Converla

AI agents & knowledge base

AI in Converla works best when it answers from your approved knowledge — not inventing policy. This guide covers content prep, enablement, and safe handoff.

Ops, support leads, knowledge owners

Before you start

  • A Converla workspace with at least one connected channel (WhatsApp and/or web widget).
  • Source content you trust: FAQs, shipping policy, product care, return rules, or internal help articles.
  • A teammate who can review AI answers during the first week.

How to

Collect source-of-truth content

Prefer current help center articles, FAQ pages, and policy PDFs over old chat transcripts. Remove outdated promos and conflicting prices before upload.

Add documents to the knowledge base

In admin Knowledge Base, upload or create documents for your brand. Wait for processing to finish so retrieval can index the content before you enable AI replies.

Enable AI replies on the right channels

Turn on AI for WhatsApp and/or the web widget where you want automated first answers. Keep business hours and after-hours behavior explicit for your team.

Define human handoff rules

Customers must be able to reach a person for billing disputes, complaints, and custom quotes. Train the team to take over the same thread without restarting context.

Review early conversations weekly

Sample AI threads in the inbox. Fix knowledge gaps by updating documents — do not paper over wrong answers with more generic prompts alone.

Tips that save time

Short, factual chunks beat huge dumps

Split long manuals into focused topics (shipping, returns, sizing). Retrieval quality improves when each document answers a clear question family.

Never invent Meta or payment policies

If a policy is not in your knowledge base, AI should not guess. Add the approved wording first, then retest the same customer question.

Examples

Good first knowledge set

Store hours, shipping regions and times, return window, payment methods, and “how to track my order.” Add product-specific care guides after the basics work.

Troubleshooting

AI answers are vague or off-topic.

Check that documents finished processing, remove conflicting articles, and ask the same question again. Prefer explicit FAQ phrasing in the source content.

AI never hands off to a human.

Confirm handoff / escalate behavior is enabled for the channel and that agents are online or queued. Test with an explicit “talk to a person” request.

Answers mention features we do not offer.

Audit uploaded docs for outdated marketing copy. Delete or revise those sections; AI should cite your current knowledge only.

Frequently asked questions

Does AI replace human agents?

No. AI handles repetitive questions from your knowledge base. Humans remain essential for exceptions, empathy-heavy cases, and anything not documented.

Can AI work in Arabic?

Yes. Provide Arabic source content when your customers ask in Arabic, and use the Arabic admin/widget experience for bilingual teams.

How is this different from OTP API?

Knowledge-base AI answers customer questions on messaging channels. OTP API is a developer verification endpoint — it is not an inbox AI feature.