Solidroad

What it is

Solidroad scores customer support conversations — human-handled and bot-handled — against one scorecard. It reads closed conversations out of the helpdesk (Zendesk, Intercom, Gladly, Gorgias, Help Scout, ServiceNow), grades all of them across chat, email, phone and video, and routes each failure to the remediation that fits the party who failed: for a human, a practice simulation that talks like your customers; for a bot, a scorecard revision.

Two founders out of Intercom, Mark Hughes and Patrick Finlay, started it in 2023. It went through Y Combinator’s Winter 2025 batch and raised a $25M Series A on 16 April 2026 led by Hedosophia, with First Round Capital, Y Combinator and Sony Innovation Fund participating. Ryanair, ŌURA, Crypto.com and ActiveCampaign are on the public customer list.

The category leader it most resembles is Zendesk QA, the former Klaus. The difference that decides the purchase is not feature depth — it is that Solidroad does not sell the AI agent it is grading.

Why it shows up in Customer Success stacks

• It grades an agent it has no stake in. Solidroad connects to Decagon, Sierra and Fin (Intercom) and evaluates their output as a third party. That matters because of how the agents now bill. Zendesk charges $1.50 per automated resolution committed or $2.00 pay-as-you-go past an allowance of 5, 10 or 15 per agent per month, and it decides what counts as a resolution using its own model after a 72-hour no-follow-up window. When the vendor defining the outcome also sells the report card, the report card is not evidence. This is the seat Solidroad fills in the support quality assurance stack.
• The scorecard is your policy, not a generic rubric. It imports policies and macros from Guru and Notion, so a conversation gets marked down against the refund rule your team actually wrote rather than a vendor template. For an AI agent that is the only version of QA that produces an actionable finding: “violated policy 4.2” is a prompt change, “tone was off” is not.
• One finding, two remediation paths. A human who lacked context and a bot that lacked context fail identically on the transcript and differently in the fix. Solidroad splits them — coaching assignment on one side, scorecard and knowledge revision on the other — instead of filing both as a training gap.
• Chat and email are scored as first-class surfaces. The voice-first QA incumbents were built for call centers and treat text as a secondary channel. For a B2B support org where phone is a fraction of volume, that inversion is the whole product.

Pricing reality

Solidroad publishes no pricing. No pricing page, no free tier, no self-serve plan — the only route to a number is a demo. Put that at the top of your evaluation, because it changes the shape of the process: you cannot size the business case before the first sales call, which is exactly what the rest of this category now lets you do.

Third-party marketplace estimates put mid-market deployments at approximately $50-150 per user per month — an estimate with no vendor rate card behind it, so treat it as a planning placeholder and not a budget line. At 40 agents that spans roughly $24,000 to $72,000 a year, a band too wide to approve.

The number to negotiate against is published and belongs to a competitor. Zendesk sells QA inside its Workforce Engagement Bundle at $50 per agent per month paid yearly, and that bundle absorbs workforce management too. Take $50 × your seat count as the reference price, then make Solidroad beat it on something other than cost — because on cost, against a bundle that also replaces your WFM line, it will not.

Best for

Support and CX leaders running roughly 25 to 250 agents alongside a third-party AI agent, where the helpdesk vendor also sells the QA product and bills per resolution. The fit is sharpest when the agent and the helpdesk are the same company: that is the case where a second contract for an independent scorer buys something a bundled tool structurally cannot, and where the reopen rate — a conversation billed as resolved that comes back and gets handled twice — is the number the audit has to defend.

Do not buy it if

Your team is all-human, under about 20 agents, and on one helpdesk. Native AutoQA inside Zendesk or Intercom covers that at no additional contract, and the independence argument has nothing to bite on when no vendor is metering your resolutions.

Skip it too if you run a voice-first contact center with real-time compliance obligations. Solidroad grades conversations after they close; it does not coach an agent mid-call or redact PCI data in the stream, and buying it for a floor of 300 phone agents means buying the wrong half of the category.

And skip it if procurement requires a published rate card to open an evaluation. That is not a knock on the product, it is a statement about your process — quote-only vendors lose those cycles on the calendar, not the merits.

Versus the alternatives

• Zendesk QA (formerly Klaus) — the volume incumbent, and the default. AutoQA across every interaction including voice and AI agents, Spotlight for churn risk and stuck loops, bot and human scores side by side, at $50 per agent per month paid yearly in the Workforce Engagement Bundle. Pick it when the AI agent resolving your tickets belongs to someone else — then Zendesk is grading a competitor’s bot, the conflict inverts, and the bundle is the cheapest credible answer.
• MaestroQA — the deepest custom scorecards and multi-step review workflows in the category, and the pick when a QA analyst team is designing calibrated rubrics as its actual job. Quote-only: Solidroad’s own comparison page puts it at roughly $35,000-$70,000 a year for 50-75 agents and marks the figure directional, which is a vendor-sourced estimate and should be treated as one. Pick it over Solidroad when human review workflow design outweighs AI-agent evaluation.
• Observe.AI and Level AI — the voice pole. Real-time in-call coaching, compliance alerts while the call is live, speech analytics at enterprise scale. Observe.AI is publicly reported in the $100-500 per seat per month range. Pick either when voice is the majority channel; both were built for the contact center floor and Solidroad was not.
• Lorikeet — the fastest-growing entrant, and the sharpest test of Solidroad’s thesis. Its Coach product, added January 2026, scores quality and runs root-cause analysis on every ticket, and it ships alongside Lorikeet’s own resolving agent. Pick it when you want the agent and its QA on one contract and you accept that the grader works for the graded. Pick Solidroad when you have decided you do not.

Watch-outs

• Independent scoring stops being independent the moment Solidroad ships its own resolving agent. The Y Combinator description already reads “AI agents for CX teams, starting with training and QA,” and the $25M raise funds the rest of that sentence. Guard: write the independence claim into the contract as a term — a notification obligation and an exit right if the vendor launches a competing customer-facing agent — rather than trusting the current product boundary.
• Machine scores move when the model or scorecard moves, with no change in agent behavior. A rubric revision can swing your quarterly average and read as a performance story. Guard: freeze a gold set of 50 hand-scored conversations at go-live and re-run it after every scorecard or model change; movement on the frozen set is a measurement artifact, not a result.
• Independent review volume is thin — G2 carried 3 reviews at the time of writing. Named logos are doing the work that a review corpus normally does, and logos tell you a deal closed, not that it renewed. Guard: ask for two references at your agent count and channel mix, and require a paid pilot scored against your own gold set before an annual commit.
• No published API or MCP server. Every catalogued integration is one Solidroad built, which caps you at their roadmap the moment you want scores in your own warehouse or in a reconciliation ledger the audited vendor cannot rewrite. Guard: make a scheduled export of scores at conversation-ID granularity a signed deliverable, and confirm the format before signing — the audit only holds if the ledger lives somewhere you control.