About Ultrahuman

Ultrahuman is the world's most comprehensive self-quantification platform. Our products include the Ultrahuman Ring AIR, the world's lightest sleep-tracking wearable; Ring PRO, our third-generation smart ring with category-defining 15-day battery life; the M1 continuous glucose monitoring platform; Blood Vision, a preventive blood-testing platform with pioneering UltraTrace- technology; Ultrahuman Home; and Jade, the world's first real-time biointelligence AI. We design our own hardware, firmware, and software in Bangalore and ship to customers in 150+ countries.

We operate like a product company in every function - including support. CX here is not a cost center to be outsourced and forgotten. It is a product surface: instrumented, versioned, deployed, and improved weekly like any other part of the stack.

Why this role exists

Most "Head of CX" roles are headcount-management jobs wearing an AI costume. This one is the inverse.

Our support org already runs on an in-house AI stack built on Claude (via AWS Bedrock). It handles hundreds of thousands of conversations a month, with the overwhelming majority resolved end-to-end by AI - no human touch. A voice-AI pilot follows later this year, with social-channel automation behind it. This is not a roadmap slide - it is the system in production right now, processing real replacements, real refunds, real medical-adjacent sensitivity, every hour.

The automation problem is largely solved - the machine already does that. The next frontier is experience quality. The mandate is to make every interaction genuinely good: sharper judgment on edge cases, faster recovery when the AI gets it wrong, and a CSAT line that visibly bends upward on your watch.

The hard problems left are operator problems:

- Where does the AI's judgment end and a human's begin?

- Which small fraction of conversations genuinely need a person - and how do you staff for exactly those?

- How do you raise quality without giving back the automation gains?

- How do you make a lean org outperform one ten times its size - and prove it with data?

You will inherit a deliberately lean senior organisation and a mandate to keep it that way. If your instinct when volume grows is to open a req, this is the wrong role. If your instinct is to open the prompt, the flow, or the policy - keep reading.

What you will own

- End-to-end CX quality and outcomes across chat, email, voice, and social - every customer conversation, AI-handled or human-handled, is yours

- The AI-ops engineering pod - you are the product owner of the support AI: prompts, flows, tooling, and evaluation changes shipped on a continuous cadence

- Channel AI Managers - player-coaches who operate the AI on their channel daily and personally handle customers when the system escalates; you hire, develop, and hold the bar

- The exception floor - the small human team for conversations that genuinely require a person; you define what qualifies as an exception, and keep that definition honest

- The quality system - automated QC, human-graded sampling, escalation forensics, and the metric system everyone (including the AI) is measured against

- Policy as code - replacement, refund, and warranty decisioning executed by the AI inside guardrails you author, with an audit trail proving it is applied correctly

- The automation frontier - deciding, channel by channel, when automation expands, backed by quality evidence rather than optimism

- The executive read - a weekly, numbers-first view of support health that leadership uses to make product and hardware decisions

Your first 90 days

Days 1-30 - Learn the machine. Sit on the exception floor. Read transcripts daily - hundreds of them, AI-resolved and escalated. Shadow each Channel AI Manager. Trace ten replacement decisions end-to-end through the flow, the policy gates, and the audit log. Ship at least one small prompt or flow improvement yourself, through the real pipeline. Deliver a written assessment: where the residual human-touch volume comes from, which slices are automation gaps versus genuine human work - and your read on the biggest CSAT levers.

Days 31-60 - Take the controls. Own the weekly quality review and the dashboard. Set quarterly targets per channel with your Channel AI Managers. Pressure-test the escalation taxonomy: kill categories that exist out of habit, tighten the ones that protect customers. Make your first structural call - a policy change, a flow rewrite, or a staffing reallocation - and instrument it so the result is measurable within two weeks.

Days 61-90 - Set the trajectory. Publish the 12-month operating plan: the CSAT improvement plan with owners and dates, voice-AI pilot success criteria and GA path, social automation sequencing, and the headcount plan that holds the org lean while volume grows. Present it to founding-team leadership with the quality evidence to back every expansion decision.

What success looks like at 12 months

CSAT - Meaningfully higher - the headline metric. Automation gains that cost quality don't count

Containment - Higher than today, with false-resolution rate measurably down - the AI resolves more AND resolves it right

Email - Automation running at quality parity or better versus the human-review baseline

Voice - Live in production past pilot, with quality parity

Org - Still lean despite volume growth - productivity per person visibly compounding

Decisions - Replacement/refund accuracy improved in both directions, with a clean audit trail

Team - Channel AI Managers you've hired or grown who could each run a channel autonomously

Standing - Leadership treats your weekly CX read as a primary product-quality signal

Must-haves

- 5+ years running customer-facing operations at scale - support, trust & safety, marketplace ops, fintech ops - with direct ownership of quality outcomes, not just dashboards

- Hands-on fluency with LLM systems. You have personally written and iterated prompts, designed conversation flows or agent behaviors, and debugged why a model did the wrong thing. "I manage people who do that" is disqualifying

- Demonstrated lean-org conviction - at least one concrete example of solving a scale problem with systems instead of headcount, with numbers

- Strong written judgment - policy, prompts, escalation criteria, and exec updates are all writing

- Comfort owning irreversible decisions - replacements, refunds, and sensitive interactions executed by an AI under your policy

- Data fluency - you can interrogate a funnel, spot a regression, and call out a vanity metric; SQL comfort helps

- Bangalore, in-office

- Willing to complete a live working exercise - real (anonymized) escalated transcripts and a live decision flow from our production stack; you propose concrete prompt, flow, and policy changes. It's the job, compressed into 90 minutes

Strongly preferred

- You've shipped something with LLM APIs yourself - a tool, an eval harness, an automation. A public GitHub or open-source contributions are a genuine plus and will be looked at

- Hardware/device support experience: RMA flows, diagnostics, warranty logic

- AI evaluation experience: golden sets, LLM-as-judge pipelines, regression gating

- Multi-geography support exposure, especially health-adjacent

- Player-coach instincts - you've taken a hard customer conversation yourself, recently

Explicitly NOT required

- A specific degree, or any degree - we will not ask

- BPO or contact-center pedigree - legacy contact-center experience can even be a headwind if it comes with headcount-pyramid instincts

- A prior "Head of" title - if you've been the de-facto operator of an AI-heavy support system and you're ready for the full mandate, apply; the live exercise outweighs title history

- A wearables background - we'll teach you the product; we can't teach judgment