How Engine and NeuraFlash delivered a human-centric voice agent to scale service

Automating chat is easy. You don’t have to worry about tone, latency, or the quality of the voice. If I type something into a chatbot and see the little dots spinning, I know it’s being processed — I’ll wait for the response. But voice is completely different. If I say something on the phone and hear silence, I’m immediately like, “Hello? Are you still there?” There’s just no patience for that.

That’s where voice introduces a whole new level of complexity. Tone matters. Timing matters. The quality of the voice matters. The more your agent engages like a human, the more value it creates — because people will actually stay on the call. And that’s our goal at NeuraFlash: Help our clients create voice agents that keep customers engaged, while still being transparent that they’re speaking to AI.

If the experience sounds robotic, it doesn’t matter how strong the technology is. On average, within about 2.5 words, the customer is asking to be transferred to a human. That’s how quickly you can lose them.

At the same time, customers keep calling because they believe their problem will get solved — even if they expect some frustration going in. What Agentforce Voice does is change that expectation. It understands conversational nuance, responds naturally, asks intelligent follow-up questions, and takes action on the back end. It’s actually solving problems.

I knew from the very first call with Engine that it was going to be a great fit. As a travel platform handling high volumes of customer interactions, voice is critical to delivering timely support. They were looking for a conversational voice agent that could answer traveler questions, manage bookings, and handle hotel cancellations end to end.

They walked us through their tech stack — which is wall-to-wall Salesforce. From a data and integration perspective, that puts them in a really strong position right away. Engine had the data to support a ton of different customer support use cases. That’s really where Agentforce Voice stands apart — it’s natively integrated with Salesforce CRM, Data 360, and customer records. It connects directly to your knowledge and your data, which is what makes it powerful.

They were already doing impactful things with Agentforce on chat, then came to us because of our experience saying, “Now we want to be first with voice.” Engine is willing to push the envelope, but more critically, they’re willing to put in the work to make it happen.

We started right before Agentforce Voice was generally available, and Engine still wanted to be live within three months. It became this pre-game huddle between NeuraFlash, Engine, and Salesforce to get the product where it needed to be. The voice didn’t sound great at launch — we all knew that — but everyone was committed to making it excellent.

Our teams started by mapping out how a customer conversation should flow — from the first greeting all the way to resolution. Then we narrowed in on specific areas Engine wanted to automate: General FAQs, booking inquiries, and hotel cancellations.

These weren’t the most technically impressive use cases, but they were the most frequent, and stood to really move the needle in terms of reducing inquiry volume. We also focused on support use cases that were highly feasible because they built on Engine’s existing AI agent Eva – the Engine Virtual Assistant – which customers were already turning to for 24/7 personalized agentic support across web and chat.

Because Engine was first to market with voice, there was a lot of customization in the build, and testing was truly all hands on deck — NeuraFlash, Engine, and Salesforce were all involved.

We literally called Eva Voice over and over again, interacting with it like real customers. Once we felt confident, we rolled it out in phases. We started by routing about 10% of calls to Agentforce Voice, watched closely, and made adjustments. For example, refining the script, and changing the voice from masculine to feminine.

Then we increased usage, directing 50% of calls to Agentforce Voice. Now we’re at 75% of incoming calls. It was very intentional — test, learn, iterate, and scale.

The Engine team is all-in. Their service reps are saying Agentforce Voice is nothing short of miraculous. When a case gets transferred, reps see a fully filled out voice transcript and summary in Agentforce for Service. Their job has simplified dramatically.

That means they don’t have to ask all those introductory questions anymore. They can skip the fact-finding and jump straight into helping the customer. So now, 100% of their time is focused on making the customer happy.

From the customer side, the results have been strong. Travelers rated their conversations with Eva Voice highly from the first month of deployment — especially notable for voice, which has traditionally had lower satisfaction scores.

First, this is not a “wait and see” technology. Companies using Agentforce Voice are saving money and improving customer satisfaction right away. Forget the old idea that voice automation frustrates people. That was the world of circular IVRs. This is different.

Second, don’t be scared of your data. You already have it, and there’s a way to use it to start solving problems. With Data 360, you can move quickly because the integration work is already there.

Third, you don’t just launch agentic AI — you onboard it. It’s not going to behave the same way every time. You need guardrails, and then you monitor and coach it, just like an employee.

And finally, always have a clear escalation path. If someone wants a human, they need to be able to get one — and that person needs full context so the conversation continues naturally.

The goal is a truly end-to-end agentic experience — and Engine is close. I want to be able to call Engine customer service and have Agentforce say, “Glad we found that confirmation number. Before we hang up, I see your company has an event in Austin in two weeks — is that what you’re calling about?”

Then I can say yes, and Eva Voice handles everything — adding rooms, removing rooms — done. That’s where this is going. It’s not just about answering questions anymore, it’s about solving problems before you even have them.