Answer routine questions, hand off the rest
AI assistants on WhatsApp, email and web chat that answer from your own help content in Arabic and English, and hand over to a person when they should.
Overview
Most support queues are full of questions your help pages already answer. Customers ask on WhatsApp at midnight, in Arabic or English, and wait until morning. A generic chatbot makes this worse by guessing.
We build assistants that answer only from your approved help content, cite the article they used, and pass the conversation to a person with full context when they cannot help or the customer asks. We measure deflection honestly: a closed chat is not a solved problem unless the customer does not come back.
Signs you need this
- Agents spend most of the day pasting links to the same help articles
- Customers wait overnight for answers that exist on your website
- A previous chatbot frustrated customers and was switched off
- Your help content is out of date and nobody owns it
What you get
- Content review
- A gap list of questions customers ask that your help content does not answer, built from past tickets.
- Grounded assistant
- An assistant that retrieves from approved content only, cites sources and declines when the answer is not there.
- Channel connections
- WhatsApp Business Platform, email and web chat connected to the same assistant and your help desk.
- Hand-off rules
- Triggers for passing to a person, such as complaints, payments, low confidence or a direct request, with the transcript attached.
- Honest reporting
- Weekly figures on resolved chats, hand-offs and repeat contacts within a set window, plus a sample of transcripts to read.
How the work runs
- 1
Ticket analysis
1 to 2 weeks
We group past tickets by topic, find the routine ones and check your help content covers them.
- 2
Assistant build
2 to 4 weeks
We build the assistant, connect channels and help desk, and test it on past questions in both languages.
- 3
Limited launch
2 to 3 weeks
We switch it on for one channel or time window, read transcripts daily and adjust.
- 4
Full rollout and handover
1 to 2 weeks
We extend to all agreed channels and train your team to update content and review reports.
Timings depend on the size of your estate. The assessment fixes them before any build starts.
What you can check afterwards
- Every assistant answer links to the help article it came from
- Hand-offs reach an agent with the full transcript and the reason for transfer
- Weekly reports show resolved chats alongside repeat contacts from the same customers
- Test questions outside your help content get a polite refusal and a hand-off, not a guess
Tools we work with
- WhatsApp Business Platform
- Zendesk
- Freshdesk
- Intercom
- Chatwoot
- OpenSearch
- pgvector
- Langfuse
- Python
- Node.js
Questions buyers ask
- Will it make up answers?
- It is built to answer only from your content and to hand off when it cannot find support for an answer. We test that with questions designed to tempt it into guessing.
- Where are customer conversations stored?
- In your help desk and your cloud account, in the region you choose. Messages sent to the model can be redacted first.
- Does it handle Gulf Arabic and mixed Arabic and English?
- Yes, reasonably well, and we test with real past messages to check. Where it struggles, it hands off.
- What happens after launch?
- Your team owns the content and reviews weekly reports. Most improvement comes from fixing help articles, which we show you how to spot from transcripts.
Talk to us about AI customer support automation.
Tell us what you run and where. We reply within one business day with questions or a quote.