AI Hub 029 · Guide

Choosing an AI development partner: 12 questions to ask before you sign

Guide

Which AI development partner can you trust to build this safely, and how would you know? You cannot tell from a pitch, a demo or a list of logos. You can tell from the answers to twelve plain questions, asked of every firm on your shortlist, in writing.

Each question below has what a good answer sounds like, what to be wary of, and our own answer, with the page where you can check it. It is the list we would want a buyer to use on us before starting AI development services.

The twelve questions

  1. Have you put AI into production for a business like ours?
  2. Who is on my team, by name and role, and will they stay?
  3. Who makes the technical calls, and who do I call when something is wrong?
  4. How do you train your people on AI, and how do you keep up?
  5. Where do you start?
  6. How will we know it is working?
  7. How will you tell me when this is the wrong idea?
  8. What checks does the work go through before it reaches us?
  9. What happens when something breaks at 2 a.m.?
  10. What does month six cost, not month one?
  11. Who owns the code and the data when we part ways?
  12. Are we tied to one AI model, or to you?

Send the same list to every firm and ask for the answers in writing. A vague answer is an answer. Then ask to speak to one of their clients.

Proof and people

1. Have you put AI into production for a business like ours?

A good answer names live tools, the kind of business, and a measured result. Be wary of demos, pilots and logos with no number attached.

Our answer: we’ve put AI into production across mortgage lending, logistics, ESG reporting, recruitment, customer service and AI-native products. For a US mortgage lender, more than 12 AI tools were in production in under nine months, saving $175,000+ a month across three core workflows. Read the case studies, and the client reviews, which Clutch verifies independently: 4.9 out of 5 across 27 reviews.

2. Who is on my team, by name and role, and will they stay?

Rotating strangers is the agency trap.

A good answer is a list of names you meet before you start, and a plain account of what happens when one of them leaves.

Our answer: on a dedicated AI development team you meet everyone before they join. If someone isn’t the right fit or moves on, we replace them at no cost, and because the team works from shared playbooks and documentation, the knowledge stays with the team.

3. Who makes the technical calls, and who do I call when something is wrong?

A good answer is two named roles: one who decides how it is built, one you can escalate to. Be wary of an account manager with no engineer behind them.

Our answer: a tech lead leads the engineers day to day, makes the technical calls, and reviews what ships. A project manager runs the sprints. Your account lead comes from our UK leadership team and is your single point of escalation. For who leads the work on your side, see do you need a head of AI before you build?

4. How do you train your people on AI, and how do you keep up?

A good answer is a written way of working that everyone follows, someone whose job is testing what is new, and proof they use it themselves. Be wary of “all our developers use AI tools”.

Our answer: four people on a dedicated AI R&D team test models, tools and ways of delivering work. What they find goes into the playbooks the whole company works to. Every engineer, designer and QA engineer comes through Compass, our own training platform, before they join a client team. And we run Offshorly on tools built this way. See how we work.

How the work runs

5. Where do you start?

A good answer is one process, a small first step at a fixed price, and a number at the end of it. Be wary of a strategy phase or a large build before anything has been measured.

Our answer: with a free consultation, where we look at how the work runs today, find the process that costs you most, and say so if there is nothing worth automating. Then a proof of concept at a fixed price, quoted before it starts.

6. How will we know it is working?

A good answer measures the process before the work starts and again after, in your numbers. Be wary of a productivity percentage quoted before anyone has seen how your team works.

Our answer: You measure the difference: the new process against the old one. We won’t quote you a productivity figure until we’ve measured it.

7. How will you tell me when this is the wrong idea?

A partner who never pushes back is selling, not helping.

Our answer: If AI isn’t the right answer for a problem, we’ll tell you before you spend anything on it.

8. What checks does the work go through before it reaches us?

A good answer has a second person reviewing every change, security scanning, and one named person answering for each thing that ships. Be wary of AI that checks its own work.

Our answer: every change is reviewed by another developer and security-scanned before it reaches you. Tools and agents produce the volume. One named developer answers for what ships.

9. What happens when something breaks at 2 a.m.?

If nobody answers, you have not built a team. You have built a dependency.

A good answer names who you contact, how fast they respond, and who fixes it.

Our answer: the team that built the tool maintains it, its working day overlaps with the UK, the US and Australia, and your account lead is your single point of escalation.

Money, ownership and leaving

10. What does month six cost, not month one?

Predictable pricing only counts if it holds.

A good answer lists what is included, what would cost extra, and how you stop. Be wary of a low first month with an hourly meter underneath.

Our answer: a fixed monthly rate per team member, with part-time roles charged pro rata. There are no setup fees and no minimum term beyond the current month. If something would cost extra, you will hear it before it happens, never on an invoice. Everything we charge is on our pricing page.

11. Who owns the code and the data when we part ways?

If the answer is not “you do,” walk away.

Our answer: What we build for you is yours from the first commit, and you hold the repository. Which data any AI tool may touch, and where it is processed, is your call.

12. Are we tied to one AI model, or to you?

A good answer shows you can change AI provider without a rebuild, and that the way of working is written down and yours to keep. Be wary of a platform you have to keep renting to keep your own tools running.

Our answer: our workflows run regardless of which model sits underneath, so moving to a better or cheaper one is a configuration change, not a rebuild. The playbooks live in GitHub, and your team works from them from day one. We hold nothing back to keep you dependent on us.

Ask us all twelve

Bring this list to the first call and hold us to it. See how our AI development services work, or book the free consultation and start with question one.

Offshorly is a UK-led AI development company that builds dedicated AI development teams in the Philippines, trading since 2014.

GET STARTED

Tell us what you want to build.

Whether it’s a quick question or a detailed brief — we’d love to hear about it.

  • Honest assessment of whether we're the right fit
  • Fixed pricing, no hidden costs
  • 10+ years of trusted delivery
No sales pressure.
No lengthy process.
Just an honest conversation about technology.








    We'll use these details only to reply about your project. See our privacy policy.

    Thank you for contacting us!

    We'll be in touch with you shortly.