Rippit

What it is

Rippit is MaestroQA renamed. Same company, same founder — Vasu Prathipati — same conversation data, and a position that moved from support quality assurance to what rippit.com now calls “The AI Conversation Data Platform.” The rebrand landed in the first quarter of 2026. The exact date is muddy: third-party trackers split between 24 February and 4 March 2026, and Rippit’s own blog carries no dated announcement post, which matters if you need a paper trail for a vendor-change review.

The product claim is narrow and testable — turn unstructured conversations into “structured, queryable data,” then analyse 100% of them instead of a sample. Rippit’s engineering argument for why that needs a product rather than a warehouse query is the most useful thing it publishes: an average support transcript runs about 34,000 characters, a two-month window in its worked example totals roughly 340 million characters, and the standard Claude-plus-Snowflake approach handles that by sampling 50 conversations. Rippit sells the other 99%. The same post concedes the tradeoff in one line — “100% coverage costs more” — without quantifying it.

Named customers are Brex, Checkr, Klaviyo, Peloton, Resident, SpotOn and LegalZoom: mid-market and up, high conversation volume, support-heavy. Funding is $31.6M in total, and the last round was a $25M Series A in 2021. Five years without a raise, plus a rebrand into a more crowded category, is the context for everything below.

The split that decides your evaluation

As of today there are two live websites selling two motions from one company, and no comparison page says so.

• maestroqa.com still sells the QA suite — Auto QA, Scorecard Builder, Calibrations, Coaching, QA Reporting, Screen Capture, Root Cause Analysis — across 32 named use cases and roughly 60 integrations: helpdesks (Zendesk, Intercom, Gladly, Kustomer, Front, Freshdesk), AI agents (Ada, Decagon, Forethought, Sierra, Agentforce), phone systems (Five9, Talkdesk, NICE inContact, Aircall, Dialpad, Twilio, Zoom), warehouses, and WFM (Assembled, Calabrio, Playvox). Pricing is a contact form.
• rippit.com sells Agent Apps on conversation data with published prices, a free tier, and exactly two one-click sources: Zendesk or Intercom.

Ask which SKU your contract sits on before you ask anything else. The answer determines your rate card, your feature set, and which of the two roadmaps you are buying into.

Why it shows up in CX stacks

• Ingestion is not the constraint. 35+ one-click integrations across calls, messages, phone and chat; sync from Snowflake, Databricks, BigQuery or Redshift; direct push through developer SDKs. Rippit reads the conversation wherever it already lives, which is the difference between a two-week connection project and a quarter of data engineering.
• It ships seven Claude Skills, named and downloadable. /rippit-double-click builds an evidence-backed answer on a theme, /rippit-escalation-finder posts conversations heading sideways into Slack, /rippit-support-shoutouts names and quotes the agent, /rippit-chatbot-analysis grades bot-handled conversations against escalations and the knowledge base, /rippit-aeo-seo-optimization drafts content grounded in questions real customers asked, /rippit-tool-bakeoff runs the same analysis in competing tools, and /rippit-qa-program-revamp rebuilds evaluation beyond per-agent scoring. The page states the constraint plainly: they run only if your Claude setup already has MCP access to the conversation data. Rippit does not ship a documented MCP server of its own alongside them.
• AI credits are passed through at cost, 1:1. Rippit’s argument against running the same analysis on a warehouse is that Snowflake marks the tokens up 10-25%. That is a vendor claim about a competitor’s pricing, so verify it against your own Snowflake bill — but the pass-through on Rippit’s side is stated on the pricing page, including $100 of credit on the free tier.
• Chatbot QA is a first-class surface, not a bolt-on. Native connections to Ada, Decagon, Forethought, Sierra and Agentforce mean the thing being scored can be your AI agent rather than your humans. For teams that deployed an agent in 2025 and still have no evaluation loop over it, this is the shortest path to one.
• The QA depth is genuine and independently rated. The MaestroQA listing holds 4.8 out of 5 across 324 G2 reviews, with conditional scorecard logic and evaluator calibration as the repeatedly-cited strengths. That is a decade of QA product, and it did not disappear when the logo changed.
• Brex runs it past QA. The customer story reports churn risk detected in 3% of conversations, with the operations lead framing the old motion bluntly: “We were spending hours on QA every week, but it wasn’t driving customer outcomes—it was just generating scores.”

Pricing reality

Two rate cards, and they are not close.

