Voicify cuts costs with Google Cloud's Gemini models

Voicify has adopted Google Cloud's Gemini models and Vertex AI for its telephone AI platform, cutting costs and improving reliability for restaurant and healthcare customers.

Founded in 2018, Voicify started with voice assistants across multiple channels. During the pandemic, demand pushed it toward phone-based services in sectors facing heavy call volumes and staff shortages. Restaurants could miss up to 20% of calls and lose orders, while healthcare providers struggled to manage appointment traffic with the accuracy required by practice management systems, according to the company.

That shift brought new requirements. Voicify needed its assistant to handle complex customer requests across point-of-sale and practice management systems with complete accuracy, while keeping response times short enough to prevent callers from hanging up.

Traffic spikes created another operational challenge. Restaurants and healthcare groups often see abrupt swings in inbound demand, forcing suppliers to balance cost with service continuity when large language model use rises suddenly.

Voicify initially used Google AI Studio before moving to Vertex AI and Gemini Enterprise as call volumes grew. It sought stronger assurances around scale, security and compliance, particularly for healthcare clients.

Under the new setup, Voicify uses its own orchestration layer to manage the phone AI stack and connect back-end services, including speech recognition, text-to-speech and text generation. The system also combines generative AI with programmatic processes rather than relying on open-ended text generation alone.

One of the main technical decisions involved restaurant menus. Instead of placing an entire menu in a single prompt, Voicify includes limited information at the start of a conversation and fetches more detail as an order develops.

This is intended to reduce delays and limit the complexity of larger orders. In a telephone ordering system, even small pauses can determine whether customers stay on the line.

According to Voicify, Gemini Flash improved latency and reduced user waiting times. The switch also cut costs by about 25% to 30% compared with its earlier use of other large language models.

Voicify also linked the move to faster onboarding for restaurant clients. Once access to a point-of-sale system is in place, preparing a restaurant for testing now takes one to two days, down from one to two weeks, it said.

Handling peaks

Seasonal surges are a key test for restaurant suppliers. Voicify used a mix of provisioned throughput and pay-as-you-go capacity on Vertex AI to handle record demand on the day before Thanksgiving without rate-limiting problems, according to the company.

Those measures have also helped it avoid dropped responses and service interruptions during periods of high use. Under the current configuration, Voicify said it has achieved 100% uptime.

Healthcare adds another layer of complexity. Voicify said it has built its platform to meet HIPAA, SOC 2, ISO 27001 and PCI compliance requirements, reflecting the need to handle sensitive patient and payment information under tighter controls.

The platform was designed from the outset with those standards in mind, the company said. It also supports a multicloud environment as part of its effort to maximise availability.

Sector focus

The focus on restaurants and healthcare reflects a broader commercial pattern in applied AI, where buyers often prefer narrowly defined tools tied to specific workflows rather than general-purpose assistants. For Voicify, that means validating restaurant orders against point-of-sale systems before submission and processing appointment-related requests in healthcare settings.

These tasks leave little room for error. An incorrect takeaway order or a mis-entered appointment can create immediate costs for operators and weaken confidence in automated systems.

Voicify said the updated architecture reduces reliance on its own in-house tools for gathering context and building menus. The increased stability of the Gemini-based service has also made it easier to bring clients onto the system, it added.

"Our conversational orchestration platform builds and validates restaurant orders against a point-of-sale system before submission to ensure accuracy," the company said.

On the economics of the move, Voicify said the higher cost of enterprise services compared with AI Studio was justified by uptime and easier scaling. "These enterprise-grade services may have carried an increased cost over AI Studio, but they were well worth it to ensure reliable uptime, and the premium pay-as-you-go feature made scaling much easier for us," it said.