Enterprise

Enterprise Virtual Assistant

Google ADKCX Agent StudioVertex AIDialogflowCloud Run

chatbot sessions a day: 300K+; chatbot sessions a day
containment: 78%; containment
field-tech chats a day: 10K+; field-tech chats a day
field-tech assistant, 0 → launch: 4 mo; field-tech assistant, 0 → launch

Overview

A Fortune 50 telecom company's omnichannel virtual assistant, used by customers in its app, on its website and over messaging, and by its field technicians. It handles more than 300,000 sessions a day. The team is moving it from a legacy chatbot stack to Google's ADK / Conversational Agents platform and adding generative AI for the first time. Client details are confidential.

Problem

The legacy chatbot couldn't support generative experiences. The customer and field-tech assistants were built separately, with no shared playbook layer. Five teams across three orgs had to migrate together without disrupting a channel that handles hundreds of thousands of conversations a day.

My Role

Product Manager. I own roadmap items and feature scoping and coordinate across 5 teams in 3 orgs. I turn leadership's priorities into product increments we can ship and work through trade-offs with engineering, design and operations.

What it looks like

01 · Generative answer

Virtual Assistant

Online

My internet keeps dropping in the evenings. Third night in a row.

Sorry about that. I checked your connection, and here's what I found:

Connection check · last 7 days

Gateway statusOnline
Outages in your areaNone
Reconnects14 · mostly 7–10pm

An evening-only pattern with no outage usually means Wi-Fi congestion, not a line problem. Moving your network to a less crowded channel often fixes it. Want me to try that?

Optimize my Wi-Fi Restart gateway Talk to an agent

Type a message…

02 · Handoff with context

Virtual Assistant

Connecting you…

I was charged twice for my plan this month.

I can see two charges of $64.99 on Oct 3. Billing disputes need a specialist, so I'm connecting you to one now.

Shared with the agent

Customer reports a duplicate monthly charge. Two identical charges on Oct 3 confirmed. Identity already verified. No earlier contact about this.

Billing Verified Sentiment: frustrated

Alex from Billing joined the chat

Alex · Billing

Hi! I've got the details, so no need to repeat anything. I'm reversing the duplicate charge now.

Type a message…

03 · Field-tech assistant

Tech Assist

Work order #48213

Downstream SNR is 29 dB at the tap. Customer still has intermittent drops. What next?

29 dB is below the 33 dB minimum, so the problem is likely between the tap and the home. Check in this order:

Inspect the drop connectors for corrosion
Replace any splitters with more than 2 ways
Re-test SNR at the ground block

Source: Install & Repair Guide §4.2

Replaced a 4-way splitter. SNR is 36 dB now.

That's within spec. I've added the fix to the work order notes.

Close work order Run speed test

Type a message…

One assistant, three jobs A generative answer based on live account checks, a handoff to a person that carries a summary so customers don't repeat themselves, and a field-tech assistant that answers from repair guides. Illustrative · sample data

Approach & Key Decisions

▹ Scoped and led a 4-month program to take the field-tech chatbot from nothing to launch. It now serves 100+ technicians and handles 10,000+ chats a day.
▹ Prototyped the generative Conversational Agents Playbook layer and pushed for it to go to production, where it reduced escalations to human agents.
▹ Led the Sales and affordable-plan enrollment modules of the omnichannel IVR migration to Dialogflow / ADK.
▹ Built a Gemini-powered chat transcript analysis tool so the team can see how the assistant performs in production conversations (see the separate case study).
▹ Kept one migration backlog across all teams so cutovers stayed in order and risks stayed visible.

Outcomes

★ 300K+ chatbot sessions a day at 78% containment, sustained during the migration.
★ Field-tech assistant launched in 4 months: 100+ technicians, 10,000+ chats a day.
★ Playbook prototype taken to production, with lower escalation to human agents.
★ Omnichannel IVR migration in progress, with the first generative AI features live in production.

More work