Proof
We have no customers to quote yet. So here is the machinery instead.
Every platform in this category publishes a resolution rate. It is the vendor's number, about the vendor's product, measured by the vendor. This page does the opposite: it shows you the parts you can inspect — a run, step by step; how a release is scored; what stops the agent; who signed off; and what gets counted.
What is missing from this page, and why
- No customer logos
- We have none we can name. A logo wall assembled from pilots, investors or stock brands is the oldest lie in B2B software, and you would find out.
- No testimonials
- When a customer is willing to put their name to a sentence, it goes here with their name and their job title on it. Until then, nothing.
- No resolution rate
- Ours would be a number about our product, measured by us, on conversations we chose. The number worth having is the one you get when you replay your own past conversations through an evaluation. That is a feature, not a billboard.
A run, end to end
Every reply leaves a record you can open.
Scout's Activity lists every run with its outcome and a sanitised trace. This is one run: a customer asking where an order is, answered without a person. Each row is what actually happened, in order, with what it read and what it called.
- Disagreeing with the agent becomes an argument about evidence
- Not about impressions. The row that caused the problem is in the record, with the source it read and the result it got back.
- Runs are separated by mode
- Live, Sandbox and Check. A test run and a customer-facing run are never confused with each other, and a check run is never billed.
- Skipped runs say why
- Scout turned off, a person already assigned, the conversation already handed over, the channel's reply window closed. Silence is always explained.
Every run ends in one of these
- Running
- Answered
- Handed over
- Skipped
- Failed
- Asked a question
- Waiting for approval
Illustrative demo with fictional data.
Scoring
A release is scored before it ships, and a failing score blocks the release.
Checks are saved test questions. They run automatically against the draft before every publish. If any check fails, publishing is blocked — not warned about, blocked.
What one check can require
- What should happen: Scout answers, or Scout hands over.
- The scenario that should have matched.
- The tool that should have been called.
- The knowledge source the answer should have come from.
- Phrases the answer must include.
- Phrases the answer must not include.
Three checks, as they appear
| Test question | Expected | Result |
|---|---|---|
| What's your return policy? | Scout answers | Passed |
| I want to speak to a person | Scout hands over | Passed |
| Do you offer gift wrapping? | Scout hands over | Passed |
And a published version can be undone. Every publish creates a numbered version with a note on what changed. Restore an earlier version and the current draft is preserved. A regression is a rollback, not an incident.
Before any of that, the playground. A private test chat against the draft. Nothing reaches a customer and tools are never called for real. Select any reply to see what Scout did: knowledge search, scenario selection, guardrail, AI decision, safety check, tool and handover.
Limits
The things Scout is not allowed to do, and what happens when it tries.
A guardrail is a rule plus an action. You write the rule; the action decides what the platform does when the rule matches. Platform safety rules stay on even if you configure none of your own.
Four actions
- Block and hand off
- Blocks the reply and hands the conversation to your team.
- Ask for clarification
- Keeps Scout in the conversation, but makes it ask one clarifying question instead of answering.
- Escalate
- Hands the conversation to your team immediately.
- Record only
- Logs the match in the run's activity and leaves the reply unchanged.
Ready-made guardrails
| Rule | Action |
|---|---|
| Never promise a guaranteed refund | Block and hand off |
| Escalate stolen card and fraud reports | Escalate |
| Never reveal internal pricing or cost | Block and hand off |
| Hand off legal and medical requests | Escalate |
| Ask for an order number first | Ask for clarification |
| Record competitor mentions | Record only |
Guardrails are testable
Run a message through them before it is live.
“The lamp arrived broken. I want my money back today.”
Final action: Block and hand off. Reason: Never promise a guaranteed refund.
Sign-off
Anything that changes something waits for a person.
When Scout wants to use a tool that changes state — adding a private note, adding a tag, assigning a team, resolving a conversation — it does not do it. It asks. A reviewer sees the exact requested operation, can leave a comment, and chooses approve and execute, or reject.
