How to Choose AI Software for Your Laboratory

Start with the problem, not the technology

Write down the bottleneck you want to change — for example backlog at sample login, time spent on second-person review, or report turnaround — and how you measure it today. A feature that cannot be tied to a measurable outcome is hard to justify, validate or retire later.

Realistic use cases for testing laboratories

• LIMS/ELN integration: extracting fields from client paperwork or instrument files into structured records, with a person confirming each entry
• Sample accessioning: reading chain-of-custody forms and labels, flagging missing information or holding-time risks
• Method selection assistance: suggesting candidate methods from a request description for a qualified analyst to confirm against the client's specification and your accreditation scope
• Scheduling: forecasting instrument and analyst capacity from historical workload
• QA review support: highlighting results, calculations or audit-trail entries that deviate from expected patterns so reviewers can focus attention
• Report drafting: assembling narrative text from approved results and templates; a responsible signatory still reviews and releases the report
• Anomaly detection: trending control charts, system suitability or proficiency results to surface drift earlier
• Equipment maintenance: predicting service needs from instrument logs and usage

Where caution is essential

Generative tools can produce fluent but incorrect text. Any output that becomes part of a result, opinion or report must be reviewed by a competent person, and the review must be recorded. ISO/IEC 17025 places responsibility for results and reported opinions on the laboratory, regardless of the tools used.

Evaluation checklist

• What exactly does the AI feature do, on what inputs, and what does it output?
• Which model or service is used, where is it hosted, and can the vendor change it without notice?
• Is client data used to train or improve the vendor's models? Can that be contractually excluded?
• Are prompts, outputs and user actions captured in a secure, time-stamped audit trail?
• Can the feature be switched off, and can records be produced without it?
• What evidence of accuracy does the vendor provide, on what data, and does it resemble your samples?
• How are errors reported, corrected and communicated to customers?

Data security and client confidentiality

• Map which client data the tool would see: identities, sample descriptions, results, reports
• Confirm encryption in transit and at rest, access control, data residency and retention
• Review the vendor's subprocessors, especially third-party AI model providers
• Check that contracts with your clients permit the processing, and update confidentiality terms if needed
• Ask for independent security attestations (for example SOC 2 or ISO/IEC 27001) and read their scope

Data integrity and audit trails

Regulators describe data integrity in terms of ALCOA+ — attributable, legible, contemporaneous, original, accurate, plus complete, consistent, enduring and available. For AI-assisted work, keep the original input, the AI output, the human decision and who made it. If the software only stores the final edited text, you lose the ability to show how a record was created.

Regulated contexts and 21 CFR Part 11

Where your laboratory supports FDA-regulated clients and records fall within 21 CFR Part 11, electronic records and signatures require controls such as validation, audit trails, access limits and signature linking. Part 11 does not prohibit AI features, but the system that holds those records must meet its requirements. FDA's Computer Software Assurance guidance encourages a risk-based approach: focus testing effort on functions with the greatest impact on product quality and patient safety. Confirm with your clients which regulations apply to work you do for them.

Model validation and fitness for intended use

• Define the intended use in one or two sentences and the acceptable error rate
• Test on a representative set of your own historical records with known correct answers
• Record performance, failure modes and the human review step that controls them
• Re-assess when the vendor changes the model, or your sample mix changes
• Keep validation proportionate to risk, following a GAMP 5–style categorisation where it helps

Vendor due diligence and procurement

• Request a written description of AI functionality — not a marketing brochure
• Ask for references from laboratories of similar size and accreditation scope
• Clarify pricing basis (users, samples, API usage) and how AI usage costs scale
• Negotiate data ownership, export, exit assistance and notification of model changes
• Confirm support for your existing LIMS, instruments and accreditation obligations

ROI and a pilot evaluation

Run a time-boxed pilot on one workflow with a baseline measurement, a defined success threshold and a named owner. Track time saved, error or rework rate, reviewer burden and any new risks. Include validation, training and subscription costs in the comparison. Expand only if the pilot meets its threshold; a well-documented decision not to adopt is a good outcome too.

Sources

• FDA — Part 11, Electronic Records; Electronic Signatures: Scope and Application (https://www.fda.gov/regulatory-information/search-fda-guidance-documents/part-11-electronic-records-electronic-signatures-scope-and-application)
• eCFR — 21 CFR Part 11 (https://www.ecfr.gov/current/title-21/chapter-I/subchapter-A/part-11)
• FDA — Data Integrity and Compliance With Drug CGMP: Questions and Answers (https://www.fda.gov/regulatory-information/search-fda-guidance-documents/data-integrity-and-compliance-drug-cgmp-questions-and-answers)
• FDA — Computer Software Assurance for Production and Quality System Software (https://www.fda.gov/regulatory-information/search-fda-guidance-documents/computer-software-assurance-production-and-quality-system-software)
• NIST — AI Risk Management Framework (AI RMF 1.0) (https://www.nist.gov/itl/ai-risk-management-framework)
• ISO — ISO/IEC 17025 General requirements for the competence of testing and calibration laboratories (https://www.iso.org/standard/66912.html)