200+ daily field appointment tickets. A WhatsApp message handles the routine. An LLM-powered voice call handles the exceptions. Staff handle only what actually needs a human.
200 outbound calls made by staff daily — one per ticket.
Most calls were routine confirmations indistinguishable before dialling.
Staff had no way to triage — every ticket looked the same upfront.
Voice calls are the most expensive channel — cost compounded at scale.
Real escalations were delayed because staff were buried in routine calls.
Process collapsed under leave or high-volume days.
Three tiers. Each tier filters out what the next tier doesn't need to handle.
At the scheduled time, an automated WhatsApp message is sent to each field partner asking if they've reached the site. Partners who reply YES are logged and the ticket closes. No staff involvement, near-zero cost. The majority of tickets resolve here.
Triggers Tier 2: on NO reply or no response within window.
Only partners who flagged a problem or didn't respond get a voice call — from an LLM bot, not a human. It conducts a natural two-way conversation, probes for specifics, and logs everything as structured data. No menus, no "press 1 for yes." The bot understands natural speech.
Triggers Tier 3: only when bot determines human judgement is needed.
Staff only enter the process when the bot has determined a situation warrants it. By the time a coordinator picks up, they have full context: what the issue is, what the partner said, what action is needed. They resolve real problems — not routine check-ins.
Volume: ~30% of original call volume — all meaningful.
| Metric | Before | After |
|---|---|---|
| Staff calls daily | ~200 — every ticket, manually | ~60 — escalations only |
| Employee effort | Full coordination time on routine calls | 70% reduction |
| Per-transaction cost | High — voice call for every ticket | Low — WhatsApp for majority, voice only for exceptions |
| Issue capture accuracy | Dependent on individual staff notes | Structured, consistent, auto-logged |
| Escalation response time | Delayed — staff tied up on routine calls | Immediate — staff free when issues surface |
| Resilience | Degraded under leave or high-volume periods | Fully automated — unaffected by staff availability |
"Start with the cheapest intervention. Escalate only when the situation requires it."
By making WhatsApp the first touchpoint and voice the exception, per-transaction cost dropped dramatically — with no loss in coverage or issue detection.
Most transactions are routine. Handle them cheaply. Reserve the expensive channel — voice call, human time — for the minority that genuinely need it. This principle applies to any process with high transaction volume and variable complexity.
WhatsApp first: "Confirming your 3pm booking tomorrow — reply YES or NO." Only non-responses or reschedule requests trigger any further action. The majority confirm instantly.
Automated WhatsApp to a missed client — did something come up, would you like to rebook? Only if there's no reply does it escalate to personal outreach.
A WhatsApp message after each visit. Positive response triggers a review request. An issue flags for personal follow-up. Good signal routing at zero staff effort for the majority.
Preferences, health notes, and consent gathered via WhatsApp before arrival. Therapists start with context instead of a clipboard. Complex cases are flagged for a pre-visit call.
90-minute AI Opportunity Audit. No pitch — just findings you can act on.