Whatsteam customer support

WhatsTeam for Customer Support

A complete playbook for running customer support on WhatsApp with a team — from the moment a customer messages in, to ticket resolution and reporting.

When to use

Activate when the user:

• Operates a support team, helpdesk, customer service, or success function on WhatsApp.
• Asks about SLAs, first-response time, resolution time, escalation, CSAT, tickets.
• Mentions auto-reply, business hours, out-of-office, FAQ, bot handoff.
• Wants routing by language, product, region, or department.
• Asks "how do I report on what my team is doing?"

For outbound sales / lead qualification, route to whatsteam-sales-bot. For broadcasts, use whatsteam-campaigns. For order-related support inside an e-commerce flow, see whatsteam-ecommerce Recipe 5.

Prerequisites

• whatsteam-setup complete.
• Team members configured at https://app.whats.team/team.
• (Recommended) Departments configured at https://app.whats.team/device/departments (Tier 1, Billing, Tech, …).
• (Optional) A knowledge base / FAQ source for deflection.
• For WABA: pre-approved Utility templates for out-of-hours, escalation acknowledgement, ticket-closed-feedback.

The support lifecycle

NEW ─► PENDING ─► IN PROGRESS ─► WAITING ON CUSTOMER ─► RESOLVED ─► CLOSED
            │           │              │
            └──────► escalate ──► assigned to senior agent / dept

Map this to WhatsTeam primitives:

Lifecycle stage	WhatsTeam state
NEW / PENDING	chat.status = active, no assignedTo
IN PROGRESS	chat.status = active, assignedTo = agent
WAITING ON CUSTOMER	label waiting-customer, status still active
RESOLVED	chat.status = resolved
CLOSED	chat.status = archived (no further action expected)
ESCALATED	label escalated + reassign

Recipes

Recipe 1 — Auto-reply on first contact

"When a brand-new customer messages us, send a greeting and tell them what to expect."

on message:in:new where chat.messageCount == 1:
  if within_business_hours():
    reply: "Hi! Thanks for reaching out. An agent will respond within 15 minutes."
    label chat "new-contact"
  else:
    reply: "Hi! Our hours are 9am-7pm CET, Mon-Fri. We'll reply first thing tomorrow."
    label chat "out-of-hours"
  ack 200

Use your own DB to track "first message" — chat.messageCount is not a reliable first-contact signal (it may include outbound messages, depending on how the chat was created). For "is this the first inbound?" logic, track first-inbound in your own DB, or fall back to chat.lastInboundAt (which exists) being unset before this event.

Recipe 2 — Business hours + holiday calendar

function within_business_hours():
  now = utcnow()
  local = now.in_timezone("Europe/Madrid")
  if local.weekday() in [Saturday, Sunday]: return false
  if local.date in holiday_calendar: return false
  return 9 <= local.hour < 19

# On message:in:new outside business hours:
  send template "out_of_hours" with [agentName, next_business_day]
  label chat "out-of-hours"

Don't auto-reply twice to the same out-of-hours window for the same chat. Track last-auto-reply timestamp per chat.

Recipe 3 — Route to the right department

"Messages mentioning billing → Finance dept. Messages mentioning technical issues → Tech."

on message:in:new:
  intent = classify(message.body)   # rule-based or LLM
  if intent == "billing":
    assign chat to Finance dept
    label "intent:billing"
  elif intent == "technical":
    assign chat to Tech dept
    label "intent:technical"
  else:
    assign to Tier 1 round-robin

Use the chat's analyze_whatsapp_chat_messages tool for LLM-based intent extraction. Cache the classification on the chat so subsequent messages skip the LLM call.