Rippit self-serve is published: Free at $0 for 100 Agent Runs a month, unlimited Q&A, one Zendesk or Intercom connection, and $100 of at-cost AI credits; Starter at $185/mo for 300 Agent Runs; Growth at $495/mo for 1,000 Agent Runs; Enterprise quoted, for 1,000+ runs plus a dedicated AI Agent Specialist, SSO and advanced integration. The meter is the Agent Run — not the seat, not the conversation. Nothing published defines how many conversations one run consumes, and that single number decides whether $495/mo is generous or a rounding error against your volume.

Legacy MaestroQA stays quote-only. Vendr’s dataset, updated February 2026, puts average contract value at $23,520 across 55 transactions, banded at $6,000-18,000/yr for 5-25 agents, $18,000-60,000 for 25-100, and $60,000-200,000+ above 100. Observed mid-market deals at 50-75 agents landed between $35,000 and $70,000. Buyers took 15-30% off opening quotes, and multi-year commitments came in 15-25% below single-year pricing. Open at the bottom of your band and cite the free tier — a vendor publishing a $0 entry point has weakened its own floor argument.

Best for

Support, CX and customer operations leaders at mid-market and enterprise volume — roughly 25 agents and up, on Zendesk or Intercom, with a conversation archive nobody reads. It is the strongest pick when you already run a QA program and want the analysis layer over the same data rather than a second vendor, when the thing you need scored is an AI agent you deployed on Ada, Decagon, Sierra or Forethought, and when your analytics team wants conversation data in Snowflake as structured rows instead of transcripts.

Skip it if you have under 15 agents and no analyst — the self-serve tier will read as an expensive query box. Skip it if what you actually want is agent training simulation and role-play, which is a different product. And skip it if your buying committee needs a vendor whose category identity is settled; the ground moved here in the last six months and it will move again.

Versus the alternatives

Zendesk QA (formerly Klaus) is the volume incumbent and the cheapest first test for any shop already on Zendesk — take it when QA scoring is the whole job and the data never has to leave the helpdesk. Solidroad is the fastest-growing entrant in support QA and the sharper pick when coaching and simulated practice matter more than analytics breadth; run it head-to-head with Rippit if your problem is agent performance rather than customer insight. Enterpret is the closer substitute for the new Rippit — adaptive-taxonomy feedback analytics with a published MCP server and a far wider source list — and it wins when your question is product-shaped (“what should we build”) rather than operations-shaped (“what went wrong in the queue”). Assembled overlaps only at the edges; it owns forecasting and scheduling and is a complement, not a swap. Gong covers the same analytical move on sales calls instead of support tickets, so if your conversation volume is revenue-side, start there. Rippit’s own blog picks the fight it wants — a post titled “Walking Dead: Qualtrics and Medallia” — and against survey-led VoC the argument is fair: mining what customers already said beats asking them again. The honest baseline is your warehouse plus an LLM and an analyst, and Rippit’s answer to it is coverage: sampling 50 conversations answers a different question than reading all of them.

Watch-outs

• You bought a QA vendor and the vendor stopped calling itself one. Founder Vasu Prathipati’s public framing was that QA reads as the past in an AI-first world. Scorecard Builder, Calibrations and Screen Capture are still on maestroqa.com today, but roadmap attention follows positioning. Guard: name the specific QA features you depend on in the renewal document, attach a written end-of-life notice period of at least 12 months per named feature, and refuse a renewal that only references “the platform.”
• The free tier is not the product you would sign. Free and Starter connect one Zendesk or Intercom workspace; the QA suite, the phone-system connectors and SSO sit on quoted tiers. Guard: run the evaluation inside the tier you intend to buy, on your own data, with your own scorecards loaded — a pilot on the $0 tier validates a query box, not a QA program.
• The Agent Run is an undefined billing unit. Every published tier meters it and no published page says what one covers. Guard: get the definition in writing before signature, model one month of your real conversation volume against it, and put a ceiling in the contract with overage priced in advance rather than at renewal.
• The company has not raised since 2021 and headcount reporting is inconsistent — trackers list 71 and 33 employees for overlapping periods, a spread wide enough that at least one is stale. Guard: run the standard viability checks — request a contractual data-export right, keep the Snowflake or SDK export path live from day one so your conversation history is never trapped, and re-check the funding position at each renewal.
• Every performance number here is vendor-published. 100% coverage, the 10-25% Snowflake token markup, Brex’s 3% churn-risk detection: all company material, none audited. Guard: pick one question you already know the answer to from a manual review, run it through Rippit during the trial, and compare conclusions rather than coverage percentages. A tool that reads everything and reaches the wrong conclusion is worse than a sample that reaches the right one.
• The Claude Skills depend on plumbing you provide. They are downloadable, they are specific, and they do nothing until your Claude workspace has MCP access to the conversation data. Guard: confirm the exact connection path during the trial, with your security team in the room, before the skills appear in a business case.