Skip to content
BlendDesk
Book a demo

Platform

The customer operations platform for teams that run on messaging.

One conversation, four stages. Answer it in a shared inbox, Resolve it with Scout, Act on the order or booking behind it, then Improve on what happened. Every stage leaves a record you can open.

Book a demoOpen the shared inbox
One conversation moves through four stages: it is answered in the inbox, resolved by Scout against approved knowledge, acted on with a live order lookup, and scored afterwards in an evaluation.

Illustrative demo with fictional data.

Answer

Every message about one customer lands in one thread.

WhatsApp, Instagram, Messenger, email and web chat arrive in the same conversation for the same person, with their orders, previous threads and custom fields attached before anyone types. Each channel keeps its own behaviour — delivery and read receipts on WhatsApp, quoted replies on email — instead of being flattened into a generic ticket.

  • Inbox

    Queues for Mine, Unassigned and All, assignment by rule, private notes, and a handoff that carries the whole thread to the next person.

  • Channels

    What each channel needs before it will connect, what Meta charges to message on it, and how the 24-hour service window changes what you are allowed to send.

  • Knowledge

    The approved answers your team and Scout both read, so a customer gets the same sentence whichever one replies.

The 24-hour customer service window is Meta's rule, set out in Meta's WhatsApp developer documentation.

Illustrative demo with fictional data.

Resolve

An AI agent you can rehearse, limit and audit.

Scout answers from knowledge a person approved, and quotes the passage it used. Before it talks to a customer you rehearse it against real questions in the playground and declare, case by case, how it should behave. Guardrails and approvals decide what it may do alone. Afterwards the run is on the record: what it read, what it decided, which tool it called.

  • Scout

    Retrieval with citations, a confidence threshold, the handoff rule — and the playground where you test a change before anyone sees it.

  • Scenarios

    Write down how a case is handled: the questions it covers, the steps to take, what to collect. Scout follows it instead of improvising.

  • Guardrails & approvals

    Hard limits on what Scout may say and do, an approval step for anything that touches money, and a named person at the end of the escalation path.

Scout's own view of the run it just finished. Each row opens.

Illustrative demo with fictional data.

Act

The work behind the question actually gets done.

Looking up an order, confirming a cash-on-delivery address, recovering a cart, sending a reminder: this is the work where a wrong answer costs money, so it runs as deterministic automations against your real systems instead of a model's guess. Tool calls are real HTTP calls, and they appear in the run with their arguments, their result and their latency.

  • Automations

    Rules and workflows that route, tag, confirm, chase and close — triggered by a message, a status change or a schedule, and editable without a developer.

  • Integrations

    Shopify, your ERP or CRM, and your own HTTPS endpoints, exposed to Scout as tools with a defined input and a defined result.

Illustrative demo with fictional data.

Improve

Score what happened, then change it on purpose.

A resolution rate you cannot inspect is a marketing number. Replay conversations your team has already handled against the agent you are about to ship, read the runs that failed, and watch the operational numbers the AI is supposed to move.

Replay sets
Pick conversations you have already resolved, run the current agent against them, and compare its answer with the one your team actually sent.
Run traces
Open a single run: retrieval scores, tool calls with latency, guardrail checks, approvals, and the exact text that went out.
Operational reporting
Volume, first response time, resolution time and handoff rate by channel, team, tag and agent, with Scout counted beside your people rather than in its own dashboard.
  • Evaluations & traces

    Scored replays on your own conversations, a diff against the live agent, and the trace behind every run.

  • Analytics

    What your team handled, how fast, on which channel — and how often a conversation reached a person.

Why this is the product

Governance is a category expectation, not a feature.

Every vendor in this market publishes a resolution rate about its own product, and none of them let a buyer verify one before signing. The useful answer is not a bigger number. It is five things you can do to an AI agent before and after it speaks to your customer.

Rehearse
Run a draft agent against real questions in the playground and read the answer before a customer does.
Declare
Write the scenario for a case, so the behaviour is a decision you made rather than an emergent property.
Limit
Guardrails bound what Scout may say; approvals stop anything that moves money or changes an order until a person signs off.
Score
Replay handled conversations and grade the runs, so shipping a change is measured instead of felt.
Audit
Every run keeps its trace, and every automated conversation keeps a route to a named human.

A human escalation path out of automation is required by the WhatsApp Business Messaging Policy, updated 23 September 2026.

  • Guardrails & approvals

    The limits, the approval queue and the escalation path, with the off switch in plain sight.

  • Evaluations & traces

    How a run is scored, how a regression is caught, and what a trace actually records.

Engineering

Arabic and right-to-left are engineered, not translated.

Switch a person to Arabic and the workspace mirrors: the conversation list, the thread and the customer panel swap sides, and Arabic, Latin and digits keep their own direction inside a single line. This is a capability we built properly, not a position — teams run BlendDesk in English everywhere else.

  • Interface language is per person, and the workspace mirrors without a second build.
  • Arabic knowledge and Arabizi-tolerant retrieval, so a question typed in Latin letters still matches the right passage.
  • The chat widget and the Help Center flip for your customers too.

Illustrative demo with fictional data.

Go deeper

Questions

What people ask about the platform.

What does customer operations mean here?

Answering a question about an order and changing that order are the same job, so they live in one product: the inbox, the AI agent, the deterministic automations, and the governance around both. The span is the point — an inbox alone cannot act, and an automation platform alone cannot answer.

Do I have to let Scout reply to customers?

No. Scout can run as draft-only, suggesting replies that an agent reads and sends. When it does reply directly, it answers from approved knowledge, stops the moment a person takes the thread, and anything that changes another system waits for an approval you configured.

Which channels can I connect?

WhatsApp through the Cloud API with your own Meta business account and number, website live chat, email over Gmail, Microsoft or IMAP, Telegram, SMS through Twilio, and a generic API channel. Instagram and Messenger are set up with our team and depend on Meta's approval.

What can I see after Scout handles a conversation?

The run trace: the scenario it matched, every passage it retrieved with its score, every tool call with its arguments, result and latency, each guardrail it checked, any approval it waited for, and the final text that was sent.

Does BlendDesk work in Arabic?

Yes. The workspace, the chat widget and the Help Center all run in Arabic with right-to-left layouts, and Scout retrieves from Arabic knowledge. English is the default and nothing about the product assumes an Arabic-speaking team.

Bring your customer conversations together.

Tell us a little about your team. We'll show you how BlendDesk could fit the way you already talk to customers.

Book a demoTalk to us