How It Works
A four-step loop that keeps humans in charge.
Sioona sits on top of the chats you already have. It finds the signals, your team reviews them, and the Signal Score tells you whether it’s ready to trust.
The four steps
Upload an export from Intercom, WhatsApp or a CSV. Sioona redacts sensitive details on ingest. You don’t need live integrations or credentials to start.
Each conversation is scanned for lead, retention and upsell signals. Every detection keeps the exact words that triggered it.
Your team marks each signal Correct, Incorrect or Partially Correct. Those decisions are what produce the Signal Score.
Confirmed signals come with a Suggested Next Action. Your team follows up in the tools it already uses. Nothing is sent from Sioona.
01
Connect Chats
Upload an export from Intercom, WhatsApp or a CSV. Sioona redacts sensitive details on ingest. You don’t need live integrations or credentials to start.
02
Detect Signals
Each conversation is scanned for lead, retention and upsell signals. Every detection keeps the exact words that triggered it.
03
Human Review
Your team marks each signal Correct, Incorrect or Partially Correct. Those decisions are what produce the Signal Score.
04
Act
Confirmed signals come with a Suggested Next Action. Your team follows up in the tools it already uses. Nothing is sent from Sioona.
Shadow Mode
Detection that runs beside your process, not instead of it.
Sioona is not a chatbot. It never talks to your customers.
- Sioona detects signals and shows the evidence.
- A person reviews each one before anything is acted on.
- Suggested Next Actions are display-only. Follow-up happens in your own tools.
- Marking a signal Acted records that your team followed up. It doesn’t send anything.
Product Surfaces
Four places your team works.
Each surface has one job, so a review session stays fast.
Insights
The review queue. Pending, needs follow-up and done, with the transcript, the words that fired each signal, and a Suggested Next Action.
Upload
Bring chats in from Intercom JSON, WhatsApp Business text exports, or CSV and JSON files. Sensitive details are masked on ingest.
Eval
Signal Score, measured as your team reviews. The go or no-go view for a pilot.
Taxonomy
The phrases and keywords behind each signal type, with negative phrases that rule out noise. Tuned from review feedback.
A Two-Week Pilot
What a pilot looks like.
A typical cadence. We agree the details with each team before you start.
Day 0
Share one redacted export
Tell us how many opportunities you find today. That’s your baseline.
Days 1–3
Review in Insights
Your team marks signals Correct, Incorrect or Partially Correct, and marks the confirmed ones Acted.
Around day 7
Add a second export
Upload new messages from the same conversations so the queue keeps moving.
Days 10–14
Read the Signal Score
Compare your score to the agreed threshold and your baseline, then decide whether to continue.
Pilot scope and evidence
A pilot starts from one redacted chat export and runs for two weeks. Scope and price are agreed before it starts. No public list price is published.
Supported exports: Intercom JSON, WhatsApp Business text exports, and CSV or JSON files. No mailbox access, chat-tool API keys or CRM credentials are required.
Every signal shows the exact source words behind it. Evidence anchors identify the source message and text span so reviewers can reconstruct the evidence from persisted source text.
Signal Score is precision measured against human review: correct signals divided by reviewed signals. Pilot default: go at 80% or higher, with at least 5 reviews. Thresholds are agreed with each team.
Shadow Mode never sends messages to customers. Suggested Next Actions are display-only; humans decide and follow up in their own tools.
Request a pilot
Find out what your chats already know.
Start with one export. Review the signals with your team, and let the Signal Score tell you whether to go.