Mark Hughes, Co-Founder and CEO at Solidroad – Interview Series

Mark Hughes, Co-Founder and CEO at Solidroad, is a repeat entrepreneur and go-to-market leader who has built his career at the intersection of sales, customer experience, and technology. He previously founded Gradguide, a career and recruitment platform that raised €2 million and was later acquired, and held senior commercial roles at Chargify, where he led EMEA operations following its merger into Maxio. Earlier in his career, he developed deep expertise in high-velocity and enterprise sales at Intercom, managing complex deal cycles and helping scale revenue teams. Today, he leads Solidroad from San Francisco, applying that experience to rethinking how companies evaluate and improve customer interactions using AI, with a focus on raising the quality bar across both human and automated support systems.

Solidroad is an AI-powered quality assurance and training platform designed for modern customer experience teams, built to analyze and improve every customer interaction at scale. The platform ingests conversations across channels, evaluates performance against company standards, and generates personalized training simulations to continuously optimize both human agents and AI systems. Rather than replacing customer support teams, Solidroad focuses on enhancing them, turning each interaction into a feedback loop that improves outcomes like response quality, efficiency, and customer satisfaction. The company positions itself as infrastructure for high-performing CX teams, helping organizations deliver consistent, high-quality experiences without increasing operational costs.

Most companies still rely on reviewing a tiny sample of customer interactions to evaluate performance. What convinced you that this approach is fundamentally broken, and how did that realization lead to building Solidroad’s approach to continuously monitoring, scoring, and improving both human and AI agents?

Patrick and I had both spent years around customer support teams before starting Solidroad, and one problem kept standing out: companies were having hundreds of thousands of conversations with customers, but if you asked how those conversations were actually going, they didn’t really know. And what they did know was based on such a small sample of the interactions that it wasn’t accurate.

That was already a shaky foundation when humans were handling every ticket. Now, with AI entering the picture, stakes have changed. One bad pattern could play out across thousands of conversations simultaneously, and go unnoticed because most teams are only reviewing 1-2% of interactions.

We saw this happening at companies over and over again and they were losing customers because of it, so we decided to create a solution. We recognized this ultimately was an infrastructure problem. Companies simply didn’t have the systems needed to truly understand the performance of their customer support agents.

Solidroad was built to address that blind spot. We give companies the tools to see what’s really happening at scale, and ensure every conversation—human, AI, or both—delivers value.

Solidroad has been described as a “flight simulator” for customer-facing teams. Can you walk us through how your AI actually simulates real customer interactions, and what makes those simulations effective for training at scale?

The flight simulator analogy works because the core idea is the same. You don’t want someone’s first experience handling a difficult situation to happen in front of a real customer.

What makes our simulations effective is that they’re grounded in what’s actually happening in that company’s real conversations. When an agent gets something wrong in a live interaction, the system generates a targeted simulation around exactly that type of scenario so they can practice it before it happens again. It’s not generic training content.

The feedback loop is what drives the learning. Agents run through a scenario, get specific guidance on what worked and what didn’t, and try again. Mistakes happen in a safe environment, and the learning sticks because it’s tied to real situations rather than abstract classroom exercises.

Your platform doesn’t just train agents, it also scores live interactions against custom guidelines. How do you design these scoring systems to reflect real business outcomes like CSAT, retention, or revenue?

To create these guidelines, we always start with what that specific company actually cares about. A financial services company handling billing disputes has different quality standards than an e-commerce brand managing returns during peak season. So scoring is built around each organization’s own guidelines, their policies, their brand voice, and what a good resolution looks like for them.

And business outcomes like improved CSAT scores and retention come from high quality customer interactions. Rather than tracking scores retroactively, we focus on the behaviors that predict those outcomes: consistent performance across agents, following the right processes, and the soft skills that shape how customers feel at the end of a conversation.

The goal is to give leaders a clear, practical picture of what good looks like for their company, so they can coach their teams, replicate success, and grow it across their organization.

Many companies only review a small percentage of customer conversations. How does Solidroad enable full coverage analysis, and what kinds of insights become possible when you move from sampling to analyzing everything?

Our recent State of CX survey showed that around 81% of support conversations are never reviewed for quality, leaving teams reviewing such a small fraction of conversations that they’re essentially sampling and hoping it’s representative. When a company starts to evaluate every customer conversation, patterns that would never surface in a random sample become obvious. Teams start to see which types of requests are consistently handled poorly, where AI and human agents are diverging in unexpected ways, and which issues are recurring before they ever show up in a customer complaint.

Solidroad uses AI to automatically review every customer conversation across chat, email, and voice, making full-coverage analysis possible. That shift from sampling to 100% review is what reveals consistent patterns in quality, customer friction, and performance that would otherwise stay hidden.

Crypto.com is a great example of what full coverage actually unlocks. Before Solidroad, their team had no reliable way to measure agent quality at scale or verify agents were ready before handling live tickets. Problems were only surfacing after they had already affected customers. By moving to automated scoring across 800,000 monthly conversations, they could catch quality issues early, validate agents before deployment, and confirm improvements were actually sticking. The results were a 18% reduction in average handling time and a 3% increase in CSAT (which is significant at the scale they operate at). As their conversation volume continues to grow, their quality visibility scales with it rather than falling further behind.

That’s really what full coverage changes. It shifts quality from something reactive, to something proactive that can be managed ahead of time.

You work with companies like Crypto.com and Ryanair, where customer experience is mission-critical. What patterns or common weaknesses have you identified across large-scale support teams?

