60% of every customer message, sent by AI.

Support at 100,000-customer scale, on a brand built on medical credibility.

A premium supplements brand at 100,000+ customer scale has a specific support shape. High inbound volume, time-zone spread, questions that run from delivery tracking to ingredient and dosage guidance.

• High inbound volume across a customer base of more than 100,000, with a time-zone spread that needs answers around the clock.
• Health-adjacent questions on ingredients and dosage, where a wrong or off-brand answer carries reputational cost.
• Draft-only AI was not enough. AI that suggests replies but will not send them still leaves the customer waiting.
• Send-anything AI was a risk. AI that sends replies but cannot be trusted on health-adjacent topics puts the brand on the line.

They needed AI that sends, and can be trusted to.

The team needed AI that could answer fast, answer around the clock, and not embarrass a brand built on medical credibility. Two things ruled out the standard alternatives: AI that drafts replies but won’t send them, so the customer still waits, and AI that sends replies but can’t be trusted on health-adjacent topics, so the brand pays the reputational cost.

Richpanel’s autonomous-mode-with-SOPs posture committed to both at the same time. The SOPs that govern refunds, replacements, and brand voice are encoded up front, and a 50% resolution guarantee was written into the contract if it didn’t deliver.

Full email autonomy, governed by the brand’s own SOPs.

• Autonomous email AI. autoMode on globally. The AI sends and closes email without waiting for human review, on the only active channel, for every customer.
• Encoded SOPs. Refunds, replacements, and brand voice are governed by the team’s own rules, sampled against historical tickets before launch to confirm the AI’s drafts matched what the human team would have written.
• A human layer for the hard 40%. Medical-context cases, subscription edge cases, and named wholesale accounts stay with the human team by design.

AI sends the majority of messages, at higher CSAT than humans.

• 60% of all output sent by AI. 4,881 of 8,074 messages over 42 days, across 6,376 conversations.
• AI CSAT 4.43 / 5, higher than the human team’s 4.25, with a team-level CSAT of 4.32.
• 28.1-second median first response from AI, pulling the team-level median to 40.5 seconds.
• ~3 FTE-equivalent of capacity added at ~45 messages/agent/day, without hiring.
• 100% autonomous from day one. No co-pilot phase, no rate-limited ramp.

18 days to set up. Day 19 to full autonomy.

Side by side.

Per-layer volume and quality on the messages each side sent over the same 42 days. Team-level CSAT, weighting both layers, is 4.32 / 5, what customers actually experience.

Both layers above 4.20. The SOPs translated cleanly to AI behavior. Each side’s metrics reflect only that side’s messages, not a head-to-head on identical conversations.

The 40% the team still owns.

Of the 8,074 messages Aeons handled in 42 days, the AI sent 4,881. The other 3,193 went to the human team by design. The 60/40 split isn’t a target the team set. It’s what the AI will autonomously resolve given the current SOPs and the channel mix.

The boundary between the two layers is deliberate. Categories where the right answer depends on customer-specific medical context, subscription edge cases that need negotiation, and wholesale inquiries with a named account owner all stay on the human side. As the team trusts more categories to autoMode, the share moves.

100% autonomous from day one.

autoMode is on globally. AI sends and closes email without waiting for human review. Over 42 days that’s 4,833 emails sent and closed autonomously on the only active channel. The 28-second first response and 60% volume share come directly from that decision.