether to continue, repeat or discard.
They are also the steps most likely to introduce an error that matters.
The record is still self-reported
Then there is the record of those steps. Regulated labs are not undocumented. There are batch records, worksheets, witnessed signatures, LIMS entries and at many pharma sites an MES enforcing the sequence. Contemporaneous recording is not an aspiration in these labs, it is a requirement, and a lab recording after the fact has a finding rather than a habit.
The gap is not an absence of documentation. It is what the documentation is made of.
Every one of those records is self-reported. A person performs the step, then attests to the step, and the attestation is the evidence. The system holds what the operator entered, not what the operator did. Where the two diverge, through a genuine error, a habit that drifted from the SOP or an entry made twenty minutes later from recall, nothing in the record shows the divergence. Instrument logs cannot close it, because the steps in question happened outside the instrument.
That is a structural limit, not a discipline problem. A workflow can be well engineered, staffed by careful people and fully compliant on paper, and its evidence still rests on the account given by the person whose work is being evidenced.
In a regulated lab, that is not an efficiency problem. It is where compliance risk lives.
What the gap looks like in four regulated workflows
The gap is the same everywhere. What it costs depends on what the lab is accountable for.
Pharma QC and QA
The visible cost is deviations. A deviation is raised, and the investigation depends on reconstructing what actually happened at a step performed weeks earlier. If the contemporaneous record is thin, the investigation widens, the root cause is recorded as indeterminate and the corrective action becomes another procedure and another signature. Batch record review absorbs the same weakness: reviewers spend their time chasing missing entries rather than assessing quality. Ahead of an inspection this compounds into the familiar scramble: months of documentation reassembled under time pressure, gaps found late.
IVF
Embryology protocols run long, across many discrete steps, often over several days, on material that belongs to one identified patient and cannot be re-created. Double-witnessing exists because a mis-identification cannot be corrected afterward. But a witness check is a point-in-time confirmation recorded by hand, and it evidences that two people signed rather than what either of them observed. When the lab has to demonstrate later that every step was performed as required, the completeness of the form is the whole case.
Genomics
Clinical genomics workflows run across multiple days and multiple instruments, with library prep, normalization and sequencing separated by hand-off points. A single mis-set parameter or a swapped index invalidates the run, and the invalidation is often discovered downstream at analysis. Finding which step failed, and proving it, needs step-by-step detail the instrument logs do not hold.
Tox and forensics
The output has to survive challenge. Chain of custody is a documentation problem before it is anything else, and defensibility rests on whether the lab can show, in order, who handled the sample, when, under what conditions and with what result. A record assembled after the fact is the hardest kind to defend.
Four segments, four consequences, one cause: the record of the workflow is produced separately from the workflow.
What closing the gap requires
These criteria apply to any vendor in this space. Two of them are places where documentation tools, the visual ones included, have real limits. Ask about those first.
1. Capture that does not depend on recall
The record has to be created at the moment the step happens, not reconstructed from it. That means capturing the human actions in the workflow, not only the instrument outputs, and doing so without asking the scientist to stop and type. Any approach that adds a documentation task to the bench will be completed inconsistently, because it is competing with the actual work.
2. Attestation that ties an action to a person, a time and a material
Capture on its own is data. What QA needs is proof: a specific action, performed by a specific person, on a specific sample or lot, at a specific time, in a form that cannot be quietly edited afterward. Under GxP data integrity expectations these are the ALCOA attributes. Attributable and contemporaneous are the two that self-reported documentation is structurally weakest on, because both depend on the operator rather than on the system.
3. Output an auditor will accept
A record that only makes sense inside the system that produced it has limited value at inspection. The test is whether an inspector can be handed the output and follow it without a guided tour. For ISO 17025 labs that means traceability of records and equipment through the full workflow. For CAP and CLIA labs it means demonstrable adherence to documented procedure, including personnel competency at the step level.
4. Coverage across islands, not one instrument
Because most regulated labs run task-level automation rather than total lab automation, a solution tied to a single instrument or a single vendor’s stack will document the part of the workflow that was already documented well and miss the parts that were not.