The approval record
Every request carries a status, and the status is part of the run's permanent record.
- Pending
- Approved
- Rejected
- Expired
- Executed
- Execution failed
- Revoked
- Requests expire, and are rechecked
- A request nobody reviews expires rather than sitting there, and it is checked again before it runs, so a stale approval never executes against changed facts.
- The customer is not left hanging
- While a request waits, Scout can send a waiting-for-approval message that you wrote, so nobody is on read while a person decides.
One request, as the reviewer sees it
Pending- Run 1042
- Conversation #4790
Requested operation: add_private_note
Customer is asking for a same-day refund on a damaged lamp. Guardrail blocked a promise; flagging for the team.
Illustrative demo with fictional data.
Measurement
What gets counted, and what deliberately does not.
Analytics summarises the work over a date range. These are the measures the product actually keeps — not a dashboard mock-up, and not a promise about your results.
Measured
- Daily conversation volume
- Conversations handled
- Answer rate
- Handoff rate
- Average response
- Knowledge gaps
- Tool failure rate
- Provider errors
- Outcomes
- Approvals and average review time
- Scenario frequency
- Guardrail actions
And before go-live, a readiness list
- Assistant identity configured
- Approved knowledge available
- At least one inbox selected
- AI provider configured
- Required evaluations created
- Evaluations passing
- A version is published
- Assistant is active
Not measured, on purpose
- An estimated cost figure
- Analytics shows what Scout handled, not what it supposedly saved. A saving is a claim about your business that we are not in a position to make.
- A published accuracy or resolution rate
- Your results depend on your knowledge and your setup. The honest version of that number is the one your own evaluations produce.
- Sentiment
- Scout does not score how a customer feels. Handoff rate and guardrail actions tell you more and invent less.
Market facts
Numbers we are willing to publish, with their sources attached.
Six, all from regulators or from Meta's own documentation. Each one is printed with the source that produced it, so you can go and check it rather than take our word for it.
92.7%
of Saudi internet users use WhatsApp, making it the country's most-used app.
CST, Saudi Internet Report 2025 (national regulator)
#1
WhatsApp Business was the UAE's most-downloaded app, ahead of TikTok.
TDRA, Digital UAE 2023 (national regulator)
1 Oct 2026
Meta began charging for service messages after two years free, with 1,000 free per business number per month.
Meta for Developers, WhatsApp Business Platform pricing documentation
~18×
An Egypt marketing template costs $0.0644 against $0.0036 for a utility template carrying the same sentence.
Meta USD rate card effective 1 October 2026, transcribed 2 November 2026
24 hours
The customer service window. Outside it you can only send an approved template.
Meta for Developers, WhatsApp Business Platform documentation
Required
Meta's policy requires businesses using automation to offer a path to a human.
WhatsApp Business Messaging Policy, updated 23 September 2026
The refusal list
What we will not claim.
These numbers circulate freely in this market and several of them would flatter us. We cannot trace any of them to a source that survives a check, so they do not appear on this site — and if you see them on a competitor's page, now you know what to ask for.
A cash-on-delivery share for Egypt
Vendor blogs put it anywhere between 34% and 70% for the same period. A range that wide is not a statistic, it is a vibe.
A return-to-origin reduction percentage
Every number we found traced back to a vendor's own marketing page with no methodology attached.
“30–40% of support tickets are order-status questions”
Widely repeated, universally uncited. Your own analytics will tell you your real figure within a week.
“200 million WhatsApp Business monthly active users”
A figure from an old announcement that has been recycled far past its shelf life.
Any competitor review quote attributed to this region
The review sites that would substantiate one block automated access. We could not verify a single regionally-attributed competitor review, so we quote none.
A certification we do not hold
No badge appears on this site until there is an auditor's report behind it. What is in place and what is in progress is written out on the trust page.
The fastest way to verify this is to run it.
Bring your own past conversations, put them through an evaluation, and read the traces. That produces a number about your business rather than ours.