Startup Spotlight: Solidroad drops QA after it consumed 90% of the work

Why it matters

Solidroad's shift shows the cost of serving adjacent enterprise needs: QA became a large part of the workload while training grew faster. Its new bet is that interactive practice for both conversations and software workflows can stand on its own.

Solidroad is dropping its quality-assurance product to focus on training simulations, a sharp narrowing after it raised a $25 million Series A to build tools for evaluating customer-support conversations. CEO Mark Hughes (@markhuqhes) announced the change on September 22nd, saying QA had consumed about 90% of the effort across engineering, sales, support and customer success even as the training business grew faster.

That decision puts Hughes's original thesis back at the center: people learn customer-facing work by practicing it, not by sitting through materials and hoping the live queue is forgiving. Solidroad began with simulated customer conversations for support reps. Its next move is to reproduce the software and workflows those reps must use, giving them a place to practice the clicks as well as the conversation.

The shift is consequential because Solidroad's previous pitch linked QA and training in a feedback loop: assess live interactions, find skill gaps, then create practice around them. Now Hughes says Solidroad is willing to abandon one half of that loop to concentrate on the part customers appear to want more. Whether the new simulations can sustain Solidroad without QA is the central bet.

A founder returning to the problem

Hughes has spent much of his career close to customer-facing teams. He was an early sales hire at Intercom, then led sales, support and customer success for EMEA at Chargify, now Maxio. At Solidroad, his co-founder and CTO Patrick Finlay had worked as a product engineer at Intercom; the two had collaborated there before starting the company.

Hughes also has already built and sold a startup. Before Solidroad, he founded Gradguide, a coaching and mentorship network for recent graduates. In his Y Combinator founder profile, Hughes says Gradguide raised about EUR 2 million, reached 20 employees and exited in late 2022. The profile does not give a sale price or describe the transaction structure, so the exit establishes experience building and selling a company, not a valuation benchmark.

That earlier work helps explain why Hughes describes Solidroad as a skills company as much as a customer-service tool. In a September 22nd post on LinkedIn, he called its original ambition a "flight simulator for work": a place to practice hard conversations and make mistakes before they affect a customer. Gradguide and Solidroad serve different markets, but both put practice at the center of the product.

Finlay brings a related builder's background. The YC profile says he co-founded Monaru, a platform for creators selling live classes, before joining Intercom as a product engineer. Together, Hughes's experience selling into support teams and Finlay's experience building software for them gave Solidroad a direct line to the workflow it set out to improve. Their prior work put them close to the customer-support workflows Solidroad set out to improve.

From conversation practice to system practice

Solidroad's initial product lets support representatives rehearse customer interactions in simulated scenarios. Solidroad says its AI can act like a customer, score a practice interaction against criteria such as accuracy and process adherence, and give targeted feedback. The same scorecards can be applied to real interactions, connecting training to what happens in customer channels.

The new System Simulations product extends that idea from what an employee says to what they do. Hughes says Solidroad has developed a way to turn screen recordings of software into interactive digital replicas. Employees can click through buttons, complete workflows and make mistakes in a simulated environment, then receive feedback and try again. The intended result is practice with the systems a job actually requires, rather than another video or static training document.

That product move has a clear operational target. New support hires need to learn company-specific rules and tools before handling real cases; BPOs, which deliver outsourced contact-center services, often need to bring large cohorts up to speed. Solidroad's YC profile says the company reduced onboarding times by 50% at BPO customers PartnerHero and Tech Mahindra. It also says Crypto.com's go-live customer-satisfaction score rose three percentage points to above 90%.

Those results are company-reported, and the profile does not spell out the measurement windows, baselines or comparison groups. Solidroad's Crypto.com case study gives further detail: it says Crypto.com uses simulations to validate new agents before live work and scores more than 800,000 conversations monthly, reporting an 18% drop in average handling time and a three-point CSAT gain. The case study names Alex Dimitrov, Crypto.com's head of training, but the figures remain presented through Solidroad's own materials rather than an independently described evaluation.

The outcomes are measures operators track: training time, readiness to handle real cases, customer satisfaction and handling time. They do not, by themselves, establish that simulations caused the improvements or that the same results generalize across customers. Solidroad's sharper test now is whether practice can cover both the human interaction and the software process without becoming a bespoke content project for each customer.

QA was the detour and the data source

Hughes says Solidroad followed customer demand into QA after starting with training. Solidroad built software to evaluate real customer conversations, and that work became a meaningful part of the business. In Hughes's account, however, QA was growing more slowly than training while consuming roughly nine-tenths of company effort. That allocation, rather than a lack of relevance to customers, is the stated reason for cutting it.

The tradeoff is real. QA gave Solidroad visibility into where interactions broke down and helped identify what employees needed to practice. Hughes says the work taught the company how organizations operate, where processes fail and where training falls short. That accumulated knowledge is part of the rationale for System Simulations: if the product understands the work well enough, it can turn operating procedures into practice environments.

Yet customers do not necessarily buy the pieces in the same way, and a training simulation must prove its value independently of the QA product that surfaced the need. Hughes wrote that Solidroad had spoken with QA customers and had a plan to find them a new home. He also said some were continuing with Solidroad by expanding into training. Those are management's descriptions of the transition; they do not establish how many customers will move, what happens to every QA contract, or how much revenue is being left behind.

There is also a visible product-transition wrinkle. On October 10th, Solidroad's homepage continued to market custom scorecards and described QA and training as working together, while Hughes's post said the QA product would be sunset. That may reflect a transition in how existing features and customers are handled, but the public materials do not settle the product's precise sunset scope. Customers deciding whether to adopt Solidroad still need clarity on which QA features will remain available and when the sunset will take effect.

The capital and the next proof point

Solidroad entered Y Combinator's Winter 2025 batch. Solidroad announced a $6.5 million seed round in June 2025, led by First Round Capital with participation from Y Combinator, and said that brought its funding to $8 million. On April 16th, 2026, Solidroad announced a $25 million Series A led by Hedosophia, with First Round, Y Combinator and Sony Innovation Fund participating. Solidroad has not disclosed a valuation in those announcements.

The Series A came when Solidroad was pitching automated QA across human and AI-led support. The September decision redirects the team's effort toward training, now expanded to interactive software replicas. The capital gives Solidroad resources to make that bet, but the round itself does not establish demand or prove the new product's economics.

The contrast with Solidroad's 2025 pitch is notable. The seed announcement described a platform analyzing hundreds of thousands of conversations each month for more than 50 customers and cited outcomes at Crypto.com, Podium and ActiveCampaign. Those were Solidroad-supplied figures and examples at the time. Solidroad's new case is more focused: turn the complexity of customer operations into realistic practice, then measure whether people become ready to do the work sooner and perform it better.

That narrower claim fits Hughes's experience. He has sold customer-support software, built a prior coaching business and watched Solidroad customers pull the product toward QA. Now he is choosing to spend Solidroad's engineering effort on the training problem he says motivated its founding. The next evidence will be whether Solidroad can turn its promising onboarding and customer-service results into repeatable simulations for both conversations and workplace systems, without relying on the QA product it has chosen to leave behind.