Voice AI Won the Argument. The Harder Question Comes Next.

A year ago, the question in enterprise customer experience was whether voice AI was ready. That question has been answered. Funding into the category has surged, and brands are demonstrating measurable results from these solutions. Boards that once treated the phone channel as a cost to minimize are now asking their CX leaders why they aren’t further along. 

The market has made its decision. Voice AI is the buy. 

That consensus should feel like good news, and in one sense it is. But it also means the interesting question has quietly shifted, and most buyers haven’t caught up. It is no longer whether to invest in voice AI. It is which approach brands choose, and what they stand to gain or give up based on that choice over the next eighteen months. 

Two Ways to Buy Voice AI 

Enterprise buyers evaluating this category are really choosing between two different products wearing the same label. One gives you powerful general-purpose AI technology and a blank canvas, then asks you to build your system on top of it. The other shows up with your industry’s workflows, integrations, and edge cases already built in. 

Both can sound identical in a sales demo. The technology underneath is often genuinely comparable. What is not comparable is what happens after the contract is signed. 

With the first approach, the vendor and buyer learn the business together, often requiring significant implementation work to build industry-specific workflows, integrations, and edge cases. For some enterprises, that investment is worth the added flexibility. With the second, much of that learning has happened: common workflows, customer intents, and integrations have been built and refined for specific industries and use cases. 

The tradeoff is not as simple as one approach being universally better. McKinsey has argued that enterprises should buy standardized AI capabilities while reserving custom development for the domain-specific logic and proprietary workflows that actually create competitive advantage. The question for CX buyers is how much of that customization they want to take on themselves. 

What the Learning Curve Changes 

This distinction is not theoretical. It shows up directly in deployment outcomes. Research from CX analytics firm AmplifAI found that only 25% of call centers have successfully integrated AI automation into their daily operations.  

That gap points to a problem enterprise buyers often underestimate: buying capable AI technology and operationalizing it are two distinct tasks. Data quality, integrations, governance, change management, and industry-specific configurations can all slow deployment. Generic platforms that require substantial configuration add implementation time and professional services costs, while purpose-built systems can start further along because common workflows, integrations, and edge cases have been addressed. 

Real-world deployments of purpose-built systems across retail, healthcare, and transportation are already demonstrating automation rates of 70 to 90%, particularly when the system can handle a company’s highest-volume use cases end to end. General-purpose AI is not incapable of reaching the same outcome, but enterprises should look beyond the sophistication of the underlying model and ask how much domain knowledge is baked into the product on day one and how much will need to be built after the contract is signed. 

Domain Knowledge Isn’t a Configuration Setting 

Many of the underlying capabilities of voice AI transfer across industries: speech recognition, reasoning, latency management, conversational design, and escalation logic. But the operational context in which those capabilities have to work is very different.  

That matters because more capable AI does not automatically create better customer experiences. Simply layering AI onto the same operating philosophy risks automating deflection rather than eliminating it. PwC has similarly argued that successive waves of IVR, workflow automation, and chatbots have too often automated broken processes rather than fundamentally improving service. Avoiding that trap requires treating each industry as its own operating environment, not simply a configuration setting. 

In retail, that means handling order status, returns, exchanges, subscriptions, and loyalty questions at scale, including peak periods like Black Friday, across systems like Shopify, Salesforce, and Zendesk. One cookware brand automated 80% of call volume this way and saved nearly $100,000 within six months. 

In healthcare, it means completing scheduling, rescheduling, insurance verification, and location questions directly inside systems like EMRs, without staff intervention. Missed appointments cost the U.S. healthcare system more than $150 billion annually, and the difference between a patient who books and one who hangs up and forgets is often just whether the AI can complete that booking in the same call. An urgent care ecosystem saw roughly an 80% reduction in both missed calls and abandoned calls after deploying purpose-built AI. 

In transportation, the bar is daily operational reliability at a scale most platforms never get tested against. A 90% automation rate has to hold on a day like New Year’s Eve, when call volume spikes and any failure can cascade across an entire network. 

The Channel the Market Underrated 

The broader AI conversation has tilted heavily toward chat, but voice continues to matter precisely because customers often call when an issue is urgent, complex, or requires an action to be completed. 

It is also a demanding environment for AI, requiring low latency, strong error recovery, and the ability to maintain context in real time. As adoption grows, the important distinction will be less about whether an interaction happens through voice or chat and more about whether the AI can resolve the customer’s problem. 

The Question That Matters Now 

The value of AI in customer experience ultimately depends on what it can accomplish. Answering a call is not the same as resolving it. Routing a customer is not the same as completing a return, booking an appointment, or modifying a reservation. Those outcomes depend not simply on how powerful the underlying AI is, but on how well that intelligence is connected to the workflows, systems, and edge cases of a particular business. 

The market has largely moved past the question of whether voice AI is worth buying. The question enterprise buyers should be asking next is narrower and more consequential: how much does the AI they are evaluating already know about their industry and workflows, and how much will they have to build themselves?  

Brian Schiff is the cofounder and CEO of Flip, a conversational AI company built for retail, healthcare, and transportation. Flip has automated more than 300 million customer interactions for over 250 enterprise brands.