Think Build Implement Repeat
London, UK +44 7367 067226
WhatsApp FOLLOW f in X
AI Integration

Twelve Questions Before You Sign

Last updated:

The four that matter most

  1. How will we know it is working? A good answer describes an evaluation set with real cases.
  2. What happens when it is wrong? A good answer describes confidence routing and a review interface.
  3. Who owns the code and credentials? The answer should be you, in your accounts.
  4. What will it cost to run at our volume? A modelled figure, with the assumptions stated.
A supplier who cannot answer the first two has built demonstrations, not systems. That is worth knowing before rather than after.

Eight more worth asking

  • Which model provider, and can we change it later?
  • Where is our data processed, and is it used for training?
  • What is retained, by you and by the provider, and for how long?
  • How do we turn it off in an emergency?
  • What does maintenance cost annually?
  • Can we speak to a client running something similar?
  • What is not included in this price?
  • What happens if accuracy is below what we agreed?

Warning signs

Accuracy promised as a number before anyone has seen your data. No mention of review or correction anywhere in the proposal. Credentials held in the supplier's accounts. Vague language about the model doing the work.

None of those is necessarily disqualifying on its own. Two or more together usually is.

What a good proposal contains

SectionWhat to look for
ScopeWhat is in and explicitly what is out
Success measureA number agreed before the build starts
EvaluationHow correctness will be established
Failure handlingWhat happens on outage, error and low confidence
Running costModelled at your volume, with assumptions
MaintenanceA yearly figure, not a footnote
HandoverCode, credentials, documentation, all yours

Ask for a small first phase

A proof of feasibility on your real data, priced separately, before a full build. Any competent supplier will suggest it themselves.

It de-risks the decision for both sides and it makes the eventual estimate considerably more accurate.

Frequently asked questions

Should we get three quotes?

Two or three, on the same written scope. Quotes on different scopes cannot be compared and comparing them is worse than not trying.

Is the cheapest quote usually the problem?

Often it is missing the evaluation, review interface or maintenance rather than being efficient. Compare what is included, not the total.

What if a supplier will not answer these?

That is a complete answer. These are ordinary questions for anyone who has delivered this kind of work.

Can you review a proposal from someone else?

Yes, and we do. It is a short piece of work and it usually surfaces two or three specific gaps.

Keep reading

Holding a proposal you are not sure about?

Send it over. We will tell you what is missing, whether or not you work with us.

Book a free 30-minute call Get a project estimate WhatsApp us

Related services

What we build for problems like this one

AI IntegrationEnterprise AIAI Agents