Katalyst is not a chatbot. It is an autonomous concierge that resolves the messy half of e-commerce support and hands the rest to your team with the context already written.
Katalyst AI handles the fast, routine work. The Katalyst Neural Engine steps in when sentiment, value, or complexity spikes — all mid-conversation.
Vector search over your product titles, descriptions, and sizing notes means recommendations are tied to your actual inventory, not the model's training data.
Issue prepaid labels, detect the first carrier scan, and release refunds — all before the parcel reaches your warehouse.
When a case needs a human, your supervisor receives a full brief: what happened, customer value, emotional state, and a recommended action.
Cluster repeated complaints into actionable alerts: defect spikes, carrier delays, or SLA breaches — not just another ticket queue.
A branded chat widget that inherits your colours, fonts, and voice. Drop it in via iframe or a one-line JavaScript snippet.
Products, policies, promotions and resolved cases are ingested into one embedded knowledge graph and refreshed automatically on your chosen cadence.
A phone agent with human prosody, instant barge-in handling and caller-ID memory that greets repeat customers by name and order.
Sensitive actions are gated behind knowledge-based checks and one-time passcodes delivered over SMS or email.
Watch every active call and chat as it happens, whisper private coaching to the agent, or take the conversation over mid-flight.
Every conversation is scored against a rubric and summarised with issue flags, so review takes seconds instead of a full read-through.
Shopper language is detected on each message and answers are generated natively in that language across chat and voice.
Bring your own Shopify, Stripe, EasyPost, Twilio and Resend keys for real catalog sync, refunds, labels, SMS and email — or stay in sandbox.
Run Katalyst as a booking concierge instead of a store concierge: your intake questions, lead qualification, real calendar availability and confirmed appointments.
Parallel agents scale to your line limit so no shopper or caller waits, with real-time metering of every active chat, text and call.
A texted link turns hold into a screen: live wait clock, queue position, and a station carousel that swaps the audio on the live phone line without dropping the call.
Mood chips, hold duration, station skips and tap telemetry combine into one weighted score that routes hot callers straight to a supervisor.
Callers confirm their order number and ZIP while waiting, and the supervisor screen-pop opens with a verified order, the core problem, and a ready opening line — or the caller skips the queue with a callback.
Screens inbound numbers before an agent picks up, blocking known robocallers and debt-collection scripts and quarantining the rest.
Record a short sample in your browser and clone it, so the phone concierge speaks in your own voice — or a natural sandbox voice until you connect a key.
Decide how much the concierge may do alone. In co-pilot mode it drafts and waits; in autopilot it acts within the caps and confidence floor you set.
Test edge cases before customers do, watch an accuracy scorecard over time, and turn plain-English feedback into rules the agents follow.
Katalyst staffs your store with a roster of specialists. Each conversation is classified by department and routed to the right agent, who can warm-transfer upward when the case outgrows their authority.
Departmental roster
Styling, logistics, returns and billing specialists — each with their own name, avatar, title and voice.
Tiered escalation
Frontline agents hand off to a team lead, and the team lead escalates to your human supervisor with a dossier.
Per-agent tool permissions
Decide which agent may issue labels, look up orders, or release a refund. Nothing else is callable.
Concession caps
Every agent carries its own refund authority, capped by your store-wide ceiling. Above it, a human approves.
Shoppers see the transfer as a friendly "X joined the chat" card, and your inbox shows the full handoff trail so supervisors know exactly who touched the case and why.
Autonomy is a dial, not a switch. Every autonomous decision is measured, capped and auditable, so you can start in approval-first mode and open it up as the accuracy scorecard earns your trust.
Automation level
Co-pilot drafts every reply and action for your approval. Autopilot acts on its own inside the limits you set.
Caps and ceilings
Refund, concession and per-conversation spend caps are hard limits the agents cannot cross without a human.
Confidence floor
Below your floor the concierge refuses to guess, defers to a human and files the gap for you to answer once.
Audit log
Every autonomous action, guardrail trip and simulated edge case is recorded with the reason it was taken.
Every conversation is measured, every escalation is timed against the SLA you set, and every shopper is invited to rate the concierge. Nothing is estimated — the numbers come from your own data.
Resolution rate
Share of conversations closed without a human, tracked daily over 7, 30 or 90 days.
CSAT ratings
Shoppers rate the chat 1–5 with an optional comment; averages and detractors surface in reporting.
SLA compliance
Median and worst supervisor response times, met vs breached, plus a live at-risk queue.
CSV export
Export the selected range — headline metrics, daily volume, intents and raw ratings.
Create a free store, run the onboarding wizard, and chat with the live demo widget in minutes.