5. A validation path, not a claim that validation is unnecessary
Any system producing GxP-relevant records is itself subject to computer system validation and falls within Part 11 and Annex 11 expectations. A vendor saying their product needs none is either misreading the regulation or moving the risk onto the lab. The useful question is how much of the validation the vendor has already done: qualification documentation, a specified audit trail, change control and a release model that does not reopen the validated state every quarter. What should not change is the bench routine. Rewriting protocols or retraining staff on a new way of working costs more than the problem is worth.
6. We should talk about the cameras
The first question we get in a lab is rarely about detection accuracy. It’s whether this is a quality tool or a way to watch staff. Here’s the answer: Lumi’s cameras sit above the bench and point down at the work surface, faces are rarely in shot and can be blurred, and the model is trained on your own documented procedure, so what it flags is a departure from the path, not a person. Runs are still attributable, the way a signed batch record is. The operator is named from their login, because a record that can’t say who did the work isn’t worth keeping. What the record doesn’t do is rate how well they did it. The footage and the procedures are yours, access is limited to authorized users with their own logins, and recordings are kept for one year. What we use across customers is aggregated, anonymized deviation and accuracy data. If your team wants to see exactly what’s captured before you commit, we’ll show them in a pilot in your lab.
7. Limits the vendor names before you ask
Documentation tools evidence what was observable. They are not evidence of intent, of a reagent’s true concentration or of anything out of view, and they do not remove the need for method validation, calibration records or instrument qualification. A vendor claiming to document everything is describing a product that does not exist.
These criteria separate two jobs. Orchestration tools run the workflow. Documentation and attestation tools evidence it. Most labs have bought well for the first and have not yet treated the second as a purchase at all.
Where Lumi fits
Lumi is a Visual AI Copilot for the lab. It observes the steps a scientist performs at the bench and turns them into a structured record of what was done, when and by whom, with attestations meant to stand as proof rather than as a second-hand account of the work. The scientist does not stop to type, and the protocol does not change.
The detection layer is computer vision, not generative AI. Nothing in the record is generated or inferred. Every attestation traces back to the observation that produced it, so a reviewer can check the evidence rather than take the output on trust.
A Lumi install is a set of cameras positioned above the bench, with the count and placement worked out lab by lab so every step of the procedure is properly in view. Video is processed in the cloud rather than on hardware in the lab. Each run produces an attestation record: the procedure and version followed, the sample or batch identifier, the operator, named from their signed-in account, a timestamp for each step, and any yellow or red flags with the clip they point to.
Lumi is not an ELN, a LIMS, a quality management system, or an electronic batch record system, and it doesn’t replace automation hardware. It covers the part of the workflow those systems assume has already been captured accurately: the manual and semi-manual steps between automated runs. Labs generally start with one high-risk workflow in one lab, where the documentation burden is heaviest and a gap costs the most, and extend from there.
FAQ
What is lab automation?
Lab automation is the use of robotics, instruments and software to perform laboratory tasks that would otherwise be done by hand. It spans liquid handling, sample and plate movement, instrument control, scheduling and data transfer. Task-level automation covers a single step or assay. Total lab automation connects most of the sample path.
How does automation reduce manual errors in lab settings?
Automation reduces manual errors by removing operator-to-operator variance from repetitive physical steps such as pipetting, dilution and plate handling. The instrument performs the step the same way every time. It does not reduce errors in the steps around the run, including reagent preparation, deck setup and the documentation of what happened.
How do labs ensure compliance with regulatory standards in automated workflows?
Labs demonstrate compliance through validated methods, qualified equipment, trained personnel and contemporaneous records showing each step was performed as documented. Automation supports the first three. The fourth still depends on what the operator enters, because instrument logs record instrument activity, not the human actions around it.
What is the best software for lab workflow automation?
There is no single best, because two different jobs are usually bundled under the phrase. If the constraint is running the workflow, the category is scheduling and orchestration software, and the criteria are instrument driver coverage, error recovery and how fast a workcell can be reconfigured. If the constraint is evidencing the workflow, the category is documentation and attestation software, and the criteria are contemporaneous capture, attributable attestations, an output an auditor accepts and a validation path. Regulated labs usually need both. Most have only the first.
How does lab automation work?
A scheduler sequences the run, handling hardware moves labware between positions and instruments perform the individual steps, with results passed to a LIMS or ELN. Total lab automation connects most of the sample path, usually on a track.
Most labs already know which workflow they would least like to defend at an inspection. That is the one to look at first.