A few things come up consistently. The first is the gap between what companies think is happening in their customer conversations and what’s actually happening. Most teams are confident in their support quality right up until they get full visibility, then they realize the picture is more complicated than their metrics initially suggested.

There’s also a consistent disconnect between how teams measure performance and what actually drives customer outcomes. Speed metrics and ticket counts are easy to track, so they tend to dominate evaluations. But those numbers don’t tell companies if the customer’s issue was resolved, the agent represented the brand accurately, or whether the interaction left the customer feeling good about the company. In high-stakes environments like fintech or healthcare, that misalignment between measurement and outcome can have serious consequences.

There’s growing concern that AI in customer service can degrade the human experience. How do you ensure that your system improves quality rather than pushing teams toward overly scripted or robotic interactions?

It’s a valid concern, but one that usually comes from QA (Quality Assurance) systems that leverage AI for the wrong things. If a company scores agents purely on adherence to a script or how quickly they close a ticket, they’ll end up with interactions that technically check the boxes but feel hollow or impersonal to the customer.

Our approach is to build scoring around what actually makes a customer leave an interaction feeling supported. We look at things like if the agent listened actively, showed empathy when needed, and actually helped the customer solve their issue.

The same applies to AI agents. The goal is to use AI to make customer interactions more consistent, accurate, and appropriately responsive to what the customer is actually experiencing. When quality oversight is built around those outcomes rather than just process compliance, it tends to push interactions in a better direction, not a more scripted one.

Solidroad sits at the intersection of human agents and AI agents. How do you see the relationship evolving between the two, especially as AI begins to handle more frontline interactions?

It’s important to view human agents and AI agents as a team, splitting division of labor. I think the future for human and AI agents is hybrid.

AI handles high-volume, straightforward requests very well, and the best AI agents are even resolving most of those conversations on their own, which is really impressive.

But the result of that is that the interactions reaching human agents are increasingly the complex, emotionally charged, high-stakes ones. The customer who’s frustrated, the situation that doesn’t fit a standard template, the conversation that requires real judgment. So the bar for human agents is actually rising, not falling.

That’s exactly why oversight across both becomes so important. AI needs to handle its volume consistently and accurately. Humans need to be well-prepared for the harder conversations they’re now primarily handling. And there needs to be an independent layer sitting across both, giving companies a clear picture of what’s working and what isn’t. That’s the hybrid model we think will define CX going forward, and it’s what we’re building toward.

Your system provides real-time feedback and coaching. How important is immediacy in improving agent performance, and how does that compare to traditional training and QA workflows?

Immediacy is really important, and the research supports that feedback is most effective when it’s connected to the specific situation that generated it. Traditional QA workflows break that connection almost by design. A manager reviews a conversation days or weeks after it happened, shares feedback in a periodic review, and by then the agent has had dozens of other interactions where that mistake was likely repeated. Without providing feedback in real time, mistakes persist and agents have to unlearn practices they’re always used to.

What we’ve found is that feedback is most effective when it surfaces immediately after a real conversation. It works best when it’s tied directly to what the agent just handled and paired with a specific simulation they can run right away. In this format, it translates into actual behavioral change much faster. Agents aren’t just hearing that they need to improve, they’re practicing the improvement in context, while it’s still fresh.

The contrast with traditional onboarding is the clearest example of this. Our data shows that over half of human agents say the hardest part of onboarding is applying what they learned in training to real customer situations. That discrepancy exists because classroom learning and live support feel nothing alike. Continuous, situational feedback solves this in a way that periodic reviews never really could.

You’ve shown improvements like reduced onboarding time and higher CSAT. Which metric do customers care about most when adopting your platform, and how do you demonstrate ROI early in the relationship?

It varies by where the customer’s pain points lie, but the two things that come up most consistently are quality assurance coverage and time savings. Teams that are manually reviewing a small fraction of conversations immediately see the value in moving to 100% coverage, both for the insights Solidroad surfaces and for the hours it gives back to their QA teams. Across our customer base, we typically see a 20x increase in QA coverage and 90% reduction in manual review time.

On the ROI side, we try to connect outcomes to metrics that already matter to that business. At Podium, the headline was onboarding time. New agents hitting performance benchmarks in 60 days instead of 90, and resolving issues 33% faster once they went live. With Crypto.com, it was resolution time and conversation volume visibility. For Ryanair, it showed up in recruiter hours saved.

The specifics differ, but the pattern is the same: when you can actually see what’s happening across your customer interactions and act on it quickly, improvements follow in the metrics that matter.

Looking ahead, do you see Solidroad remaining focused on training and QA, or evolving into a broader orchestration layer for managing both human and AI customer interactions?

Training and QA are where we started, but the vision is bigger than that. The way we think about it, quality oversight is going to become key infrastructure for any company running AI in customer support. In the same way security certifications became non-negotiable once companies moved data to the cloud, quality certification will become essential as customer conversations move to AI.

The north star is that “Solidroad Certified” carries real meaning: proof that a company meets a high bar for how it treats customers, regardless of whether the interaction was handled by a person or an AI.

What that means practically is that we want to be the system organizations rely on to manage and improve customer interaction quality across the board, not just flag issues on the back end. That’s a big part of why we raised our $25 million Series A led by Hedosophia—it gives us the runway to build toward that vision. That includes expanding our product capabilities and teams to help even more companies evaluate 100% of their customer interactions.

Thank you for the great interview, readers who wish to learn more should visit Solidroad.