.avif)
Summarize with:
Most animal research facilities already collect and maintain the data an analytics or artificial intelligence (AI) tool would draw on: approved protocols, census counts, animal health records, standard operating procedures (SOPs), and environmental data. In many programs, this information is distributed across free-text notes, separate applications, and spreadsheets that staff reconcile manually.
Before evaluating any product, a facility needs to know whether its research data is AI-ready and which datasets fall short. This guide presents four questions to answer first and two additional questions to address before selecting a tool. Together, they work as a short research data readiness assessment for the animal facility director and the IT lead who will share the decision.
What AI and Analytics Need from Animal Facility Data
Predictive models, language models, and operational analytics each read facility data differently, so the demands on that data vary by use case.
All three draw on information the program already keeps for its own oversight. The Public Health Service (PHS) Policy on Humane Care and Use of Laboratory Animals and the Animal Welfare Regulations (9 CFR Parts 1 to 3) both set recordkeeping requirements. Those include Institutional Animal Care and Use Committee (IACUC) meeting minutes, protocol decisions, and semiannual reports. Daily husbandry and veterinary care produce far more data than that.
A veterinary technician reading one note that says "BCS 2" and another that says "thin" knows both describe body condition. A model processing a year of entries counts them as unrelated values unless they were recorded in the same field.
Analytics and AI draw on data in different ways. Descriptive analytics summarizes what has already happened, such as census trends by room. Predictive models estimate what is likely to happen next, such as housing demand next quarter. Large language models work with text, which is why they suit retrieval across protocols and SOPs.
What Questions Should Research Facilities Ask Before Adopting AI?
Each of the four questions below points to a data gap that any AI tool would inherit. The animal facility director and the IT lead should answer them together: the facility team knows how data is entered on the floor, and IT knows where it is stored, how it moves, and who can access it. The attending veterinarian and the IACUC office should review the answers too, since much of the data in question supports their responsibilities.
Question 1 : Is your operational data recorded consistently, with attribution and a date?
A model looking for early welfare signals needs the same observation recorded the same way every time. Suppose one technician writes "hunched" in a note, another selects "abnormal posture" from a list, and a third records nothing until the animal is treated. The program saw one pattern, but the model sees three unrelated events.
Check each data set the facility plans to use for three properties.
• Structured fields versus free text : Free text has a place in clinical notes. Clinical signs, treatments, body condition, and room conditions should still come from controlled lists wherever possible. Structured SOAP (Subjective, Objective, Assessment, Plan) documentation in veterinary care is one format that supports later analysis.
• Attribution : Each entry should show who made it. Shared accounts and entries made on behalf of a colleague weaken that link.
• Timing : Each entry should carry the date and time of the observation, captured by the system rather than typed in. An entry transcribed from a paper sheet two days later carries the wrong time.
Facilities with FDA-regulated work will recognize these expectations. The Food and Drug Administration (FDA) 2018 guidance on data integrity for drug manufacturing describes reliable data as attributable, legible, contemporaneously recorded, original or a true copy, and accurate, often shortened to ALCOA. The guidance was written for drug manufacturing, and FDA guidance is not legally binding, so it does not govern academic animal care programs. The five attributes are still relevant because they describe what makes any record reliable: who created it, when it was created, whether it can be read, whether it is the original entry, and whether it is accurate. Clinical observation records in an animal care program need these same qualities, which makes the attributes a practical standard for evaluating that data.
One practical way to assess consistency is to export six months of clinical observation records and review a field that should contain a limited set of standard entries, such as the observed clinical sign or the body condition score. If that field holds many variations of the same entry, such as different spellings or abbreviations of one observation, the data is not being recorded consistently. The results indicate how much data cleanup is needed before the records can support analysis or AI use, and which fields should be restricted to standard values going forward.
Question 2: Do protocol, facility, and animal health records link to each other, or are they reconciled by hand?
Several routine oversight questions need data from more than one system. Is the census on an approved protocol within the approved animal numbers? Did animals with a clinical event share a room, a protocol, or a procedure? Are animals still being ordered against a protocol that is close to expiration?
