A call comes through QA with a strong score. The agent greeted the customer warmly, read every disclosure, and promised a refund by Friday. The transcript checks every box. But the transcript only shows what the agent said. It can't show whether they pulled the right account, ran the verification, or actually submitted the refund they promised. That's the other half of every call, and audio-only QA has never been able to see it.
Video QA is how you score that half. It looks at what the agent did on screen alongside what they said on the call. Run the two together, and for the first time you're grading the whole interaction instead of half of it.
What is video QA?
Video QA evaluates what an agent did on screen (the systems they opened, the steps they ran, the actions they took) alongside what they said on the call. It analyzes the screen recording a contact center already captures and scores on-screen behavior against the same scorecard used for the conversation. Where audio QA reads the talk track, video QA reads the work. This screen-aware layer is what turns a transcript score into a complete one.
The half of the call a transcript can't see
A transcript can prove an agent said "let me verify your identity." It cannot prove they ran the verification. Picture a call that sounds flawless end to end: warm open, clean disclosures, confident close. On the transcript it scores well. On the screen, the agent never opened the record, never checked the second factor, never logged the action they described. The words were right. The work did not happen.
That's the gap every audio-only program lives with: said versus did. QA scorecards often ask about on-screen work, like whether the verification was run or the action was logged, but a transcript has no way to answer those questions. A transcript tool can confirm the words. It can't confirm the rest. Video QA closes that gap.
What QA can finally see
Pairing the screen with the conversation puts the parts of the scorecard that used to be unscorable into view:
- Required steps, actually performed: verification, disclosures, and consent run in the system, not just spoken (and correctly waived when the contact self-served or was transferred).
- The right action, in the right order: the correct record selected, the valid option chosen, the transaction logged where it belongs.
- Follow-through: whether a promised callback, refund, or submission was genuinely completed, rather than committed to on a call that "sounded fine" and then left open.
- Navigation and pacing: where an agent detoured, landed on the wrong screen, got stuck, or rushed, read from the path they took instead of inferred from their words.
- Multi-leg journeys as one interaction: an inbound, a consult, and a follow-up stitched into a single scored interaction, instead of each leg graded in isolation as if it were the whole call.
- Back-office work: the no-customer-on-the-line tasks that call-only QA never saw at all.
The point is coverage of the entire scorecard, not a sampled slice of it. Audio tells you how the call sounded. Screen tells you whether the job got done.
From clicks to coaching
The screen recording isn't there to catch people. It shows the route. You can see the exact path an agent took through the call: where they clicked into the wrong tab, where they stalled, where they went hunting for something that should have been one click away. That turns a vague "needs improvement" into a specific coaching point you can deliver right after the call. "When a customer asks about a refund, open the order history first, then the refund form. You went to billing, and that's where the call stalled."
The same view works across the whole floor. If one agent stalls on a screen, that's a coaching conversation. If every agent stalls on the same screen, it's not an agent problem. It's a sign the process, the system, or the training needs to change, and now you know exactly where to look.
Closing the loop: the flagged call becomes the practice
Knowing where an agent went wrong is half the fix. The other half is giving them a safe place to get it right. In Reddy, when video QA flags a workflow that broke down, that exact flow becomes a practice simulation inside a digital twin of the agent's real systems, built from the screen recording itself.
The agent runs the flow again until it's right. In guided mode, Reddy only allows the correct next step, so they learn the path without getting lost. In open practice, a wrong click is allowed, caught, and coached, because the wrong turn is what shows you what they misunderstood. Reddy scores the practice the same way it scored the live call.
That's the loop Reddy closes. QA finds the gap, practice fixes it in a replica of the real job, and the next live call shows whether it stuck. QA and training run in one platform instead of two separate tools.
Harte Hanks runs this loop with Reddy: automated QA ended manual call scrubbing, agents rehearse their workflows in custom simulations, and QA scores rose 6%. (Read the Harte Hanks story)
What your QA team gets back
Today, a QA reviewer spends most of the day listening. They pull a handful of calls, play them back, pause, rewind, and try to work out from the audio whether the agent actually opened the right account or just said they did. A lot of the job is detective work, and the calls they never get to stay a mystery.
With video QA, that question is answered before a reviewer opens the call. The screen shows what happened, and the hours spent guessing come back. That time goes into the work that makes the floor better: coaching an agent through the step they missed, spotting a pattern across a team, tuning the scorecard with supervisors, and taking what they find back to training and operations so the same mistake stops showing up.
