Before anyone signs anything, the question that decides the budget is the AI project timeline. Most vendors answer it with a number that sounds good in a meeting and collapses on contact with your actual data. Here are the honest ranges we see on Kuwait and GCC projects, week by week, plus the specific things that add a month without anyone noticing.
The honest ranges for an AI project timeline
Almost every AI build in the Gulf falls into one of three bands. The band is set by how many live systems the AI has to touch, not by how clever the model is.
- 2 to 3 weeks — a single-purpose assistant on content you already own: a bilingual FAQ bot on your website, a WhatsApp responder that handles hours, location and pricing, or internal search over your policy PDFs. No new integrations, one approval chain, one owner.
- 6 to 10 weeks — an agent that reads from or writes to a live system: your booking calendar, CRM, inventory, or accounting tool. This is where most real business value sits, and where most of the schedule risk sits too.
- 3 to 6 months — multi-department deployments with security review, role-based access, audit logging and a formal procurement process. Common in banks, government entities and large groups.
If a vendor quotes four days for the middle band, they are showing you a demo, not a deployment. The gap between the two is evaluation and integration. If you are still deciding whether you need a build at all, ChatGPT vs a custom AI assistant is the cheaper question to answer first, and it can cut your timeline to zero.
Where the weeks actually go
Model work is the smallest slice. On a typical 8-week integrated project, the split looks roughly like this:
- Week 1 — scoping and data access. Pick one measurable job, write the failure cases down, and get credentials. Access requests are the single most common reason week 1 becomes week 3.
- Weeks 2 to 3 — retrieval and the answer layer. Your documents get cleaned, chunked and indexed. Expect to find contradictions: three price lists, two refund policies, an Arabic version that disagrees with the English one.
- Weeks 3 to 5 — integrations. Booking, CRM or ERP. Older on-premise systems need a middle layer, which is where AI ERP integration work earns its line item. If the AI is going inside software you already run, adding AI to an existing app has a different critical path than a greenfield build.
- Weeks 5 to 7 — evaluation in Arabic and English. Build a golden set of 150 to 300 real customer questions, including Kuwaiti and Gulf dialect phrasing, and score every change against it. Microsoft documents the same staged approach to evaluating generative AI: test the model, then the system, then the live deployment.
- Weeks 7 to 8 — pilot with a real team. Twenty staff, two weeks, with an escalation path to a human. This is the only honest test.
None of this is optional padding. McKinsey research on AI adoption consistently finds that the organisations capturing value are the ones redesigning the workflow around the AI, not the ones bolting it on. Redesign takes calendar time.
What stretches the timeline in Kuwait and the GCC
Five things account for most of the slippage we see locally:
- Knowledge that only exists in WhatsApp threads. If your pricing lives in a sales manager’s chat history, someone has to write it down before the AI can use it. Budget a week.
- Dialect testing. Modern Standard Arabic is the easy case. Customers type Kuwaiti, Saudi and Egyptian phrasing, often mixed with English and Arabizi. Testing that properly adds days, and skipping it adds months of cleanup.
- Approval chains and the calendar. Ramadan, Eid and summer travel genuinely move go-live dates. Plan sign-off gates around them rather than discovering them.
- Security and procurement review. Regulated sectors add four to twelve weeks before a line of code is written. AI automation for banks and AI automation for government both cover what those reviews ask for.
- Unclear ownership. Projects with one named internal owner finish. Projects owned by a committee drift by a factor of two.
How to compress the timeline without breaking the result
Narrow the first release to one job with a number attached: cut first-response time, or deflect 40 percent of repeat questions. One job ships in weeks; a platform ships in quarters. Then phase the rest.
Phasing is how long enterprise rollouts stay sane. On our factory timesheets, wages and budgets system, going live department by department meant the first group was working in the new system while later groups were still being scoped, which kept the build honest and the feedback loop short.
Finally, settle money before week 1, not week 5. Our breakdowns of AI automation pricing in Kuwait and AI consultancy pricing exist so that budget conversations do not consume the schedule. If you want a timeline mapped against your own systems and data, that is the first thing we produce on every AI engagement — usually a one-page plan with dates, owners and the measurable outcome, before any build begins.