Each answer depends on shared identifiers, usually the protocol number and an animal or cage identifier, shared consistently across the protocol system, the census and ordering system, and animal health records.
Where those systems are separate, staff link the data by hand. A coordinator exports census to a spreadsheet, matches it against protocol approvals and resolves the differences before an inspection or a billing run. Experienced coordinators catch most mismatches this way because they know the program.
When an analytics tool receives three separate exports, it does not have the necessary context, and as such inherits every mismatch the coordinator would have corrected manually. Typical examples are a protocol number entered in a different format, an animal moved between rooms without a census update, or an amendment approved after an order was placed.
Linkage is likely the main gap if someone runs a recurring reconciliation outside the systems, such as a monthly census check against protocols. The same applies if the IACUC office and the facility report different figures for animals on protocol. For the IT lead, the follow-on check is whether each system shares data through a supported interface or only through manual exports. If exports are manual, every analytics run will need a staff member to prepare the data first.
Question 3: Can you produce a complete history for one protocol or one animal without assembling it first?
Pick one active protocol and ask for its full history: original approval, amendments, personnel changes, animals ordered and received, current census, clinical events, and post-approval monitoring findings. Then choose one animal and ask for its history from arrival to final disposition. Note how long each takes and how many people are involved.
If both histories can be retrieved directly from the systems, Questions 1 and 2 are largely answered. If staff have to assemble them, that effort previews what an AI project would need to resolve first. Semiannual program reviews and accreditation site visits call for the same histories, often on short timelines.
Predictive models also need enough past data to find patterns, and data definitions must remain consistent over time. A system migration, a change in observation categories, or a renumbered set of rooms can split a five-year history into short segments that do not compare. Before a pilot, the facility director and IT lead should agree on how far back usable data actually goes.
PHS Policy requires most IACUC records to be kept for at least three years, and records tied to approved protocols for the duration of the activity plus three more years. Those are minimums, and a predictive model may need a longer and more consistent history.
Question 4: Have you defined the first problem AI should help solve, and who is responsible for acting on the results?
A defined first question sets the scope of the data work and gives the facility a way to judge whether a pilot succeeded. Examples:
• Which rooms are trending toward capacity over the next two quarters?
• Which early changes in observations preceded clinical cases in the past?
• Which protocols are approaching their approved animal numbers while orders remain open?
• Which SOP topics do staff look up most often, and how long does it take them to find an answer?
Each question implies its own data set. The capacity question needs census and ordering history. The welfare question needs consistent clinical observations. Starting from one question lets the facility check readiness for one data set instead of every system at once.
Every output also needs a named owner who will act on it, and that owner has to match institutional responsibility. A model that flags animals for closer observation gives the veterinary team an earlier prompt, and the clinical judgment stays with them. The Animal Welfare Regulations require each research facility to give the attending veterinarian appropriate authority to ensure adequate veterinary care (9 CFR 2.33).
A capacity forecast informs the facility director's planning. A flag on approved animal numbers goes to the IACUC office, which decides whether follow-up is needed. If no role in the program would act on an answer, the facility should choose a different first question.
Two More Questions Before Choosing a Tool
Once a facility has a first use case in mind, two further questions come up before vendor evaluation. Both mainly concern the IT lead.
Question 5: Are the protocols and SOPs an AI tool would search current and version-controlled?
Natural-language retrieval is one of the more practical early uses of large language models in an animal facility. A technician asks how often post-operative animals on a given protocol should be checked, and the tool returns the relevant protocol section and SOP. The tool returns whichever version of a document it finds. If superseded SOPs sit in the same shared drive as current ones, or approved amendments live only in email, the tool can return an outdated answer with full confidence.
Before a pilot, confirm three things. Each SOP has one current version with an effective date. Approved protocols, including amendments, are stored where the tool can read them. Existing access controls are preserved, so staff retrieve only the documents they are permitted to see. Staff should open and act on the approved protocol and current SOP the tool points to, rather than on its summary.
Question 6: Who governs the data, and where will it go?
An analytics or AI tool may copy facility data into a new environment, send it to an external model provider, or combine it with data from other systems. The IT lead should settle the following before a contract is signed:
• Who owns the definition behind each field, such as what counts as a clinical event or an active cage?
