Intryc

What it is

Intryc is an AI quality-assurance platform for customer support teams. It reads conversations out of the helpdesk, scores them against your own scorecard with an LLM, and turns the failures into coaching sessions, training simulations and root-cause reports. The pitch is the one every AI-native QA vendor makes — stop sampling 2-5% of tickets and score all of them — but Intryc sells it to the mid-market with two unusual terms attached: usage-based pricing instead of per-agent seats, and a written accuracy guarantee.

The founders come from the operator side of the problem. CEO Alex Marantelos ran Customer Success for Northern Europe at Confluent; CPO George Pastakas led fraud detection at Revolut. The company went through Y Combinator’s Summer 2024 batch and announced a $3.1M seed in January 2025 led by General Catalyst, with Sequoia scouts, Episode 1 and 500 Emerging Europe participating, for $4.2M raised in total. Y Combinator lists a team of 12. Named customers include Deel, Preply, Blueground, Factorial, Deliverect, Lodgify and BVNK.

Why it shows up in CX and support stacks

• Human agents and AI agents on one scorecard. AutoQA evaluates voice, chat, email and secure-message conversations handled by people and by bots against the same criteria. Once a bot resolves a meaningful share of your queue, that stops being a feature and becomes the only way to compare the two fairly. It is the job Intryc fills in the support quality assurance stack.
• The loop closes inside the product. Scores feed AutoCoaching, which drafts and schedules the coaching session, and Training Simulations, which let a new agent practise against AI customers before touching live tickets. You are not exporting a CSV of failures into a separate LMS.
• Helpdesk-first integrations. One-click connectors cover Zendesk, Intercom, Freshdesk, Salesforce, HubSpot, Kustomer, Gorgias and Aircall, with knowledge sources such as Notion, Confluence and Guru, reporting into Looker, Power BI and Metabase, and alerts into Slack. The vendor counts 40+ integrations.
• Accuracy is a contract term, not an adjective. The “90% Accuracy Promise” waives 100% of the first month’s fees if AutoQA does not exceed 90% accuracy on your real scorecards and tickets in month one, and a separate clause refunds the whole contract if you are unhappy in the first 60 days.
• Security paperwork for the buyer it targets. The vendor lists SOC 2, GDPR and HIPAA compliance.
• Reviewers rate it highly, on a small base. 4.8 out of 5 across 37 G2 reviews, with the move from manual sampling to full coverage and the responsiveness of the team as the recurring praise.

Pricing reality

There is no public rate card. Intryc meters usage — evaluations run — rather than agent seats, bundles every module, and charges nothing for integrations. The only hard numbers it publishes are the conditions of the accuracy guarantee: an annual agreement, at least 1,000 target evaluations a month, and no more than 15 scorecard criteria. Treat 1,000 evaluations a month on an annual contract as the practical floor for the deal the marketing describes.

The vendor’s own buyer’s guide says per-agent QA platforms charge $30-125 per agent per month and that Intryc typically lands about 50% below them at equivalent volume. That is a vendor claim, not a benchmark, but it gives you a planning band: for a 50-agent team, per-agent tools at that range cost roughly $18,000-75,000 a year, and half of that is $9,000-37,500. The reference price to beat is published: Zendesk sells QA inside its Workforce Engagement Bundle at $50 per agent per month paid yearly, or $30,000 a year at 50 agents, with workforce management included.

Usage pricing favours a large team with moderate volume per agent, and punishes the opposite. Before signature, get the per-evaluation rate in writing and multiply it by your real monthly conversation count — every channel, every bot conversation — at 100% coverage.

Best for

Support operations and QA leads at digital-first companies running roughly 20-150 agents on Zendesk, Intercom, Freshdesk or HubSpot, who want 100% coverage, coaching and onboarding simulations from one vendor, and who now have an AI agent handling part of the queue that nobody is grading. It fits best when the team is growing faster than conversation volume per agent, because seat growth does not raise the bill.

Skip it if voice is the majority of your volume and your telephony runs on Five9, Genesys or NICE — its connector list starts at the helpdesk, not the CCaaS layer. Skip it if you run fewer than about 1,000 conversations a month: you fall below the guarantee’s floor and the full-coverage argument has little to cover.

Versus the alternatives

Zendesk QA (formerly Klaus) is the volume incumbent and the default for a Zendesk shop: AutoQA on every conversation, bot and human side by side, bundled with workforce management at $50 per agent per month. Pick it when you are on Zendesk Suite and do not need coaching workflows or simulations beyond what the bundle ships. MaestroQA, now operating as Rippit, is the other long-standing leader — 4.8 out of 5 across 324 G2 reviews, the deepest custom scorecards and calibration workflows — and is the pick when a QA analyst team designs rubrics as its main job and wants humans firmly in the scoring loop. Solidroad is the fastest-growing entrant, with a $25M Series A in April 2026; it makes the same human-plus-bot pitch with simulation-led remediation, and is the pick when grading a third-party AI agent such as Decagon or Sierra is the budget line. Level AI is the pick when the floor is voice-heavy and 100 seats or more.

Intryc wins against all four on the combination of usage pricing and a contractual accuracy floor. If none fit, export one week of transcripts from a single queue, score them with an LLM against your rubric, and double-score 200 by hand. If the AI scores do not change a coaching decision, no QA platform will earn its price yet.

Watch-outs

• The accuracy guarantee has entry conditions. It needs an annual contract, 1,000+ monthly evaluations and a scorecard of 15 criteria or fewer, and it waives one month’s fees — not the contract. Guard: write into the order form how accuracy is measured (agreement with a named human reviewer on a set sample, per criterion), who holds the reference scores, and what happens if accuracy drops below 90% after month one.
• Usage metering can outrun the headcount saving. Adding an AI agent, a new channel or a higher sampling rate raises evaluations without adding a seat. Guard: model your 12-month volume at 100% coverage including bot conversations, and contract a committed volume with the overage rate fixed in advance.
• The vendor is small and last raised in January 2025. Twelve people and $4.2M in total, in a segment where Solidroad has just raised $25M and Zendesk bundles the category. Guard: contract for a full export of scores, scorecard definitions and coaching records in a documented format, and test that export during the pilot.
• The headline outcomes are vendor-reported. Deel’s 40% lift in audit output and the claim that simulations halve onboarding time come from Intryc’s own pages. Guard: set your own baseline — time to proficiency for the last onboarding cohort, reviewer hours per week — before go-live, and review against it at day 60, inside the refund window.