Choosing the wrong AI partner in Jordan wastes a year and a budget. Seven questions every Jordanian business should ask before signing — and how to read the answers.
1. Do they understand your business, or just the technology?
A good AI partner can describe your operations back to you in the first meeting. They ask about your existing tools, your team's workflow, your customers' language, your bottleneck. A weak one talks about transformer architectures and shows you a generic deck.
If you can't get a clear answer to "where do you think we'd see ROI in our specific business?" within 30 minutes — that's a signal. The global AI agents pillar is useful background to have read going into these conversations.
2. Can they show you a working agent — not a deck?
This is the single highest-signal question. Demos are cheap; a working agent in a real production environment is not. Ask to see a live system they've shipped (preferably with a Jordanian client, with permission). Watch it handle an unscripted question.
If everything they show you is a slideshow or a recorded demo, you're talking to a reseller, not a builder.
3. Are they fluent in Arabic — and your industry's language?
This is where most generic agencies fail in Jordan. "We support Arabic" means nothing if their Arabic was built on news articles and not on Levantine customer messages, medical voice notes, or legal contract language.
Ask: "Show me your evaluation results on Levantine Arabic for our use case." If they can't, they haven't done the work. The Arabic problem is covered specifically in our Jordan customer-service tools comparison and the Jordan healthcare AI guide.
4. Do they own the model layer, or rent it?
Most agencies wrap the OpenAI/Anthropic API. That's fine for many problems — the LLM is genuinely commodified. But it has implications:
- You're at the mercy of pricing changes from US providers.
- Your data flows through US/EU servers unless explicitly architected otherwise.
- You can't tune model behavior beyond prompt-engineering.
A more capable partner can do all of: rent a frontier model when it's right, run open-weight models in-region when needed, and tune them when the use case justifies it. For the technical context, our LLM agents explainer covers the layer they're working in.
5. What does "done" look like — and when?
A red flag in any AI engagement is fuzzy success criteria. A real partner will write down, before kickoff: "In 8 weeks, the agent will handle X% of incoming tickets at Y% accuracy, measured by Z, with these fallback paths." If they won't, you'll be paying for a forever-prototype.
The honest milestones look like the Jordan SME automation playbook: a scoped pilot in 30, hardening in 60, expansion in 90. The global automation playbook covers the broader methodology.
6. Who owns the IP, the data, and the model after delivery?
Read the contract — and have a lawyer read it. The clauses that matter:
- IP ownership of any custom code, prompts, and fine-tuning data.
- Data ownership — your data should be yours, full stop, with clear deletion rights.
- Exit terms — can you take the system to another vendor, or is it locked to theirs?
- Confidentiality — especially important in healthcare, finance, and regulated sectors.
Vendors who push back on these clauses are telling you something. The good ones have a standard answer ready and don't blink.
7. What happens 6 months after launch?
AI systems are not "ship and forget." Models drift. Customer behavior changes. New edge cases appear weekly. The vendors that quietly ship a great launch and then disappear leave you with a system that's quietly degrading.
Ask: "What's your operating model after launch — monitoring, retraining cadence, support SLA, escalation path?" The answer should be a real document, not improvisation.
For a comparative view of which platforms are designed for long-term operation versus rapid demo, see our global ranked platforms list.
Putting the seven together
You're not looking for a vendor who scores 7/7 on slick answers. You're looking for one who answers honestly — including "we haven't done that, but here's how we'd approach it" — and whose previous clients corroborate the answers. Reference checks in Jordan's small market are unusually easy; use them.
A short list of three vendors, each running through these seven questions, is enough to make a good decision in two weeks. It's worth the time.
How MedGAN AI answers these questions
We built MedGAN AI specifically because we kept watching Jordanian businesses sign contracts with vendors who couldn't answer questions 3, 4, or 7. Our default scope is a fixed-outcome 8-week pilot, with clear success metrics, full IP and data ownership transferred to you, and a defined post-launch operating model. If you'd like to run the seven questions through us, book a discovery call or email contact@medgan.co. The Jordan business pillar guide is a good warm-up read before the call.