• Where will the data be stored and processed, and does that meet institutional security requirements?
• Does the vendor or model provider retain or reuse the data?
• How will outputs be reviewed, and who signs off before an output informs a program decision?
• Does the work fall under an existing institutional AI policy or review process?
Institutions may already have guidance on AI use through their IT, research, or legal offices. The National Institute of Standards and Technology (NIST) also publishes a voluntary AI Risk Management Framework that some organizations use to structure these reviews. These frameworks should be applied within the existing oversight: the IACUC, the attending veterinarian, and the Institutional Official keep the responsibilities they hold under PHS Policy and the Animal Welfare Regulations.
Reading the Answers: A Practical Research Data Readiness Assessment
Answers to the six questions classify each dataset into one of three readiness levels. A facility's animal health data may be ready for a pilot while its census data still needs linkage work.
Facilities already running electronic systems are not automatically in the first readiness level. Older systems may store data in formats that are hard to export, hold free text where structure is needed, or keep protocol, census, and animal health data in separate products that do not exchange information.
Programs at the third readiness level need to improve their data and systems before introducing AI. These improvements also support current compliance work. For example, when animal census counts are linked to approved protocols, animal health records are entered consistently, and a protocol’s full history can be retrieved when needed, the program is better prepared for IACUC semiannual program reviews and facility inspections, accreditation site visits, and day-to-day oversight.
How Key Solutions Supports AI and Analytics Readiness
The Key Solutions Animal Research Suite captures the operational data that analytics depends on, and the Key Solutions Professional and Consulting Services (PCS) team provides data analytics, AI and machine learning (AI/ML), and data infrastructure services alongside the software. Four points are relevant to the questions above.
Structured capture inside daily work: Lab Animal Health Software (LAHS) keeps animal health records under a unique ID for each animal and uses structured SOAP documentation for veterinary care. That gives the consistent, attributable data Question 1 asks for as a by-product of routine clinical work.
Protocol-linked facility operations: Lab Animal Resources Software (LARS) validates animal orders against IACUC-approved protocols and supports census, transfers and housing management. Facility data therefore carries the protocol link Question 2 looks for. eProtocol IACUC and IACUC-PAM sit in the same suite, which keeps protocol review, facility operations, animal health, and post-approval monitoring within one product family.
Analytics from a team that knows the data model: For institutions unsure how ready their data is for AI, the starting point is an AI readiness assessment from Key Solutions Professional Consulting Services (PCS). The assessment reviews data quality, systems, and governance. It identifies the use cases best suited to AI and sets out the steps needed before implementation. PCS then designs and delivers analytics and AI/ML solutions around each program’s priorities. Because PCS works with the same platform the facility already uses, projects can begin from established data definitions and identifiers. Where facility data is stored outside the suite, LAHS can integrate with electronic lab notebooks (ELN) and laboratory information management systems (LIMS). PCS data infrastructure and hosting services support the wider data environment.
Security and sector focus: Key Solutions is ISO 27001:2022 certified, which means its information security management system has been independently audited against an international standard. For CIOs and IT security teams, the certification provides verified evidence for vendor risk and data security reviews before facility data is moved or integrated. With 15+ years focused on life sciences research, Key Solutions builds its products and services around the committee structures, regulatory requirements, and review cycles under which animal care and use programs operate.
More detail on the services side is available on the AI/ML Integrated Services and Data Analytics pages.
Where to Begin
The export test in Question 1 and the history test in Question 3 require no new tools and can be completed by existing staff. Their results give an initial view of the program’s AI readiness level and indicate whether the next step is an AI pilot, targeted data cleanup, or moving core workflows into structured systems. Where the results are mixed or unclear, a formal AI readiness assessment can confirm the readiness level and identify which use cases and preparatory work to prioritize. The tests also allow the facility director and IT lead to evaluate AI tools against their own data rather than a vendor’s demonstration data.
To see how structured protocol, facility, and animal health data can support your program's analytics plans, talk to an expert.
Frequently Asked Questions
