AI Readiness and Selection
Decide what AI has to deliver, and which data it may touch, before anyone sells you a platform. Then choose the provider by testing on your own data, and sign terms that still hold when the model changes.
Book a conversationWhat AI Readiness and Selection Covers, and When You Need It
Buying AI starts before the first demo. Readiness means a use case, a measure of its value, and clear rules for what data the supplier may see. Those answers narrow the options. AI features in software you already license run under that supplier's terms, and may route requests to other model providers. With a model provider's API, or the same model run inside a major cloud, the supplier runs the model and sets limits, retention and retirement. If you host a model yourself, your team runs the capacity, security and upgrades.
When you need it
- AI spending is rising, and few can show what it delivers.
Only 29% of organizations currently measure the value of AI software
Flexera 2026 State of ITAM - A platform or model commitment is close to signature, and nobody has settled which uses it serves.
- Staff are using personal AI accounts for work, so company data is moving under consumer terms.
- You need processing kept in the US or the EU, and have to choose a deployment type and region that keeps it there.
- A model the business depends on is being retired, or a renewal is coming on terms that predate current AI clauses. In our judgment, that is the moment to revisit the choice.
When you don't
In our judgment, a formal selection may not be needed when:
- Your AI use is limited to features built into software you already license, under business terms already reviewed, with no sensitive data in scope.
- Your own AI platform team already tests models on your tasks and has enterprise terms in place with its providers. You may need us only when the next renewal or new need arrives.
If the AI you already own covers the need, our advice is to use it.
What Decides Which AI Platform Fits
Settle these before the pilot, not after the first invoice.
A value measure set before the pilot
One analyst firm names four reasons generative AI projects are abandoned after proof of concept: poor data quality, weak risk controls, rising costs and unclear business value. Name the use case, how its value will be measured and who owns the result before the pilot starts. End the pilot with a go or no-go against that measure.
Data that is ready, and permissions that are right
Match data sources to the use case before a supplier asks for access. An AI assistant inside SaaS can surface what a user is already allowed to see, so fix oversharing before you switch one on.
Security and privacy requirements written first
Write down which data may leave the company, where it may be processed, whether the supplier may keep it and whether it may train on it. Each answer rules deployment options in or out. Ask whether synthetic data could stand in for the sensitive data.
Your data and tasks, not the leaderboard
A public leaderboard does not test your data or your tasks, and scores can be inflated when test questions leak into training data. Decide what you are testing for, run each candidate on your own work, and ask suppliers for the evidence behind their capability claims.
A pricing model that matches the risk
Pay per token and you carry the volume risk. Reserved capacity buys steadier throughput, but it is a fixed commitment, often tied to one model or publisher, that usually cannot be canceled mid-term. Batch pricing typically costs less, and may carry no real-time service level. Sizing the commitment is covered on our technology savings and cost optimization page.
Rate limits are a capacity decision
Providers cap requests and tokens per minute and per day, and some raise the caps as spend grows. At some providers, shared capacity can return capacity errors that do not count against the service level. Plan for peak volume before launch.
Residency depends on the deployment type
Choosing a region is not the whole answer. Global deployment types may process requests anywhere, newer models tend to reach residency-bound options later, and AI features in SaaS can send requests to third-party models outside a data boundary. Ask where inference runs, not only where data is stored.
One model, several retirement dates
Hosted models retire on the supplier's schedule, and the same model can retire on different dates depending on where you buy it. Some deployments leave the migration to you. Keep the option to move between models, and test the successor well before the date.
One AI Purchase, Two Sets of Questions Answered
For IT and data leaders: a provider that fits the work
- A provider chosen by testing on your own data and tasks, not a public leaderboard.
- Residency settled by deployment type and region, with inference location checked.
- Provider demos, proposals and term comparisons handled by TREU, so your team's time goes into testing models on your own work.
For finance: a commitment sized to the risk
- A value measure set before the pilot, and a go or no-go against it at the end.
- A pricing model matched to the volume risk you can carry.
