1. Can you show me 3 AI systems in production — not demos?
This is the most important question, and you should ask it in the first five minutes. A team that genuinely builds AI agents has live URLs. They can show you a WhatsApp bot that answered 2,000 messages last month, or let you log in to a document processor that a client's legal team uses daily.
If the answer is "we can show you our demo environment" or "here is a video of what we built," ask why you cannot see the live system. Demos are rehearsed; production systems have edge cases, error rates, and real usage patterns that reveal far more.
Look for evidence specific to the AI type you need: if you want a WhatsApp agent, ask for a live example; if you need document intelligence, ask to see a running production system.
2. What does "delivered in 60 days" actually include?
Fast delivery is a common promise in the Israeli AI market. Before accepting it, ask exactly what "delivery" means. Is it a working system deployed in your environment, or a handover of code you then need to deploy yourself? Does it include staff training, a test period, and handover documentation?
A trustworthy answer includes: a deployed system in your production environment, documented handover, two weeks of post-launch support, and a training session for whoever manages it day-to-day. A clear definition of "done" matters more than the headline number of days.
3. How do you handle Hebrew? Show me a real example.
Hebrew is technically harder than it appears. Right-to-left text, Hebrew idioms that large language models handle inconsistently, mixed Hebrew-English sentences, and Israeli regulatory terminology are real engineering problems. "We use GPT-4" is not an answer; every vendor uses GPT-4. The question is how they handle the edge cases.
Ask to see a conversation transcript from a Hebrew-speaking AI agent they have actually deployed. Look for: correct RTL rendering, handling of Hebrew slang and abbreviations, graceful fallback when the AI is uncertain, and whether the system defaults to English under pressure. Vendors who take Hebrew seriously can show you their test results.
4. What is the fixed price — and what makes it change?
The Israeli AI development market has a scoping problem: vendors often quote low to win the project, then find reasons to expand scope. Protect yourself by asking for a fixed-price contract with a written definition of what changes would trigger additional cost.
Legitimate AI custom development in Israel typically costs:
| Project type | Typical cost (ILS) | Timeline |
|---|
| WhatsApp AI agent (single use case) | ₪80,000–₪120,000 | 45–60 days |
| Document intelligence system | ₪80,000–₪150,000 | 60–90 days |
| ERP integration agent (Priority or SAP) | ₪100,000–₪200,000 | 60–90 days |
| Full multi-channel AI platform | ₪200,000–₪400,000+ | 90–150 days |
If a quote is significantly below these ranges, ask what is excluded. If it is significantly above, ask what justifies the premium.
5. Who owns the code, the data, and the model?
This question separates vendors who build custom solutions from those who resell platforms. You need clear, written answers to three distinct ownership questions:
- Code: Is the source code delivered to you at project end, or hosted on the vendor's servers? Do you have the right to hire another developer to modify it?
- Data: Who has access to the messages, documents, and business data that flow through the system? Where is it stored? What happens to it if you end the engagement?
- Model: If they fine-tuned a model on your data, is that fine-tuned version yours, or does it remain the vendor's property?
A vendor who cannot answer these clearly — or whose contract answers them in the vendor's favor — is not building something you own. You are renting a dependency.
6. What is the post-delivery model?
AI systems are not static. Language model APIs update, business processes change, integration endpoints shift. A WhatsApp bot that worked perfectly on day 60 may develop edge-case failures by day 180. Ask what support looks like after delivery:
- What is the SLA for a production bug?
- Is ongoing maintenance a separate contract, or included?
- If the LLM provider changes an API, who handles the update?
- Can you extend or modify the system yourself after delivery?
Monthly retainer costs for Israeli AI maintenance typically run ₪3,000–₪15,000 depending on complexity. Factor this into your total cost-of-ownership analysis before comparing vendors on day-one price alone.
7. What is the tech stack — and can I see an architecture diagram?
You do not need to be an engineer to evaluate this answer. You need to understand three things:
- Proprietary lock-in points: If the system can only run on the vendor's infrastructure, or uses a custom framework only they understand, you have created a dependency that follows you for years.
- Stack openness: Standard stacks (Node.js, Python, mainstream cloud providers, standard LLM APIs) mean you can hire any developer to work with the code later. Exotic or proprietary choices narrow your options permanently.
- Deployment model: Cloud (AWS/Azure/GCP), your own servers, or the vendor's servers each carry different security and ongoing cost implications for Israeli businesses.
A vendor who refuses to share a basic architecture diagram before signing is signaling something. Legitimate builders have nothing to hide about their technical approach.
What to do after the meeting
After asking these 7 questions, you should have: at least two production examples, a fixed-price quote in ILS, a clear scope definition, confirmed code ownership, and a post-delivery plan. If any of these are missing or evasive, that is your answer.
To see what a working WhatsApp AI agent looks like in practice — including cost ranges and a 60-day build model — the service pages cover both. For a broader checklist that applies to any software vendor — IP ownership, team structure, how to spot firms that underquote then overcharge — the software house selection guide covers those fundamentals. For a 15-minute, no-commitment conversation where you can ask these questions directly, a consultation is the zero-risk first step.
FAQ
What is the typical cost of an AI software house project in Israel in 2026?
Custom AI development in Israel runs ₪80,000–₪400,000 depending on scope: a focused WhatsApp agent or document processor at the lower end; a full multi-channel platform with ERP integration at the upper end. Monthly maintenance typically adds ₪3,000–₪15,000 after delivery. Always request a fixed-price quote tied to a defined, written scope.
How long does it take to build an AI agent for an Israeli business?
A focused AI agent — WhatsApp automation, document intelligence, or ERP integration — typically takes 45–90 days from signed contract to production deployment. Broader platforms take 90–150 days. "60 days" is a common market promise; ask exactly what is delivered at the end of that period before treating it as a binding commitment.
What is the difference between an AI software house and a regular software house in Israel?
A specialist AI software house has a track record of deploying AI systems — not just writing code. They understand LLM prompt engineering, RAG architectures, AI agent workflows, and the specific challenges of Hebrew NLP. A general software house may offer AI as an add-on but rely on off-the-shelf tools without deep expertise in production AI systems.
Should I choose a large agency or a smaller AI-specialist team in Israel?
For AI development specifically, smaller specialist teams often outperform larger agencies. Large agencies typically staff AI projects with junior developers trained on standard tools; specialist teams have senior engineers who have shipped AI products in production. A smaller team that shows you 3 production examples is almost always preferable to a large agency showing you slide decks.
