What to ask before signing
How will we know if quality drops? If the answer does not involve an evaluation set, there is no answer. Ask to see one from a previous engagement.
What happens when the model provider deprecates? It will happen inside your engagement. Ask how they have handled a migration before.
Who owns the prompts, the evaluation data and the code? It should be you, all of it. Some arrangements quietly keep the evaluation set with the vendor, which makes leaving expensive.
How is token spend monitored? Costs move in the wrong direction quietly. Ask who watches it and what triggers an alert.
What happens after launch? Most AI engagements are priced to deliver a feature. Ask explicitly what the arrangement looks like in month twelve.
Can this run inside our infrastructure? If your data cannot leave, ask about open-weight models early rather than after scoping.
Who is actually doing the work? Ask whether the people in the pitch are the people assigned, and whether they stay assigned.
The ongoing work after launch — evaluation, monitoring, retraining — is covered under managed AI services.
Reading the answers
The pattern to listen for is specificity. A partner who has run AI in production will answer with a story about something that broke. A partner who has not will answer with capability language.
What Azendo delivers under this model is set out under AI engineering services.
Two of the seven questions are really about the data rather than the model, and a partner who cannot answer them usually needs managed data services more than another AI engineer. The commercial shape of the engagement matters as much as the technical answers, and staff augmentation vs managed services sets out which of the two you are actually being sold.