Product1 publisher3 min readPublished
Five9's CEO reserves complex, high-value and sensitive calls for humans
Amit Mathradas, who took over Five9 in February, says the misconception is that humans are going away from the contact center. The customer result he offers is a containment rate above 50%.
The Product Desk · Product desk
What happened
- Amit Mathradas, who became Five9's chief executive in February after leading Nintex and serving as president and chief operating officer of Avalara, laid out the company's Humantic model in a theCUBE interview.
- Five9 keeps three kinds of interactions with people, sorted on complexity, value and vulnerability, and sends high-volume low-stakes work such as password resets and balance checks to AI.
- He said a 25-year moving and storage customer is on track to route about 100,000 calls a year through Five9's agentic stack by year's end, with containment past 50% and agent turnover falling.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- contradiction SiliconANGLE describes the contained-call share as too narrow a target, and the containment rate is the one figure Mathradas produces to show the human-plus-AI floor working. The deal is still sold on the metric under criticism.
- exposure A team reporting improvement from an internal practitioner survey is exposed if leadership ever asks the customer side the same question, because Five9's own research puts practitioners 33 points ahead of users.
- decision Buyers negotiating a containment target now have a vendor-supplied test for carving out exceptions, and they have to decide whether grief calls and top-account calls sit inside the number or outside it.
- cost Once AI absorbs password resets and balance checks, what is left in the human queue is complex, high-value and sensitive work. That queue changes who gets hired, how long they train, and what a seat costs.
Mathradas made the case with Five9's own research: callers who want a person cannot get one. He said 99% of business practitioners deploying AI think their contact centers and serviceability have gotten better. Only 66% of actual users think the contact center has gotten better [5]. "More than 50% of them are saying the reason is, I want to access a human and I don't get that," Mathradas said [6]. He credited "the research we have done" without giving a sample size or naming who ran it [9].
Those two figures sit 33 points apart [7], and the higher one is a self-assessment by the people who bought and installed the system [5]. On the user side, 34% do not think the contact center got better [8].
Mathradas presents the Humantic model as an adjustment the market was already making. Full automation was never really the goal even if it looked that way for a while, he said, describing the move as a market correction [14]. "I think there is a misconception that all humans are going away in the contact center. That is not true," he said [3]. The interview ran on theCUBE, SiliconANGLE's studio. SiliconANGLE wrote that contact centers have spent the past two years trying to maximize the share of calls AI can handle without a human ever picking up [13].
The customer proof point is measured in that same share. A 25-year moving and storage company is on track to route about 100,000 calls a year through Five9's agentic stack by year's end, Mathradas said. Containment is climbing past 50% and agent turnover is falling as routine work shifts to AI [11]. At better than half containment, more than 50,000 of those calls a year end without a person on the line [12]. Containment is the only number the interview puts on that deployment [11].
For buyers who do not want to move to a new stack, Five9's answer is the open platform, including opening its AI Agent Connect program to third-party voice AI vendors [10]. Mathradas said orchestration is "the next big layer that a lot of customers should be thinking about" [16]. The segment was sponsored. The disclosure as published gives the sponsor's name as the placeholder "[ADD COMPANY NAME]", alongside a statement that neither Five9 nor other sponsors have editorial control over content on theCUBE or SiliconANGLE [17].
The usable part is the sort. Five9 splits interactions on complexity, value and vulnerability. A company's highest-value customer, a person navigating a death in the family, a request that requires walking through complex financial options: all warrant a person, while password resets and balance checks go to AI [4]. An operator can run their own top call reasons through those three columns before agreeing to a containment target. A target set against total call volume scores every one of those interactions as a call that got away. And the two survey numbers are worth keeping apart when the quarterly review comes around: practitioners at 99%, users at 66% [5].
What to watch
- Whether Five9 publishes a sample size and method for the research behind the 99% and 66% figures.
- Whether the moving and storage customer holds containment above 50% as harder call types arrive, and whether Five9 puts a number on the turnover drop.
- Whether third-party voice AI vendors in AI Agent Connect get routed the calls Five9 classes as complex, high-value or vulnerable.