# Questions to ask an AI vendor
A checklist of plain questions to put to any AI vendor before you sign.
## Data and privacy
- [ ] Do you use our data to train your models, and can we opt out?
- [ ] Where is our data stored, and in which country?
- [ ] How long do you keep it, and can we have it deleted on request?
- [ ] Who inside your company can see our data?
- [ ] Do you pass our data to any third parties, including the model provider behind your product?
## Reliability and support
- [ ] What is your uptime, and what happens when the service is down?
- [ ] How do we report a problem, and how fast do you respond?
- [ ] When the AI gets something wrong, how do we get it fixed?
- [ ] Do you tell us before you change or update the underlying model?
## Cost and commitment
- [ ] What does this cost at the volume we actually expect?
- [ ] What happens if we go over our plan, and is there a cap?
- [ ] What is the minimum commitment, and how do we cancel?
## Exit and control
- [ ] If we leave, can we export our data and our configuration?
- [ ] Do we keep access to our own historical results?
- [ ] What of ours is deleted when we leave, and when?
## Safety and limits
- [ ] What can your product not do well, honestly?
- [ ] What guardrails stop it taking an action it should not?
- [ ] Can we see it handle a messy real example from our own work?
## Good enough to move on when
- [ ] Every data and exit question has a clear written answer.
- [ ] You understood the answers without needing to trust a buzzword.
- [ ] Nothing on the list was met with a dodge.
In 30 seconds
Take this list into the sales call. Ask for answers in writing, in plain language. A good vendor will not mind. A vendor who dodges these is telling you something. Tick each one only when you have a clear answer you understand, not a reassurance.
Data and privacy
Do you use our data to train your models, and can we opt out?
Where is our data stored, and in which country?
How long do you keep it, and can we have it deleted on request?
Who inside your company can see our data?
Do you pass our data to any third parties, including the model provider behind your product?
Reliability and support
What is your uptime, and what happens when the service is down?
How do we report a problem, and how fast do you respond?
When the AI gets something wrong, how do we get it fixed?
Do you tell us before you change or update the underlying model?
Cost and commitment
What does this cost at the volume we actually expect?
What happens if we go over our plan, and is there a cap?
What is the minimum commitment, and how do we cancel?
Exit and control
If we leave, can we export our data and our configuration?
Do we keep access to our own historical results?
What of ours is deleted when we leave, and when?
Safety and limits
What can your product not do well, honestly?
What guardrails stop it taking an action it should not?
Can we see it handle a messy real example from our own work?
What this does not solve
Good answers here reduce risk; they do not remove your own duty to specify the workflow, set its boundaries, and review its output. Use the answers to compare vendors with how to choose an AI vendor.
Good enough to move on when
Every data and exit question has a clear written answer.
You understood the answers without needing to trust a buzzword.