Senior Product Operations Manager – LLM Automation

Responsibilities:

- Own the contact-centre deflection roadmap. Translate operational pain points (volume peaks, repeat queries, cost drivers) into a prioritised intent-automation roadmap, measuring success by deflection rate, cost per interaction, and customer satisfaction — not feature counts.

- Design and own the intent taxonomy and evaluation framework. Build the classification hierarchy underpinning every agent, establish pass-rate-by-intent metrics and confusion matrices, and instrument the hard distinction between containment (agent solves it) and deflection (client doesn't contact us) across live and shadow traffic.

- Drive LLM agent deployment and iteration cycles. Partner with Engineering to ship agents to production, own go/no-go decisions using multi-metric eval gates (accuracy thresholds, containment rates, customer sentiment, compliance risk), and write PRDs that survive legal and engineering scrutiny simultaneously.

- Navigate multilingual evaluation complexity. Oversee agent performance across 15+ languages and five regulatory jurisdictions; distinguish genuine linguistic gaps from regulatory differences or domain drift, and prioritise fixes accordingly.

- Drive cross-functional consensus under conflicting incentives. Hold the conversation where Customer Success demands safety-first, Engineering wants velocity, Compliance requires audit trails, and Exec needs the headline number. Document decisions in ADRs that explain the trade-off, not just the conclusion — own closure, not consensus theatre.

- Build and maintain regulatory evidence packs. Produce documentation that satisfies FCA Consumer Duty obligations, EU AI Act high-risk categorisation assessments, and internal governance — showing that your agents classify intent accurately, don't discriminate, and degrade gracefully under uncertainty.

- Author high-bar product artefacts. Produce PRDs, ADRs, exec memos, and board briefing materials that survive scrutiny from product, legal, finance, and regulators. Your written output is evidence.

- Establish and challenge your own reasoning. Distinguish first-principles analysis from market heuristics, run pre-mortems, and quote uncertainty explicitly. Act when you can test; defer when you can't.

Requirements:

- 5+ years in product roles owning customer-operations or contact-centre automation — accountable for deflection outcomes, measuring cost per interaction, and iterating on actual operational metrics, not just "worked on a support project".

- Proven track record of shipping and measuring cost/deflection outcomes, speaking in outcomes (cost reduction, containment, customer friction) rather than outputs (calls handled).

- Intent taxonomy and eval design fluency — reading confusion matrices, distinguishing accuracy from containment, and designing holdout test sets and eval gates ahead of production deployment.

- Comfort in regulated environments (FCA, CySEC, ASIC, SCB, CMA) — able to discuss Consumer Duty, vulnerable-customer treatment, AI governance, and data governance with compliance teams without a translator.

- Demonstrated ability to ship LLM/AI agents to production, including awareness of the gap between lab and live performance, latency, token costs, and hallucination surfaces.

- Cross-functional decision closure under conflicting incentives — you produce ADRs that explain a trade-off and a decision, not consensus documents, and you're comfortable saying no.

- High written-output bar — PRDs, ADRs, and exec memos that separate fact from assumption, quote confidence levels, and survive scrutiny from lawyers, CFOs, and product leaders.

- Calibrated reasoning and first-principles thinking — you distinguish market heuristics from a reasoned case for action and quote your confidence explicitly.