TL;DR: The twelve questions below cover who builds the work, what is custom, what it connects to, how it is tested in Arabic and on WhatsApp, where your data lives, what you own, what support means, how success is measured, and how you leave. Each one has a good answer and a bad answer. Ask all twelve, get the answers in writing, and compare agencies on the same rows. If an agency cannot answer in writing, that is itself an answer.
Why these questions matter more than the demo
Every AI agency in Dubai can show a chatbot answering nicely in a demo. The things that decide whether the project works a year later are invisible in a demo: who built it, how it was tested, what happens when it is wrong, where the data goes and what you are left with if you part ways. These twelve questions surface those things before you sign, when you still have leverage.
The 12 questions
1. Who will actually build this, and can we meet them?
Why it matters: sales teams promise; engineers deliver. If you never meet the engineer, you cannot judge whether the promises are realistic.
Good answer: "Here is the engineer; they will be on the next call." Bad answer: "Our delivery team will be assigned after signing."
2. What in this proposal is custom, and what is a platform you resell?
Why it matters: configuring a third-party platform is legitimate work, but it should not be priced or described as custom engineering, and it comes with that platform's limits and terms.
Good answer: a clear split, with the platform named and its limits explained. Bad answer: "It's all our proprietary technology."
3. Which of our systems will it connect to, and have you connected to them before?
Why it matters: an assistant that cannot read your diary, CRM or ERP is a FAQ widget. Integration is where most projects stall.
Good answer: names your systems, describes the method, mentions what they learned last time. Bad answer: "We integrate with everything." See our AI agents and knowledge-base service for what connecting to your own documents and systems involves.
4. How do you test Arabic, and who does the testing?
Why it matters: Arabic quality varies enormously, and Gulf dialect, Arabizi and mid-sentence language switching break many implementations.
Good answer: native speakers, real customer messages, results you can see. Bad answer: "The model supports Arabic." Our post on why Arabic breaks language models explains the problem.
5. How does it get set up on WhatsApp, and whose account is it?
Why it matters: in the UAE most customer conversations start on WhatsApp. The number, the business verification and the message templates must be in your name or you cannot take them with you.
Good answer: WhatsApp Business Platform in your Meta Business account, your number, your templates. Bad answer: "We run it through our account for simplicity."
6. What will it refuse to answer, and what happens then?
Why it matters: an assistant that always answers will eventually say something you did not approve. The boundary and the handoff are design decisions, not accidents.
Good answer: a written list of topics it will not handle, a named route to a person, and what the customer sees while waiting. Bad answer: "It's very accurate." Our guide to human-in-the-loop design covers this in depth.
7. Where is our data stored, who can access it, and is it used for training?
Why it matters: customer and patient conversations are personal data under the UAE PDPL and sector rules. "Secure" is not an answer.
Good answer: named region, named providers, named accounts, retention period, no training on your data, all in the contract. Bad answer: "Everything is encrypted." See our note on data residency and the PDPL.
8. What do we own at the end: code, prompts, knowledge base, accounts?
Why it matters: if the agency keeps the prompts or the accounts, you do not have a system; you have a subscription with a hostage.
Good answer: you own all of it, written into the agreement, with a handover procedure. Bad answer: "Ownership is covered in our standard terms."
9. What does support cover after launch, and what does it not?
Why it matters: models change, your products change, WhatsApp policies change. Someone has to keep it working.
Good answer: a written scope: monitoring, prompt updates, integration fixes, response times, what counts as a change request. Bad answer: "We're always available."
10. How will we know it is working?
Why it matters: without agreed measures you will argue about whether the project succeeded.
Good answer: a short list of measures you both agree on before launch, reviewed on a schedule, with a dashboard or report you can read yourself. Bad answer: a guaranteed outcome quoted before anyone has seen your data. Our guide to measuring an AI assistant lists sensible measures.
11. Can you show us a live system doing work like ours?
Why it matters: a slide is a promise; a live system is evidence.
Good answer: a working assistant or workflow you can interact with, ideally in your sector. Bad answer: a list of logos with no access to anything running.
12. What does leaving look like?
Why it matters: every relationship ends; the question is whether it ends cleanly.
Good answer: a notice period, a handover of code, data and accounts, and documentation good enough for another team to take over. Bad answer: "Why would you want to leave?"
How to use the answers
Put the twelve answers from each agency side by side on the same rows. Patterns show quickly: the engineering firms answer questions 1, 3 and 8 easily and get specific on 6 and 7; resellers are strong on 11 and vague on 2 and 8. Neither type is wrong for every project. Our comparison of agency, freelancer, in-house and platform helps you decide which kind of builder your project needs before you compare within the kind, and our first-meeting guide shows what a useful scoping conversation looks like.
Where Soluvide fits
We are an AI engineering firm in Abu Dhabi working with businesses across Dubai and the UAE. We build WhatsApp assistants, agents connected to your own data and the integrations behind them, in Arabic and English, in-house. We are happy to answer all twelve questions in writing before you sign anything; you can read about us and how we work first. Bring the same questions to everyone on your shortlist and compare the answers.
Talk to us
Comparing AI agencies?
Send us the workflow you want automated. You get a straight answer on whether it is worth building, and a written scope with a fixed price if it is.
Questions
Frequently asked.
Ask who will actually build the work, what is custom and what is a platform they resell, which of your systems it will connect to, how Arabic and WhatsApp are tested, where your data is stored and who can see it, what you own at the end, what support covers after launch, how they measure whether it is working, and what leaving looks like. Insist on written answers; a demo cannot answer any of these.
Where this applies