Perla for Customer Support: 24/7 Tier-1 Coverage on WhatsApp

The setup

A small consumer brand running customer support over WhatsApp — the channel their customers actually use — was hitting the same wall every founder hits: messages come in at 11pm, on weekends, in the middle of a flight. Each unanswered message is a small loss of trust. Hiring a 24/7 human team is a five-figure monthly line item that doesn't make sense at their scale.

They hired Perla as their first AI employee for tier-1 support.

This use case is anonymized — composite of the early-access teams using Perla for this exact pattern. It's published as a template for what the deployment looks like, so future named customers have a shape to slot into.

What Perla does

She runs as a participant in the brand's customer-facing WhatsApp number. For each incoming message:

1. Classifies intent — sales question, support question, complaint, spam.
2. Pulls the relevant context from the organization knowledge base — product info, return policy, shipping cutoffs, brand voice.
3. Drafts and sends a reply in the brand's voice, or
4. Routes to a human when the message is ambiguous, high-stakes (refund, complaint), or hits an explicit handoff rule.

She runs on schedule too — once a day she compiles a 5-bullet summary of the previous 24 hours of conversations and posts it to the founder's private channel. No dashboard logins required.

What it replaces

It doesn't replace a tier-2 specialist or a human relationship-builder. It replaces:

• The first 30 seconds of every customer message (the acknowledge + classify step)
• The 80% of messages that have a known answer
• The night-and-weekend coverage that was previously "we'll get back to you Monday"

The pricing comparison the team ran:

What surprised them

Three things they didn't expect.

Brand voice was the biggest win, not speed. They had a VA before. The VA was fast. The VA also wrote in their own voice, not the brand's. Customers noticed. Perla, trained on the brand's existing customer chats, was the first responder that sounded like the brand.

The handoff rule mattered more than the auto-reply. The decision rule "anything with the word 'refund' or 'broken' or 'angry' goes to a human" was where 90% of the trust came from. Customers don't mind an AI first responder if they know a human is one ping away.

The daily digest changed how the founder spent mornings. Previously, day one was 30 minutes of scrolling through overnight chats. Now it's a 5-bullet read with the open items flagged.

What it cost in setup time

• Day 1 (1 hour): Connect WhatsApp Business number. Paste in brand voice guide. Paste in FAQ. Set escalation rules.
• Week 1 (~30 min/day): Review Perla's replies before they send. Correct the ones that drift. Update the knowledge base from real conversations.
• After week 2: Switch to auto-send for known intents, review-only for the rest.

The investment isn't "deploy and walk away." It's "spend a focused week teaching her your business, then maintain in 30 min/week."

When this works — and when it doesn't

This pattern works when:

• The work is high-volume, mostly-repetitive, with a clear handoff path.
• The brand has a defined voice that can be written down.
• There's a human still on the team for the hard cases.

It does not work when:

• The "customer support" is actually senior-level account management.
• Every conversation is high-stakes or legally sensitive.
• The team doesn't have time to do the week-1 teaching investment.

The full picture

This is one of several patterns we see early teams using Perla for. The companion reads:

• What is an AI Employee? — the plain-English definition and the four work categories.
• AI Employee vs. AI Assistant — why the structural difference changes what you buy.

If this is the shape of your problem, hire Perla as your first AI employee.