The verdict in three sentences
A Dubai contact centre handling 20,000 requests a month, 60% of them repetitive, pays agents to answer the same question about a policy, an invoice or a delivery time a hundred times a day. A RAG chatbot connected to your knowledge base, in English and Arabic, costs 16,300 to 43,500 USD excl. VAT (150,000 to 400,000 MAD in Morocco), 325 to 1,630 USD a month in API fees, and goes live in 8 to 12 weeks. It resolves 30 to 50% of requests without an agent, provided escalation rules to a human are clear.
RAG: why it is not a classic chatbot
A decision-tree chatbot answers 50 to 100 planned scenarios and fails as soon as a question goes off-script. RAG (Retrieval-Augmented Generation) first retrieves the relevant passages from your documents (terms and conditions, price lists, procedures, internal FAQs, memos), then generates an answer cited and limited to those sources. If the information is not in the base, the bot says so and hands over.
| Criterion | Scenario chatbot | RAG chatbot |
|---|---|---|
| Question coverage | 50 to 100 scenarios | Anything in the knowledge base |
| Languages | One per scenario, duplicated | English, Arabic (Gulf dialect understood), others on demand |
| Updates | Edit trees by hand | Add or replace a document |
| Resolution without agent | 10 to 20% | 30 to 50% |
| Risk of invented answers | Low, but rigid answers | Controlled through source citation and a confidence threshold |
| Setup cost | 5,400 to 16,300 USD | 16,300 to 43,500 USD |
Cost per contact before and after, 2026 estimate
Assumption: fully loaded cost of an agent-handled contact of 2.00 to 2.70 USD (salary, supervision, telephony, premises), set at 2.40 USD.
| Item | Before | After (40% handled by the bot) |
|---|---|---|
| Monthly requests | 20,000 | 20,000 |
| Contacts handled by agents | 20,000 | 12,000 |
| Agent cost | 48,000 USD | 28,800 USD |
| API and hosting | 0 USD | 980 USD |
| Maintenance and knowledge base enrichment | 0 USD | 1,300 USD |
| Total monthly cost | 48,000 USD | 31,080 USD |
| Average cost per contact | 2.40 USD | 1.55 USD |
| Availability | Opening hours | 24/7 |
The monthly saving of about 16,900 USD pays back a 30,400 USD project in under two months, or frees agents for sales and retention.
Escalation rules to a human
A chatbot that traps an angry customer destroys more value than it creates. The rules below are configured from day one.
| Situation | Trigger | Action |
|---|---|---|
| Low answer confidence | Score below 75% | Handover to an agent with the history |
| Complaint or anger | Negative sentiment or keywords | Priority handover |
| Sensitive personal data | Bank detail change, cancellation | Identity check, then agent |
| Out-of-scope question | No relevant document | Honest answer and callback offer |
| High-value customer | Segment flagged in the CRM | Dedicated agent from the first message |
| Explicit request | « I want to talk to someone » | Immediate handover |
The UAE Personal Data Protection Law (PDPL) requires transparency: customers must know they are talking to an automated assistant.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Omar, head of customer service at an insurance company in Dubai, runs 45 agents and 22,000 monthly requests. Scope: a RAG chatbot on 1,200 pages of policy terms and procedures, integrated into the website, customer portal and WhatsApp Business, in English and Arabic. Budget: 31,500 USD excl. VAT, 11 weeks, then 1,200 USD a month.
After four months: 41% of requests closed without an agent, average wait time down from 6 minutes to 1 minute 40. The saving, about 20,600 USD a month at constant volume, let the team redeploy 12 agents to renewal follow-ups. Payback in under two months.
FAQ
Does the bot really understand Arabic dialects?
2026 models understand Gulf Arabic and Arabizi well, with 85 to 92% comprehension after tuning on your real conversations. Replies can be written in Arabic or English based on the customer's preference.
How do you prevent invented answers?
The model answers only from retrieved passages, cites the source and declines below a confidence threshold. A sample of 200 conversations is reviewed weekly during the first two months.
Which channels can it run on?
Website, customer portal, mobile app and WhatsApp Business. WhatsApp often carries 50 to 70% of conversations in the region and adds Meta fees of roughly 0.01 to 0.05 USD per service conversation.
How many documents do we need to start?
200 to 2,000 well-structured pages are enough. The longest task, 2 to 3 weeks, is cleaning up contradictory or outdated documents.
Does the data stay in the UAE?
The vector database and logs can be hosted in-country; model calls go through an API with a no-training commitment. Compliance documentation is prepared during the project.
Let's scope your project. A multilingual RAG chatbot on your knowledge base, with escalation to your agents: budget 16,300 to 43,500 USD and 8 to 12 weeks. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
