Intryc reads the 98% of support tickets nobody else does
Most companies grade a sliver of their customer conversations and hope the rest were fine. Intryc, out of Y Combinator's 2024 batch, points AI at all of them - and it just raised a seed round to prove the difference matters.
Every support leader has given the same speech. A furious customer escalates, an executive asks what happened, and the answer starts with a number that sounds like confidence: our quality scores are strong. What that number rarely mentions is the sample size. Most support teams review somewhere between two and five percent of their tickets by hand. The rest go ungraded. Intryc, a company that came out of Y Combinator's Summer 2024 batch, was built on the suspicion that the trouble usually lives in the unread majority.
Intryc uses AI to automate quality assurance for customer support. In plain terms: it reads support conversations, scores them against a team's own rubric, and flags the ones that went wrong - all of them, not a lucky handful. The company describes the goal as evaluating up to 100% of interactions in real time, at roughly half the cost of manual review. That is the whole thesis, and it is a deceptively large one.
01 / The ProblemThe tyranny of the random sample
A random sample is a fine tool for measuring an average and a terrible one for catching the exception. If you read 40 tickets out of 2,000, you learn roughly how the week went. You do not learn about the single conversation where an agent gave a wrong refund policy to an enterprise account that was already halfway out the door. Support quality is not evenly distributed - the damage clusters in the rare, bad interactions, which is exactly where sampling is blind.
Intryc's founders knew this from the inside. Before starting the company in 2023, they spent years in operations-heavy roles where a missed edge case had real cost. The product reflects that background: instead of asking teams to read more tickets, it changes the denominator entirely.
There is a second problem hiding behind the first. Manual QA is slow, and slow feedback is weak feedback. By the time a specialist finishes scoring last week's tickets, the agent who handled them has already worked through another week of conversations, repeating whatever they got wrong. Coaching that arrives late lands soft. Full coverage only matters if it is also fast, which is why Intryc frames its evaluation as happening in real time rather than in a monthly audit cycle.
02 / The ProductWhat Intryc actually ships
The core is AutoQA: AI-driven ticket evaluation built around customizable scorecards and sampling rules, marketed with a 90% accuracy guarantee and continuous learning. A team encodes what "good" looks like - tone, accuracy, policy adherence, resolution - and AutoQA applies it at volume. Around that sit the features that make the scores useful rather than merely damning.
Performance Insights runs root-cause analysis across interactions, with DSAT and sentiment analysis across languages and channels. AutoCoaching turns each evaluation into specific, personalized feedback for the agent, so a low score arrives with a lesson attached. And Simulations takes real past tickets and rebuilds them as training scenarios, letting new hires practice on genuine hard cases before they touch a live queue.
"Quality Assurance of Customer Service conversations has long been a headache for banks and fintechs alike. Intryc uses AI to make that problem go away."
- Tom Blomfield, Y Combinator partner and ex-Monzo founderThe newest layer is the Action Center, a single hub that pulls quality scores, training data, evaluations and AI summaries into one view, plus a QA assistant you can ask questions in plain language - "which agents are slipping on refund policy this month?" - instead of building a report. It is a small feature with a large implication: it treats the QA data as a queryable body of knowledge rather than a dashboard people glance at and forget.
The customizable scorecard is the quiet center of the whole system. Support quality is not one thing - a refund team, a technical-help queue and a fraud-review desk care about very different behaviors. Rather than shipping a fixed rubric, Intryc lets each team define its own criteria and sampling rules, then holds the AI's judgments to a stated accuracy target. That design choice is what lets a fintech and a consumer-rentals company use the same product without either one bending to a generic definition of "good."
03 / The TimingWho checks the chatbot?
There is a reason Intryc's timing looks sharp. Support is being handed to AI agents at speed, and an AI agent can be confidently wrong in ways a spreadsheet audit will never catch. Intryc scores human and AI agents on the same scorecards, which quietly reframes the company from "QA tool" to something more like a control layer for automated customer experience. As more of the frontline becomes software, the job of grading that software stops being optional.
