The Voice AI Moat Is the Phone System Itself

Key points

Key takeaways from the Verticals interview with DialStack founder and CEO Mike Sharp (October 2026):

Voice AI economics. Sharp, who has spent roughly two decades at the intersection of telecom and SaaS, argues that a voice AI call costs about $0.40 compared with $7–12 when a person answers the same call, so as latency and emotion improve, venture money is pouring into the category.

From Jive to DialStack. Sharp co-founded Jive Communications in 2005–06, growing it to around $100M in ARR before selling to LogMeIn/GoTo, then led a healthcare IT rollup of 17 brands with a couple of hundred million in revenue; DialStack grew out of the realisation that vertical SaaS platforms had no good option to embed the full phone experience.

The Stripe analogy. DialStack wants to make embedding a full phone system — hardware, 911, the whole stack — as easy as Stripe made embedding payments, for verticals as varied as shooting ranges, dental offices and auto mechanics.

The moat is the whole stack. Sharp argues that voice AI alone is "half pregnant": without owning the phone system, a platform fragments its customers' calling experience and within a year ends up competing with the phone provider, since RingCentral and AT&T are roadmapping voice AI heavily and selling it into the same accounts, making a voice-only vertical SaaS effectively a pipeline for the carrier.

Hairpinning costs. A voice AI layer bolted on top of an incumbent carrier forces calls to hairpin through the carrier and a Twilio or Telnyx number, adding latency and roughly $0.015 per minute in interconnect charges that the platform has to pass on to customers.

Use-case ladder. Sharp ranks voice AI use cases like a hierarchy of needs: upgrading voicemail is the table-stakes bottom tier, a virtual receptionist or switchboard that does basic call decisioning is next, then transactional workflows such as bill payment and account support — still the shakiest tier today — and finally augmentation, an embedded- Gong -style layer that listens to every human calland surfaces bill links or documentation to the front-desk agent in real time.

Consumer sentiment. Roughly 30% of surveyed consumers still say they would hang up on an AI, but Sharp and the hosts expect preference to shift to AI for transactional tasks — his doctor's office scheduling an appointment by text is the example — while only about a quarter of a per cent of global calls currently touch AI.

Evals are not enough. "Just standing up an agent is the easy part," Sharp says; the bar is rising because cold email open rates have halved in two years, willingness to talk to AI on the phone is down 10% in the same period, and winning means owning every phone line, every call and every front desk.

Read more: Why Voice AI Is About to Change Every Vertical | Mike Sharp — Verticals