Artificial intelligence is moving into veterinary software quickly.
That makes a previously obscure contract question much more important:
what exactly can your PMS supplier do with the data your practice generates?
The wrong way to approach this is to treat every use of anonymised data as suspicious.
Software suppliers need data to operate, secure and improve their products. Aggregated usage analysis can show which workflows fail, which features practices actually use and where performance needs attention. Done properly, that can benefit customers.
The real issue is that "using data to improve the service" and "using customer-generated data to train AI or build separate commercial products" are not necessarily the same bargain.
Practices should know which one they are agreeing to. Part 1 separates ownership from portability. This article deals with secondary use.
Start by separating five different uses
A useful contract review distinguishes:
- processing needed to provide the PMS;
- internal analytics and product improvement;
- aggregated benchmarking and industry analysis;
- AI model training or research; and
- separate commercial products or services derived from customer data.
The further a use moves away from running the clinic's own PMS, the stronger the case for explicit disclosure and customer choice.
Provet contains an important AI safeguard
Provet is a good example of why nuance matters.
Its current terms allow Provet to anonymise and aggregate customer data, use it for benchmarking and prepare reports based on that information. That is not automatically vet-unfriendly. Much of it can support product operation and improvement.
More importantly, Provet's AI clause says it will not enable settings or features that permit an AI service provider to use the customer's data to train that provider's models, unless the customer explicitly instructs Provet to do so or law requires it.
Source: Provet Terms of Service, sections 3.1.3 and 3.2.3
That is a meaningful distinction.
A clinic can allow an AI service to process information to perform a task without automatically agreeing that the AI provider may use the same information to improve its general model.
Those should be separate permissions.
Vetspire goes materially further
Vetspire's public terms are broader.
They say the customer owns its Customer Data. But they also grant Vetspire the right to use de-identified Customer Data for service improvement and for medical, veterinary and pharmaceutical research, as well as artificial-intelligence research and development.
The terms then say Vetspire owns and retains rights in services and commercial endeavours resulting from that use.
Source: Vetspire Terms & Conditions, sections 3.2 and 3.3
That is a different commercial proposition from simple product analytics.
The customer is paying for the PMS while practice-generated information can also contribute, in de-identified form, to research, AI development and potentially separate commercial value for the supplier.
That may be acceptable to some practices.
But it should be a conscious decision, not something discovered years later in boilerplate.
Covetrus' US terms are unusually explicit
Covetrus' current US Terms of Service contain explicit AI provisions.
They state that Covetrus may use de-identified, aggregated Client Data to develop and train AI models. They also grant rights to use Client Data to provide, maintain, improve, train and develop AI-enabled services and models, subject to the rest of the agreement.
The same US terms restrict customers and member practices from using certain Covetrus-provided or accessed data to train, fine-tune or benchmark AI models or transferring that data into third-party AI training or inference environments.
Source: Covetrus US Terms of Service, AI provisions
This source is US-specific and should not be presented as the contractual position for a UK Covetrus customer.
But as an industry example, the asymmetry is worth noticing: the supplier expressly reserves significant AI-development rights while also restricting some customer AI uses.
That is the kind of provision practices increasingly need to identify before signing.
IDEXX uses broad product-improvement rights, but do not infer AI training
IDEXX's Software Offering General Terms say the customer owns Customer Data while granting IDEXX rights to use it for internal purposes including providing, improving, developing and enhancing its offerings and supporting marketing and promotional activities. The terms also permit aggregation and certain de-identified uses.
Source: IDEXX Software Offering General Terms, section 4.2
That is broad.
But the public general clause does not justify automatically saying IDEXX trains external AI models on veterinary practice data.
A responsible comparison has to stop where the source stops.
If AI training matters to the practice, the correct response is to ask IDEXX explicitly and obtain the answer in writing.
Shepherd allows anonymised analysis and publication
Shepherd's EULA says the client owns its data while allowing Shepherd and its affiliates to analyse, publish and use anonymised or de-identified practice data.
Source: Shepherd Service Agreement / EULA, section 5(e)
Again, that is not inherently objectionable.
The key questions are purpose, identifiability, onward sharing and whether the resulting activity is simply improving the PMS or creating something commercially separate.
"Anonymised" deserves its own scrutiny
Anonymisation can substantially reduce privacy risk, but practices should still ask what the contract means.
Useful questions include:
- Is the information genuinely anonymised or merely pseudonymised?
- Can the supplier combine it with other datasets?
- Can it be shared with pharmaceutical companies, insurers or other commercial partners?
- Is the use limited to improving the existing service?
- Can it be used to train models offered to other customers?
- Can the resulting model, dataset or benchmark be sold?
- Can a practice opt out?
This is not an argument against veterinary data research.
Aggregated veterinary data could produce substantial benefits for animal health.
The question is who decides, who benefits and what the clinic agreed to.
AI inference and AI training are different
This distinction is particularly important.
Suppose a clinic asks an AI tool to summarise a consultation.
The model needs to process the consultation text to produce the summary. That is inference.
It does not necessarily need the right to retain that consultation and use it to improve a general-purpose model. That is training.
A good veterinary software contract should distinguish the two.
The strongest position is:
- customer data may be processed to deliver an AI feature the clinic chooses to use;
- the data is not used to train an external provider's general models by default;
- any broader training or research use is clearly disclosed;
- materially different secondary commercial use requires explicit choice; and
- the clinic can understand which AI providers receive its information.
Provet's express external-model-training safeguard is a useful example of this distinction in practice.
Seven questions every practice should ask
Before enabling AI inside a PMS, ask:
- Which data is sent to an AI provider?
- Which provider receives it?
- Is the data retained?
- Can the AI provider train its own models on it?
- Can the PMS supplier train its own models on it?
- Can de-identified data be used for research or separate commercial products?
- Can the practice opt out without losing the core PMS?
Those questions should become normal veterinary software procurement.
The aim is not to stop useful data use
Veterinary software will improve by learning from how practices work.
There is nothing inherently wrong with that.
The standard should be simpler:
Use the minimum data needed to provide and improve the product. Be explicit when the purpose becomes broader. Give the practice meaningful control when its data is being used to create a separate asset.
AI makes that distinction more important than it used to be.
And practices should not have to read twenty pages of legal terms to discover which side of the line their PMS sits on.
Contract analysis, not legal advice. Covetrus' AI example in this article is expressly drawn from its US Terms of Service and is not presented as the UK/EMEA contractual position. Sources reviewed on 2026-08-20.