Every conversation handled.
Every dollar approved.
An AI layer on top of the WhatsApp and email you already use — no migration — that answers routine questions automatically, and always puts a human in the loop the moment a conversation touches money.
- No migration
- Works on WhatsApp + email you run today
- 100% gated
- Every money action, until explicitly promoted
- 2–4 wks
- Typical onboarding, core build
Conversation feed
● Example- Answered
"What time does the clinic open Saturday?"
CHT-88201 - Confirmed
"Can you confirm my RSVP for the annual dinner?"
CHT-88207 - Awaiting approval
"Can I get a refund on my membership fee?"
CHT-88213 - Money actions are approval-gated by default, regardless of intent confidence.
- Answered
"Where's my order #4521?"
CHT-88219
Your team fields the same questions over and over — RSVPs, order status, opening hours, fee reminders — on WhatsApp and email, one conversation at a time. The obvious fix is a chatbot. The trouble is most chatbots answer everything with the same confidence, whether they actually know or not.
That's fine for "what time do you open." It's not fine for "can I get a refund" or "waive my late fee." The moment a conversation touches money, a wrong automated reply isn't a minor bug — it's a trust event with a real member, customer, or resident.
One layer between your conversations and your systems
CXIQ works inside WhatsApp and email without replacing your CRM. Money actions follow an auditable path to a human.
Customers & Conversations
Wherever a message arrives
- Web chat
Submitted
Algorims CXIQ
Amazon Bedrock · Lex v2 · Amazon Connect
- Understand
- Approve
- Respond
- Escalate to voice
or Routed to
Reviewer
CRM, Voice & Audit Trail
Where the record lands
- CRM lookup
- Amazon Connect handoff
- Append-only audit log
Every step is scoped, logged, and reversible. Voice escalations carry the full conversation context.
Generic chatbots answer confidently. That's the problem.
- Repeat questions — RSVPs, order status, hours, fee reminders — consume hours of staff time on WhatsApp and email every week.
- A wrong automated reply is fine for routine questions and dangerous the moment it touches a refund or a fee waiver.
- Without a defined approval path, automation itself becomes the thing that costs a community org or SME real trust — and real money.
Most conversations resolve instantly. The moment money is involved, CXIQ deliberately slows down.
Message arrives on WhatsApp or email
No new app for your customer or member to install — CXIQ works inside the channels already in use.
Bedrock Claude extracts intent
Matched against a per-intent, confidence-scored FAQ knowledge base — no ML training pipeline to maintain.
High confidence, no money involved
Order status, hours, bookings, and general questions are answered automatically and logged to an append-only audit trail.
Money or account action detected
The conversation always routes to a human for approval — a rule that never bends until a client explicitly promotes it to auto-run.
Understand what's being asked. Approve what involves money. Connect to the channels you run. Audit everything, always.
Amazon Bedrock (Claude)
Intent extraction against a confidence-scored FAQ knowledge base — no custom ML training pipeline.
Config-not-code rules engine
Your team edits what's automated without engineering tickets.
Amazon Connect
Voice escalation for complex queries, with full conversational context carried over.
Does CXIQ ever auto-approve a money action?
No. Refunds, fee waivers, and other money actions are always routed to a human, until a client explicitly promotes a specific rule to auto-run.
Do customers need to install a new app?
No. CXIQ works inside the WhatsApp and email channels already in use — there's no migration.
Who is CXIQ built for today?
Community and association groups and SME commerce/service businesses are the proven and validating tracks; enterprise contact centers with an existing CRM are supported as a dedicated engagement.
Where it fits
Two stacks, one product
Most deployments need the core build. Enterprise contact centers can add managed NLU, multi-step agents, and voice escalation without changing the trust rules.
Core · community and SME
A 2–4 week build using Bedrock Claude, per-intent confidence scores, and editable rules over existing WhatsApp and email. The community and association track has a proven 10–15 hours per week recovered.
SME commerce and service
Order status, appointments, FAQ deflection, and payment reminders for clinics, salons, F&B, and commerce teams. This track is modeled and currently validating with design partners.
Enterprise contact center
A scoped engagement for teams with an existing CRM and 5,000+ contacts per month. Optional Lex v2, Bedrock Agents, and Amazon Connect carry full context into voice escalation. This track is in validation.
Deployment commitments
- Repeated low-confidence extraction from the same person escalates to a human rather than looping or guessing.
- An unrecognized phone number follows a defined identity-resolution flow; automated RSVPs, reminders, and bookings have a human-visible undo path.
- English plus Mandarin, Malay, Tamil, or Hindi is scoped upfront. Consent and opt-out are explicit and designed for PDPA/DPDP compliance.
- Every automated reply and human decision enters an append-only audit trail. Idempotency prevents duplicate RSVPs or double-sent reminders.
The architecture
A defined path from intake to decision
- 01
Receive
- 02
Understand
- 03
Ground
- 04
Gate money actions
- 05
Respond or escalate
Let AI handle the routine. Let your team handle what matters.
Tell us which channels you run and what a typical week of conversations looks like — we'll show you exactly what CXIQ would automate, and what it would always leave to a human.