Zendesk QA

What it is

Zendesk QA is Zendesk’s quality-assurance add-on: it scores support conversations — human and AI-agent, chat, email and voice — against a scorecard, then routes the results into coaching. It is the former Klaus, an Estonian QA startup Zendesk bought in early 2024 and folded into its workforce-engagement line. The product page names four parts. AutoQA scores every interaction instead of a hand-picked sample, with out-of-the-box categories such as Empathy, Solution and Tone plus custom prompt-based categories. AI Agent QA (called BotQA in some third-party guides) holds AI-agent conversations to the same scorecard as human agents and flags abandoned or unresolved bot tickets. Spotlight flags churn risk, escalations, stuck conversation loops, dead air on calls and knowledge gaps. Coaching turns low scores into 1:1 sessions and scheduled training.

It is an add-on to any Zendesk Support or Suite plan, not a separate company’s contract. Zendesk lists Audacy, Kahoot!, Liberty and CityGo as customers and claims up to 80% less review time — a vendor figure, not an audit.

Why it shows up in support stacks

• It is already on the invoice. For a Zendesk shop, QA is a line on the existing order form, with users imported from the help desk and no second security review. That is the main reason Level AI, Solidroad and Intryc all have to argue against it.
• It grades bots and humans side by side. Zendesk’s help centre says it auto-detects Zendesk AI agents on messaging and voice; any other bot — Decagon, Sierra, a homegrown one — can be manually marked as an AI agent so it is scored on the AI-agent scorecard. AI Agent QA carries no extra charge inside the add-on.
• It is not Zendesk-only in practice. Klaus’s connectors survived the acquisition: Zendesk’s own docs cover connecting Intercom and Front, with up to 50 connections per account.
• Privacy controls are concrete. Customer names, emails and phone numbers can be masked, sensitive custom fields excluded by ID, and retention set from one month to one year. QA data does not count against your Zendesk storage limit.

Pricing reality

Zendesk’s pricing page lists the Workforce Engagement Bundle at $50 per agent per month, paid yearly — QA plus workforce management. It does not publish a QA-only price. Third-party 2026 pricing guides (Intercom, eesel, Richpanel) put the standalone QA add-on at about $35 per agent per month and workforce management at about $25, so the bundle saves roughly $10 a seat if you would buy both. Confirm the QA-only figure with your account executive.

Worked numbers: 20 agents on the QA add-on ≈ $8,400 a year; 50 agents on the bundle = $30,000 a year. Add that to Suite Professional at $115 per agent per month on annual billing, and a fully loaded 50-seat floor is about $99,000 a year before AI-agent resolution fees ($1.50 each committed, $2.00 pay-as-you-go). Against Level AI’s third-party band of roughly $80-125 per agent per month, Zendesk QA alone at about $35 costs 28-44% as much per seat, and the $50 bundle 40-63%.

A cheaper floor is coming from inside: at Relate in May 2026 Zendesk announced Quality Score, a lighter built-in grade on every human and AI conversation for Suite Professional and above. If a single score per conversation is all you need, check whether it covers you before adding the seat.

Best for

Support-operations and QA leads at Zendesk shops of roughly 15-200 agents, digital-first, who want every conversation scored and their AI agent graded on the same scorecard without a second vendor. It is strongest when the AI agent doing the resolving is a third party’s — then Zendesk is an independent judge of someone else’s bot.

Versus the alternatives

• Rippit (formerly MaestroQA) — the other volume leader. Pick it when a dedicated QA analyst team designs rubrics and calibration is the job, not a side task.
• Level AI or Observe.AI — pick these when voice is most of your volume and you run 100+ seats on a CCaaS platform; Zendesk QA’s voice coverage depends on Zendesk’s own voice layer.
• Solidroad — the fastest-growing entrant ($25M Series A, April 2026). Pick it when the help desk and the AI agent are both Zendesk: the vendor that sells the agent should not also issue its report card.
• Intryc — pick it for coaching and simulation workflows beyond what the bundle ships.

The pick for an undecided Zendesk shop: Zendesk QA. If none fit — say you are on a help desk with no connector — run a calibrated LLM rubric over a weekly transcript export for one queue for a month before buying anything.

Watch-outs

• The grader and the graded can be the same vendor. If Zendesk’s AI agents resolve your tickets, Zendesk is scoring its own product, and it bills those agents per resolution. Guard: double-score a monthly sample of 100 bot conversations by hand or in an independent tool, and treat a gap above 10 points as a reason to move scoring out.
• Out-of-the-box categories are generic. Empathy and Tone are not your policy. Guard: write custom prompt-based categories for your top three compliance rules, double-score 200 conversations by hand before retiring any manual review, and require 90% agreement per category to go live.
• Sync is not real time. Tickets sync every four to six hours. Guard: do not use AutoQA as a live escalation trigger; route urgent risk through help-desk triggers and keep QA for retrospective coaching.
• Deleting a connection is irreversible. It removes conversations, fields and tags, though review data stays in dashboards. Guard: ask Zendesk support to deactivate a connection instead of deleting it, and restrict connection management to one admin.
• Features get retired. Zendesk ended legacy Klaus surveys and the QA conversations Insights view on 30 January 2026. Guard: export dashboards and scorecards quarterly, and put a feature-deprecation notice clause in the order form.