Recipe 4 — SLA tracking (first response time)

Define SLAs per chat tier:

Tier	First response	Resolution
Standard	30 min	24h
Premium (label vip)	10 min	4h
Enterprise	5 min	2h

Implement via scheduled job:

every 1 min:
  pending_chats = get_whatsapp_chats_by_status(active) where assignedTo is null
  for chat in pending_chats:
    sla = sla_for(chat.labels)
    age = now - chat.firstInboundAt   # see caveat below
    if age > sla.first_response * 0.8 and not chat.sla_warned:
      ping_team_in_slack("⚠️ Chat with {customer} is {age}min old, SLA breach in {remaining}min")
      mark chat.sla_warned = true
    if age > sla.first_response:
      ping_team_in_slack("🚨 SLA BREACHED on chat with {customer} ({age}min)")
      label chat "sla-breach"

Field caveat: chat.firstInboundAt is not a reliable field — for first-contact / SLA-start logic, use chat.lastInboundAt (which exists, returned by get_whatsapp_chats action:by_id) or stamp the first-inbound time in your own DB on the message:in:new webhook. Don't assume firstInboundAt or messageCount are populated.

Recipe 5 — FAQ deflection

Before routing to a human, try to answer common questions:

on message:in:new (with bot tag enabled):
  faq_match = match_faq(message.body, threshold=0.8)
  if faq_match:
    reply with faq.answer + "Did that help? Reply YES or type AGENT to talk to a person."
    label chat "faq-attempted"
    if next inbound is "AGENT" or negative sentiment:
      remove bot tag, assign to human
  else:
    assign to human directly

Deflection reduces team load 30-50% on commodity questions. Track success rate: chats labeled faq-attempted AND ending in resolved without human assignment.

Recipe 6 — Escalation

"If an agent hasn't replied in 2 hours, escalate to senior."

every 5 min:
  in_progress = get_whatsapp_chats_by_status(active) where assignedTo not null and labels not include "resolved"
  for chat in in_progress:
    last_outbound = last message from any agent in chat
    if (now - last_outbound) > 2h:
      reassign chat to senior_agent_or_team_lead
      label "escalated"
      notify both original and new owner in Slack

Set the escalation threshold by tier. Don't auto-escalate VIPs to the same junior agent twice in a row.

Recipe 7 — Resolution + CSAT survey

When agent marks chat as resolved:

1. set chat.status = resolved
2. wait 5 minutes (let the last message land)
3. send template "csat_survey":
     "How would you rate this support experience?
      Reply 1 (bad) to 5 (excellent)."
4. on next message:in:new where chat.status == resolved:
     parse rating
     save to CSAT DB
     if rating <= 2: alert team lead for follow-up

Don't send the CSAT request more than once per chat. Track sent-at timestamp. This loop must respect the auto-reply suppression from Anti-patterns ("Auto-replying after the first message of a thread") — gate the CSAT send on bot-still-active / no-human-in-thread so the survey and a first-contact greeting don't double-fire on the same inbound.

Recipe 8 — Reporting

Daily / weekly digest queries:

- Open chats by department:
    for each dept: get_whatsapp_chats_by_status(active) filtered by dept
- Avg first response time today:
    get_whatsapp_chat_statistics with dateRange=today
- Top agents by resolved chats:
    for each agent: count get_whatsapp_chats_by_status(resolved) where assignedTo=agent
- SLA breach count:
    count chats with label "sla-breach" today

Render to a Slack message, a Google Sheet, or a Notion database via a scheduled job.

Anti-patterns

• Auto-replying after the first message of a thread. Once the human is in, the bot should be silent. Only the first message (or after WAITING for >Xh) should trigger an auto-reply.
• Closing chats too aggressively. Mark resolved only when the customer confirms (or after a clear "thank you"). Premature close kills CSAT.
• No labels = no reporting. If you can't query chats by topic, intent, or status, you have no visibility. Label aggressively, even if just intent:*.
• SLAs without escalation. SLAs that fire alerts but don't actually reassign are theater. Wire escalation to the same SLA timer.
• Mixing sales and support in one inbox without labels. Sales reps optimize for revenue; support optimizes for resolution. Label every inbound so reports can split correctly.

See also

• whatsteam-inbox — the lifecycle primitives (status, assignment, labels, departments).
• whatsteam-webhooks — the event source for all reactive logic.
• whatsteam-messaging — message + template construction.
• whatsteam-sales-bot — the sales-side counterpart, with handoff rules.
• Reference implementation: https://github.com/whatsteam/whatsapp-chatgpt-bot
• Multi-agent support article: https://whats.team/blog/multi-agent-whatsapp-support
• Scale support article: https://whats.team/blog/how-to-scale-whatsapp-support-without-hiring-more-staff