Sam Krut

The through-line

Krut's consistent argument is that voice automation fails not because the technology is weak but because it is built generically and designed against how people actually call. Flip, as he describes it, is "an Alexa like voice bot that sits on your phone channel" that handles tier-one and many tier-two requests, and passes everything it has gathered to a human when it cannot resolve the call [1]. The benchmark he keeps returning to is the gap between what he says legacy vendors deliver, "five 10% automation at best 10% when a company's doing really well 5%", and what he claims Flip does at "40% and above" [1]. His explanations for that gap are all design choices rather than model choices: narrow focus on one vertical, a core system configured per brand, a deliberately non-human voice, and a routing logic that assumes the caller does not know what they want.

On specialising in ecommerce and retail

The first reason he gives for the automation gap is refusal to generalise. Ecommerce and retail are "our bread and butter that's all we focus on", and that focus lets the product be built once and reused: "we have a core system and then it gets configured as needed" [1]. He frames this explicitly as economies of scale, where the shared system already knows the shape of retail contact volume and the per-brand work is configuration to "meet your brand preferences and your brand processes" [1]. The same architecture is what underwrites his implementation claim, that a customer who asked today could be live the same day, a promise he acknowledges other vendors make and grounds in the fact that the system is essentially already built [1]. Integration effort tracks the order management system: brands on what he calls major market OMS platforms such as Shopify, Salesforce and Magento are close to out of the box, while custom or in-house systems are "just a little more involved" [1].

On callers who do not know what they want

Krut's second explanation for performance is that Flip does not treat a call as a menu tree. "We don't view the path as straight" [1]. His example is a customer who knows only that "the shirt that they just purchased is torn" and has no idea whether that is a return, a warranty claim or something else [1]. The work happening under the hood is classifying that raw complaint into the right handling path, taking the caller "through all different avenues and different areas that that may take them", and either resolving it in place or handing the accumulated context to an agent [1]. This is also why he distinguishes the interaction from older IVR: it is conversational, closer to speaking with Alexa than to "the tell me what you need after the beep", and callers can talk to it "very similarly to how you talk to a human being" rather than shouting keywords at it [1].

On sounding like a bot on purpose

One of his more counterintuitive positions is that hiding the machine hurts results. Callers do know they are talking to a bot, and Flip has tested the alternatives, including human voices and more robotic ones, with AI able to make the human version "virtually indistinguishable" [1]. Going fully human sounding "was actually worse" [1]. His reading is psychological rather than technical: people have been trained by Alexa and Siri on how to speak to machines to get results, so a slightly more robotic voice cues that behaviour. "Although you think the CX experience wouldn't be as good the data actually shows that it provides a far better experience and the engagement is much higher" [1].

On CSAT, the agent experience and always-on capacity

Krut resists the framing that automation trades customer satisfaction for cost. Across the brands where Flip has measured CSAT, he reports scores either level with what human agents were achieving or higher overall [1]. The lift, in his account, comes as much from what the bot removes as from what it does: missed calls, abandoned calls and long holds disappear, and agents are freed to spend their time on "those tier 2 requests ones that may need a a bit more love" [1]. Because the handoff carries everything already gathered, he treats improving the agent experience as part of the product rather than a side effect, and argues that the combination of the AI experience and better-prepared agents is what moves CSAT [1]. Capacity is the other half of the case: "flip doesn't take any cigarette breaks", so the system behaves like "an infinite amount of agents that can scale as needed" through peak season and after hours [1].

Takeaways

• The performance gap Krut claims is stark: legacy IVR vendors at 5 to 10% automation against Flip "regularly with Our Brands is performing 40% and above" [1].
• Vertical focus is the strategy, not a limitation. Building only for ecommerce and retail allows one core system plus per-brand configuration, which is also what makes same-day implementation possible [1].
• Callers describe problems, not solutions. The routing logic starts from a torn shirt and works out whether that is a return or a warranty, rather than asking the caller to pick a menu item [1].
• Testing showed a near-human AI voice performed worse than a deliberately more robotic one, because people already know how to speak to bots to get results [1].
• Unresolved calls transfer to an agent along with everything the bot has gathered, so automation is positioned as an improvement to the agent experience as well as the customer's [1].
• Measured CSAT came in at or above human agent levels, attributed to eliminating missed, abandoned and long-hold calls and letting agents concentrate on higher-touch tier-two work [1].
• Integration is close to out of the box on Shopify, Salesforce and Magento, and only somewhat more involved on custom or in-house order management systems [1].