“The AI drafts replies that sound like our best agent on her best day. Humans approve, edit, or take over — the handoff is invisible to customers. CSAT is up 31% this quarter.”
The challenge
Northwind moves freight for SMBs across West Africa. Their support inbox is a tense place: customers want to know where their shipment is, when it'll arrive, and what to do when something goes wrong. Most messages have a clock running.
Before Tickki, agents copy-pasted shipment statuses from an internal dashboard into customer replies, manually. The good agents were great. The new agents made costly mistakes.
What they did with Tickki
Northwind connected the inbox to their internal ops system through Tickki's webhooks. Now when a customer asks "where's my shipment?", the AI fetches the live status and drafts a reply with the actual ETA, the actual carrier, and a polite line about next steps.
Critically, they didn't switch to autopilot. Lina insists on human approval for every customer-facing message. The AI drafts; an agent clicks approve or edits. After three months of supervised drafting, the edit rate dropped to 9% — meaning 91% of AI drafts go out untouched.
The handoff design — when sentiment dips, when the keyword list trips, when AI confidence drops — was tuned heavily during onboarding. Lina audits the weekly handoff log to refine triggers.
The result
- CSAT up 31% quarter-over-quarter.
- Median resolution time down 44% — most of the gain comes from drafting + ops-system integration, not from cutting human review.
- Edit rate is 9% — meaning the AI is reliably matching the team's voice and content accuracy.
Why it stuck
"The thing I tell other operators," Lina says, "is don't switch to autopilot. The 9% you edit is the product. That's where the AI gets better."