- Price protection for the term, negotiated before published rates change.
What to Watch For in an AI Agreement
Check these terms in the agreement, the service terms and the data processing addendum.
-
Training on inputs and outputs.
Limit use of your inputs and outputs to providing the service, not training or fine-tuning models. Check which tier the terms cover; consumer and business terms for the same product can differ.
-
Output ownership and use.
Business terms often leave ownership of output with you, but owning output is not the same as holding copyright in it, and that question is for your counsel. Some terms also bar using outputs to build a competing product or train a competing model.
-
IP indemnity and its conditions.
Some large suppliers defend customers against IP claims about output, but usually only on paid tiers and only if you meet their conditions, such as required content filters and documented testing. Modifying the model, combining it with other technology or bringing your own model can fall outside the cover.
-
Retention and deletion.
Default logs can keep prompts for a period, zero retention can need the supplier's approval or specific configuration, and some features store conversations by design. Set return and deletion rights that suit the use case.
-
Subprocessors.
The subprocessor list often sits in the data processing addendum, and a SaaS supplier may add model providers for its AI features. Keep the supplier responsible for what its subcontractors do.
-
Model change and retirement notice.
Providers set their own notice periods, some can shorten them for safety reasons, and models are updated on the supplier's timing. Negotiate notice, the chance to test a new version before it goes live and help with migration.
-
Price changes.
Published rates can change soon after they are posted. Negotiate price protection for the term.
-
Acceptable use and suspension.
Usage policies, supported-region lists and codes of conduct are separate documents the contract incorporates, so they bind you, and a breach can lead to suspension. Check your use cases and locations against them before you sign.
-
Exit.
Plan for access during wind-down, return or deletion of your data, transition help, and whether a fine-tuned model or its configuration can leave with you. Some providers can end the agreement for convenience.
This list is a starting point for your review. TREU is not a law firm. Your counsel gives the legal advice.
How TREU Brings an AI Provider to Contract
See how we work-
Supplier identification
We match your use case and data requirements to the deployment options that fit, then to the platform and model suppliers that offer them, starting with the ones you already pay.
-
Buying cycle
We put each provider's offer on the same basis, test the terms against your requirements and negotiate them with your legal team, including how your data may be used.
-
Implementation oversight and accountability
We hold the supplier to what it committed to through go-live, and check the first invoice against the agreement. We don't monitor usage or monthly invoices; we source the providers who do.
-
Client and supplier lifecycle management
Models retire and prices change. We track your terms and renewal dates, join supplier business reviews, renegotiate each renewal and source the next need, for the life of every agreement.
Questions Buyers Ask Before They Buy AI
Will the supplier use our data to train its models?
It depends on the tier and the terms. Business and API terms from major providers generally say no by default, while consumer terms for the same product can differ. Put the limit in the agreement.
Who owns what the AI produces?
Major business terms generally give or leave ownership of output with you. Whether that output is protected by copyright is a separate question, and one for your counsel.
Are we protected if output infringes someone's IP?
Some large suppliers offer an IP indemnity for output, with conditions: required safeguards, paid tiers, and exclusions for modifying the model, combining it with other technology or your own inputs. Read the conditions before you rely on it.
How should we compare models?
Test the candidates on your own data and tasks, against the value measure you set, not on public leaderboards alone. Test again when a model is replaced.
Should we pay per token or reserve capacity?
Per token puts the risk on volume. Reserved capacity is a fixed commitment, often tied to one model or publisher and usually not cancelable mid-term, bought for throughput and steadier response times. Our technology savings and cost optimization work covers sizing it.
Can we keep our data in the US or the EU?
Usually, by choosing the right deployment type and region. Global options may process requests anywhere, and newer models may reach residency-bound options later.
What happens when a model is retired?
Providers retire models on their own published schedules, requests to a retired model fail, and some migrations are the customer's job. Negotiate notice, and test the successor early.
How is TREU paid?
TREU is paid by the supplier you select, never by you.