How I sequenced 10+ GenAI features for a $10M contact centre transformation — and why I championed the counter-intuitive feature first.
HCSC needed to modernise a contact centre handling hundreds of thousands of customer interactions. TrueServe™ — Deloitte's Salesforce-based AI platform — was the vehicle, but GenAI feature sequencing was wide open.
Ran discovery across agent workflows, mapping 180+ tasks by frequency, cognitive load, and error rate. Identified that email drafting — not real-time assist — had the highest volume, lowest risk, and fastest feedback loop.
Championed smart email drafting as the first GenAI feature — not the more visible real-time agent assist. The logic: async drafting let us gather clean acceptance data without real-time risk. It became the proof point that unlocked the rest of the program.
78% AI draft acceptance by sprint 4. Program delivered 20% ahead of schedule. Key learning: the right first feature isn't the flashiest — it's the one that generates the evidence you need to de-risk everything that follows.
TrueServe™ is Deloitte Digital's AI-powered contact centre accelerator — built on Salesforce, designed to reduce handle time, prevent unnecessary contacts, and empower agents through GenAI tooling.
Health Care Service Corporation (HCSC) — one of the largest customer-owned health insurers in the US. High-volume, high-stakes contact centre with strict compliance requirements and low tolerance for AI errors.
180+ distinct agent tasks. 10+ GenAI features on the roadmap. A $10M budget and executive pressure to show early ROI. Which feature do you ship first, and how do you justify it to stakeholders who want everything at once?
Product Consultant leading GenAI feature definition, stakeholder alignment, and delivery sequencing. Worked directly with engineering, UX, and HCSC's ops leadership to prioritise and ship features end-to-end.
Everyone wanted real-time agent assist first — it's the flashier capability. But I pushed for smart email drafting as the launch feature. Here's why: async email gave us a clean signal. Agents could accept, edit, or reject AI drafts without the pressure of a live customer on the line. That meant real acceptance data, not politeness. By sprint 4, we had a 78% acceptance rate — and that number was what unlocked stakeholder confidence to go bigger.
LLM-generated draft responses for agent email workflows. Agents review, edit, and send — with a feedback loop that improved model accuracy over sprints. First feature shipped; 78% acceptance rate by sprint 4.
Automatic post-call and post-chat summaries reducing agent after-call work time. Structured output tied into CRM, eliminating manual note-taking for 180+ task categories.
Live suggestions surfaced during customer interactions — recommended responses, policy lookups, and escalation triggers. Sequenced after email drafting to build on established model trust and agent familiarity.
Virtual agent flows for high-frequency, low-complexity queries — reducing live agent volume and allowing human agents to focus on complex, high-empathy interactions.
TrueServe™ is Deloitte Digital's pre-built accelerator for contact centre transformation — combining conversational AI, omnichannel orchestration, agent assist, and advanced analytics to deliver faster ROI than custom builds.
View TrueServe™ on Deloitte Digital ↗The most important PM skill I exercised here wasn't technical — it was sequencing under uncertainty. When you have 10 GenAI features and can only ship one first, the question isn't "what's most impressive?" It's "what generates the evidence that makes everything else possible?"
Championing email drafting over real-time assist was a hard sell to stakeholders who wanted the flagship feature. But the acceptance data from those early sprints became the single strongest proof point in the program — and it de-risked every subsequent feature decision.
Also learned: in regulated industries like health insurance, trust is the product. You don't build trust by shipping the biggest feature first. You build it by being right about the small ones.