// Case studies

Field reports.

We can't always name our clients, but we can show our work. Three anonymized engagements — what was broken, what we built, what changed.

AI-native fintech·Series B · ~120 FTE

From queue chaos to a model-aware support pod in six weeks.

An AI-native fintech was deflecting ~60% of inbound chat with an in-house LLM agent, but the 40% that broke through landed in a single overloaded inbox. First response drifted past 12 hours and CSAT was sliding.

// What we built

  • Stood up a 24×5 chat + email pod, looped directly into their LLM logs and feedback queue.
  • Designed a confidence-based escalation taxonomy and rewrote the top 20 macros against it.
  • Weekly RLHF review: every low-confidence AI turn graded by the pod, exported back to the model team.
"They didn't just take tickets off our plate — they became a quality signal for the model. Our containment went up because our humans got smarter."
Head of CX · AI-native fintech

// Outcomes

<0s

First response time

+0

CSAT points

0%

Resolution within SLA

Global hospitality tech·Public · 8,000+ FTE

A 24×7 multilingual ops layer behind an automated property platform.

A global hospitality platform needed a follow-the-sun back-office and tier-1 support layer that could shadow an in-house automation roadmap without re-explaining the product every quarter.

// What we built

  • Built a senior, English + multilingual pod operating across PH and EU hours.
  • Embedded into the partner's helpdesk, billing and content systems with role-scoped SSO.
  • Weekly QBR against an integrated AI-assist / human-resolve funnel — single P&L, single dashboard.
"We treat them as an extension of the product team, not a vendor. Their seniors flag bugs before our PMs see them."
VP Operations · Global hospitality platform

// Outcomes

0/7

Coverage, single contract

0×

Backlog cleared in Q1

0%

QA score, rolling 90d

Consumer web platform·Public · 5,000+ FTE

A specialist desk for the long tail of edge cases their AI couldn't close.

A consumer web platform with a mature AI-assist support stack needed a small, senior desk to own the long tail of edge cases — the 8% of conversations that drove a disproportionate share of escalations and refunds.

// What we built

  • Stood up a small, hand-picked pod with deep product training and refund authority.
  • Built a custom triage flow: AI handles intent + context, pod handles judgment + outcome.
  • Monthly review of the long-tail taxonomy — feeding both new macros and product feedback.
"They became the team we send our hardest tickets to. That alone changed the economics of our support org."
Director of Support · Consumer web platform

// Outcomes

-0%

Repeat-contact rate

+0

NPS on edge-case flow

0 days

Kickoff to live

All client identities have been redacted. Metrics are directional and represent steady-state performance within the first two quarters of engagement.

// Your turn

Bring us the case. We'll bring the team.