She doesn’t have to explain twice.

Accuracy comes from grounding and guardrails, not the model alone. bitbybit’s agent answers from your knowledge base, product catalog, and live order data — never the open web — and you prove its behaviour in the AI Studio Playground against your real configuration before any live traffic. Topics you mark sensitive always escalate, and the agent says “I don’t know” and hands off rather than guessing.

You define bitbybit’s escalation rules in AI Studio: confidence thresholds, sensitive topics that must always go to a human, VIP routing, sentiment triggers, and explicit “ask for an agent” intent. When a rule fires, the Escalation skill writes a brief — issue summary, customer signals, what’s been tried, and a suggested reply — so the human reads and responds rather than starting over.

bitbybit counts a conversation resolved when it closes without human escalation and without a customer-initiated reopen, refund, or repeat contact in the following window. Resolution rate, first-response time, escalation rate, sentiment trend, and reopen / refund rates are tracked per agent and per skill — and conversation logs are reviewable in AI Studio, so you can see where the agent went wrong, update knowledge or rules, and re-test before the change goes live.

A real help desk, not just a bot. In bitChat, conversations arrive as tickets organized by category; routine ones resolve on their own and the rest route to the right person by role. The agent respects your operating hours, and when a human is needed the handover is clean — the brief already written.

Yes — bitbybit’s agent works from your tone-of-voice brief, your knowledge base, and your action catalog — refunds, returns, shipping rules, exceptions. Macros aren’t a separate concept; they’re behaviors you configure once in AI Studio and the agent uses everywhere. Updates go live without redeploying.

Yes — bitbybit’s agent can check eligibility, surface the policy, generate a return label, and update the order in Shopify. You decide which actions the agent can take alone and which require human confirmation. Sensitive cases (chargebacks, fraud signals, repeat-issue customers) escalate by rule.

Yes — bitbybit’s agent is bilingual by default and follows your tone-of-voice brief in both languages. It detects the customer’s language from the first message and stays in it through the thread, including the escalation brief if the human prefers a different working language.

Very little — your team opens the bitChat inbox, reads the brief, and sends the reply. The configuration work — knowledge base, rules, escalation logic — happens in AI Studio and is owned by whoever runs CX, not engineering.

bitbybit charges a flat platform plan — bitChat Growth at $99/mo is the anchor, and the all-products $299/mo Pro Bundle adds bitCRM and the rest. Starter and Growth plans carry a 7-day free trial; the Pro Bundle is demo-onboarded. WhatsApp conversation fees pass through at Meta’s published rates with no per-conversation markup — what Meta charges is what you pay. Month-to-month, no minimum commitment. Full details on /pricing.

Either works. Most teams start by routing WhatsApp, Instagram, and web-widget conversations through bitbybit while keeping their existing helpdesk for email and web tickets — the customer record stays unified across both. Teams that consolidate move email and web traffic over once the agent is trusted. No lock-in either way.

bitbybit is an Official Meta Tech Partner — WhatsApp messages run on Meta’s Cloud API, under Meta’s security and compliance posture. Customer records sit in your workspace; full export is available at any time. Conversation logs and configuration changes are auditable in AI Studio. For privacy, security, and data-processing questions, our team can share the relevant documentation during review.

A support-only inbox stops at “conversation closed.” bitbybit shares one customer record across support, marketing, and CRM — so a CS tag becomes a Meta custom audience and a resolution feeds the next campaign. The agent that resolves the issue is the same one that recognizes her next time. Side-by-side comparisons with the support-shaped tools live at /compare.