Support software should carry the load, not add to it
We built Sanad because support teams were drowning in tabs — one tool for email, another for chat, a third for WhatsApp, a wiki nobody updates, and a spreadsheet pretending to be a report.
Meanwhile “AI support” came to mean two things nobody asked for: a chatbot that invents answers, and a vendor that sends replies in your name. We think both are product failures, not inevitabilities — so we designed them out.
Every workspace is its own database with its own keys — from the first commit, not from the enterprise tier.
Fourteen capabilities that draft, cluster, predict and score. Every one of them produces a suggestion.
A person approves before anything reaches a customer, a knowledge base or a scorecard.
AI assists.
Humans decide.
Not a positioning line — a constraint we build against. It is why there is no auto-send switch anywhere in this product, and why the customer-facing bot would rather hand you to a person than guess.
Four defaults we turned down
Every one of these is a normal thing for a help desk to ship. Each is a decision we made the other way, and each is checkable in the product.
Trained on your docs, confident about everything, occasionally inventing a refund policy you never had.
The customer-facing bot answers only from articles you published, with the source attached. No article, no answer — it hands off to a person instead of guessing.
Deflection · every answer carries its sourceThe vendor decides your customer does not need a human today, and sends something on your behalf.
Drafts, scores and articles wait for a human decision. There is deliberately no auto-send switch anywhere in the product — not off by default, absent.
Assist · a draft is a draftA shared table with a tenant column, and real separation quoted at you when you ask for it.
Own database, own encryption keys, from the Starter tier up. Deleting a workspace drops its database and destroys its keys for good.
Tenancy · separation is not a line itemA rubric applied to an agent’s week by a system they cannot see, question or answer back to.
Agents see the rubric, the sample and the reasoning behind each score — and can open a dispute that a human resolves.
QA · the loop closes both waysWhat is actually shipping
- Channels
- Email, live chat, WhatsApp, Messenger, Instagram and Telegram — one conversation spine, threading that survives a channel switch.
- Self-service
- A help centre and a bot that answers with citations, or hands off to a person.
- Operations
- SLA clocks that run on business hours, automation rules, routing, approvals and side conversations.
- Knowledge
- Articles drafted from solved tickets; unanswered questions clustered into gaps a human approves before publishing.
- Insight
- Event-fed dashboards, CSV exports, CSAT prediction and rubric-scored quality review with a dispute loop.
- Extensibility
- A full REST API and an app marketplace with per-install, instantly revocable access.