The shift is not incremental. Moving from a low-single-digit sample to full coverage means every ticket - including the rare one that costs an account - lands inside the review set.
04 / The FoundersFraud models to feedback loops
Intryc has three founders, and their prior work reads like a tour of hard operational problems. Alex Marantelos, the CEO, led customer success at Confluent, where he was among the first in EMEA and helped shape how the company onboarded and renewed strategic accounts. Dimitrios Ilias, the CTO, built an early core banking platform at Revolut and worked on natural-language understanding at Amazon and augmented reality at Meta. George Pastakas, the CPO, was an early employee at Pledge and led fraud detection at Revolut, building the ML models and operations behind it.
Fraud detection is a useful lens for what Intryc does. Both problems are about finding the rare, costly event hidden in an enormous stream of ordinary ones. Support QA is fraud detection pointed at conversations instead of transactions.
The team is small and spread out - roughly 21 people working across San Francisco, London and Athens. That geography is not incidental. The founders' networks run through European fintech and West Coast enterprise software, and the company's early customer list mirrors that spread. Running a fast-moving product across three time zones is its own discipline, but it also means someone is usually awake when a customer's queue is busiest.
05 / The MarketA twenty-billion-dollar spreadsheet
Intryc sells subscription software to support and customer-experience teams, priced by role - agents, QA specialists, team leads, and the L&D and CX managers who own quality. It plugs into more than 20 helpdesk and knowledge-base platforms, is SOC 2 certified and GDPR compliant, and counts Deel, Blueground, SadaPay and Ziina among named customers - a spread across fintech and consumer that suggests the pain is not industry-specific.
Those logos are worth pausing on. Deel is a global payroll and compliance company handling money and contracts across borders; SadaPay and Ziina are consumer fintechs in markets where trust is the whole product; Blueground rents furnished apartments where a bad support experience can sour an entire stay. What they share is stakes. When the cost of a mishandled conversation is high, the appeal of grading every conversation rather than a sample stops being abstract. The businesses most drawn to full coverage are the ones that can least afford to miss the bad ticket.
The competitive set includes established support-QA and conversation-intelligence tools - MaestroQA, Zendesk's Klaus, Loris, Level AI - but the more honest competitor is the status quo: a QA specialist, a spreadsheet, and a sample. Intryc's argument is less "we are better software" and more "the entire method was wrong."
The framing the company now uses - QA infrastructure for AI-driven customer experience - is a deliberate one. It positions Intryc not as a point tool that lives inside a support org's monthly routine, but as a governance layer that becomes more necessary as automation grows. If a company is going to let software answer its customers, someone has to define what a good answer looks like and check that it keeps happening. That is a durable place to sit, and it is a different pitch than "review your tickets faster."
Whether Intryc becomes that layer or one of several tools competing for it is the open question, and it is too early to call. What is clear is the shape of the wager. The company is betting that support quality is a measurable quantity, that the measurement is worth automating, and that the moment to do it is exactly as human and machine agents start sharing the same queue.
| Founded | 2023 |
|---|---|
| Batch | Y Combinator S24 |
| Headquarters | San Francisco (hubs in London & Athens) |
| Team size | ~21 |
| Seed round | $3.1M, oversubscribed (Jan 2025) |
| Lead investor | General Catalyst; Sequoia scouts |
| Total funding | ~$4.2M |
06 / The MoneyAn oversubscribed seed
In January 2025 Intryc closed a $3.1M seed round, described as oversubscribed, led by General Catalyst with participation from Sequoia scouts and existing backers Episode 1 and 500 Emerging Europe. That brought total funding to roughly $4.2M. It is an early-stage number, but the investor list and the caliber of early logos suggest the company found real pull before it had raised much at all - early adopters reportedly doubled their evaluation output without adding headcount.
The bet underneath all of it is simple enough to fit on a sticky note. The tickets you never read are the ones that hurt you. Intryc's whole company is an argument that you should stop guessing about them.