AI that customers actually own: Lessons from five customer experience rollouts

For any organization building with AI, there is a familiar refrain they hear from vendors: "It's your data, your customer relationships, and your AI."

In practice, however, those vendors hand you a pre-built black box with a thin configuration layer on top, which means you can't inspect the reasoning, adapt the architecture, or take that agent with you. Our philosophy at Google Cloud is different: We pair your teams with our forward-deployed teams so your developers build, customize, and own the code end to end.

Over the past year, I've spent my days in stand ups and deployment sprints with customer experience leaders across banking, telecom, media, and retail. We regularly discuss just how important it is for them not only to have effective AI but also understandable AI. If they want to finally move beyond experiments and deploy AI at scale, the ability to update and modify it on the fly to truly own their own AI is essential.

In addition to working with the right engineering experts, we’ve found a mindset shift is a key ingredient to successfully embracing agentic work. Once customers are comfortable with this new way of working, we're seeing build timelines shrink from months to days and a double-digit month-over-month increase in production rollouts.

With that in mind, here are four lessons we've discovered on the ground from working with five customers we've been building the next generation of enterprise AI with. We’re confident they can help you accelerate adoption and action, too.

1. Stop building separate agents for every channel

Historically, enterprises built their chat agent and their voice agent on two completely separate software stacks, creating disconnected logic, duplicate maintenance, and jarring handoffs across both human and AI-powered agents.

In most organizations, one team owns digital chat while an entirely different team manages the interactive voice response system. Each channel has its own vendor, prompt trees, backend integrations, and release cycles. This means every time a policy or product changes, engineers must rebuild and test workflows twice. Worse, when a customer switches from an app to a phone call, the context vanishes, forcing them to repeat their story from scratch.

With Gemini for customer experience, you build your core business logic, enterprise guardrails, and backend integrations exactly once. From there, you can deploy that single agent across voice, web, mobile, and messaging — all while maintaining a shared session context. Here is what that unified architecture looks like in practice:

• Unifying voice and chat in 30 days: A major media and telecom provider was managing separate voice and digital roadmaps that couldn't share context. This forced their engineering teams to maintain duplicate integrations for billing, troubleshooting, and account management. In partnership with Google’s forward-deployed engineering team, they consolidated prompts, backend tools, and policy guardrails to bring both channels into a single Gemini experience in just 30 days.
  
  This means a subscriber troubleshooting an issue in chat can transition to voice without losing a single step. Unifying the architecture in this manner changed how leadership measured success: It moved beyond basic "containment" (which often just measures how well a bot deflects frustrated callers) to tracking true end-to-end resolution.
• Connecting the retail front door to the back office: A national retailer is taking this single-architecture approach even further across the entire customer and employee journey. Rather than stitching together separate point solutions for search, checkout, post-purchase support, and store operations, they are unifying product discovery, guided shopping, and customer service in Gemini.

2. Win on real-time reasoning first

In a live voice conversation, latency and natural expression aren't abstract benchmarks; they are the entire experience. With Gemini for customer experience, we deliver remarkably natural-sounding, low-latency voice options that can be customized to your brand. The agent can even handle when callers interrupt or pivot mid-sentence (just like in a person-to-person conversation) thanks to adaptive pacing and seamless barge-in handling, which detects speaker interruptions and natural pauses.

A natural-sounding voice is only half the equation. An agent's ability to reason through ambiguity and drive business outcomes is just as critical. Because Gemini's multimodal reasoning connects directly to enterprise skills, tools, and context, the agent doesn't just recite scripts. It reasons across live backend systems, interprets complex policies, and executes multi-step workflows in real time. Here is how that reasoning engine translates into measurable business outcomes:

• Elevating the baseline for voice agent capabilities: At a major wireless carrier, the team had already evaluated several AI vendors based on impressive voice demos. When they tested Gemini on live customer calls, we used only baseline voice configuration, leaving our advanced prosody and custom voice branding off. Despite this, Gemini delivered higher containment and resolution rates than the alternatives purely because of its superior ability to reason and drive business outcomes.
  
  Live telecom calls rarely follow a script. A caller might ask about a device upgrade, a promo credit, and a roaming charge — all in the same breath. The alternative vendor bots sounded pleasant during simple greetings but stalled the moment conversations got complicated. But Gemini reasoned across the carrier's billing logic, account state, and backend APIs in real time to resolve these issues on the first call.
  
  Once the carrier layered our custom voice and natural prosody controls on top of that reasoning engine, they didn't have to choose between an agent that sounds natural and one that drives business outcomes. They got both.

3. Earn trust on legacy systems

Enterprise leaders don't want a disruptive rip-and-replace pitch when their existing contact center handles millions of calls a day across deeply integrated telephony, CRM, and compliance systems. Traditional AI pitches often demand a risky cutover, asking leaders to gamble their baseline customer experience on an unproven deployment.

Our approach in Gemini for customer experience is pragmatic: We meet enterprises where they are. We optimize their existing deterministic flows first, then layer in generative Gemini capabilities in the Gemini platform. This blends strict compliance guardrails with real-time reasoning so live operations are never put at risk. Here is how that phased approach delivers immediate value without disrupting live operations:

• Driving a 5% containment gain on existing infrastructure: A large U.S. retail bank was already achieving 76% containment across high-volume banking inquiries on our previous-generation Dialogflow platform. They were understandably cautious about migrating mission-critical call flows to generative AI.
  
  Rather than pressuring them into an immediate cutover, our forward-deployed engineering and product teams went into the trenches on their existing stack. By tuning intent routing, tightening webhooks, and resolving drop-off points, we helped them extract five additional points of containment on their legacy setup.
  
  Proving we cared about driving business outcomes on their existing system earned deep operational trust and paved the way for a seamless, phased transition to Gemini for customer experience.

4. Own your AI

When you own your AI, your team gets direct, transparent access to the entire agentic stack. This includes the reasoning loop, prompt architecture, Agent Registry, Model Garden in the Gemini platform, and modular skills, tools, and context that connect to your enterprise systems.

Many vendors lock orchestration and tool definitions behind a proprietary black box, charging a premium every time you need to change a workflow. With the Gemini platform, you aren't trapped inside a vendor's configuration screen. Your teams can inspect the agent’s reasoning, run automated evaluation, version-control its code, and evolve every piece of the architecture as your business grows.

Our approach is closer to how a coach trains a world-class athlete. Our forward-deployed engineers and product managers run drills alongside your team. We help you identify gaps, advise your developers through the build, and transfer our knowledge directly to your staff. By the time we're done, your engineers are the frontrunners for your business. There's no question about who owns the AI: You own the source code, the full design, and the underlying architecture. Here is how this hands-on partnership accelerates time-to-market for even the most heavily regulated industries:

• From model handoff to production in under two months: In August, we gave a global financial institution direct access to our latest model via Agent Runtime on the Gemini platform. Since it was provisioned in a single Google-hosted runtime and isolated cloud sandbox, it met their strict security and compliance requirements right out of the box.
  
  Instead of handing them a finished black box or building the solution behind closed doors, our forward-deployed engineers embedded directly within their engineering org to run architecture reviews, pair-programming sessions, and live call-flow drills.
  
  Within weeks, 25 of their internal developers earned certification on the Gemini platform, mastering how to build modular skills, tools, and context, as well as manage their own evaluation and deployment pipelines. By late September, they launched Gemini for voice into live production — built, owned, and operated by their own engineers.

Moving from AI experiments to enterprise-scale impact doesn't mean handing over the keys to a black box or upending the legacy systems your business relies on today.

By unifying your voice and digital channels, prioritizing real-time reasoning, layering generative capabilities onto the systems you already trust, and empowering your own engineers to own the stack end-to-end, you can build customer experiences that adapt as fast as your business does.

To learn more about how you can put Gemini to work for your business, explore Gemini for customer experience.