Why it matters

If you've ever looked at a score and thought "I'm not sure I'd stake someone's bonus on that", you're not alone. It's tempting to wait until every automatic score is perfect before you use any of them. The trouble is that day never comes, and in the meantime your scores sit unused.

Here's the good news: you don't need every automatic score to be perfect. You need the scores you act on to be right. Voyc scores every call against your Scorecard, then your team checks the calls that matter: every call that falls below your threshold, plus a few high scorers to keep Voyc honest. You tag the calls you've checked, coach on them and report on them, and that's a set of scores you can stand behind, all the way to bonus time.

This is especially useful if:

your team is spending hours re-checking calls that already scored well
you'd like to use QA scores for performance reviews or remuneration
you're rolling out a new or updated Scorecard and want to build confidence in it week by week

Here's how it works

When	What you do	Who
Every call	Voyc scores the call against your active Scorecard	Voyc
Each review session	Grade every call below your threshold and tag it Reviewed	QA
Same session	Spot check a few calls above your threshold and tag them Reviewed too	QA
As you go	Leave feedback and escalate calls that need action	QA and Team Leads
Weekly and monthly	Report on your Reviewed calls, then use the monthly report for bonuses	QA and Managers

Before You Start

A little set-up now saves a lot of rework later.

Agree on your threshold. This is the score below which every call gets a person's check. Ambers and reds is a good place to start. Remember that score and compliance are separate: a call can score well and still fail compliance, so every compliance fail gets checked too, whatever its score.
Make sure only the right calls are scored. In your Channel's Configurations, go to Audio and set a Minimum Conversation Length so very short calls, like voicemails and drop-offs, aren't processed at all. Fewer irrelevant calls means fewer misleading scores to check.
Pick one tag for checked calls. We use Reviewed. This tag is what your reports filter on later, so everyone on the team needs to use the same one.
Set your review rhythm. Pick a regular slot, for example once a week per agent or team, and do the whole review in one sitting: grade the ambers and reds, spot check a few greens, then report. You'll be surprised how quickly it goes: reviewing one agent's calls for a full week, spot checks and report included, took us about 20 minutes.

📝 Note: Scores tend to settle over the first few weeks as you fine-tune your Scorecard. That's normal, and it's exactly what this process is designed to support.

Step 1: Let Voyc Score Every Call

Once your Scorecard is active on a Channel, every call that comes in is scored automatically. You don't need to do anything here, which is the point: your team's time goes into the calls that need a person, not the ones that don't.

Step 2: Grade Every Call Below Your Threshold

These are the calls that could affect a customer, a compliance outcome or an agent's pay, so every one gets checked.

Go to your Channel's Conversations list.
Filter Call Date to the period you're reviewing (for example Last 7 days). If you have calls that don't need checking, like a particular call type or outcome, add an Attribute or any other filter to leave them out. You can also filter by Agent to grade one agent's calls, or leave it to grade all agents for that time frame.
Sort by Overall Score so your ambers and reds are grouped together. You can save this view to open it in one click next time.
Open the first call below your threshold and start the grading task in the Workflow section.
Work through each Group. Open a Statement to see why Voyc scored it the way it did. To fail it or mark it Not Applicable, select Fail or N/A, add a short note explaining why and select Save Note. To pass it, find the phrase in the transcript, highlight it and select Link to phrase. A pass needs evidence, so linking the phrase is how you pass a Statement (selecting Pass on its own will just remind you to link it).
Select Review Scores, then Complete Grading so the corrected score is saved.
In Tags, add your Reviewed tag.

Repeat for every call below your threshold. Then add a Compliance filter set to fail and check any calls that failed compliance but scored above your threshold.

Once you're done, your Conversations list shows at a glance which calls have been checked.

💡 Tip: Your notes double up as coaching feedback for the agent, so write them as if the agent will read them, for example "You could ask for a meeting here, before the customer ends the call." If you have Agent Highlights, you can send your comments directly to the agent for seamless inline coaching.

❗ Important: Always select Complete Grading. If you don't, the corrected score won't be saved and the task stays Pending.

Step 3: Spot Check A Few High Scorers

Checking a small sample of calls that scored well tells you whether Voyc is being too generous. If your spot checks keep agreeing with Voyc, that's your evidence that the unchecked high scores are consistent.

On the same filtered list, pick a few calls above your threshold, across different days.
Grade them the same way as Step 2 and add the Reviewed tag.

These few spot checks are a quick sense check for this review. To confirm how well all your other scores line up with your team's, run the monthly spot check in Keep Fine-Tuning As You Go.

If a spot check finds a call that should have scored lower or failed compliance, don't just fix that call. Note why it was missed. It usually points to a Statement that needs adjusting in your Scorecard (see Keep Fine-Tuning As You Go below).

💡 Tip: If a call went down the wrong route on your Scorecard, use the thumbs down on Did the conversation follow the expected scorecard route? in the Grading Panel. It helps us improve routing.

Step 4: Give Feedback And Escalate

Once a call is graded, decide whether it needs anything else.

No action needed: you're done. The corrected score feeds straight into your Dashboard and reports.
Feedback for the agent: add it in your Workflow's feedback task or in an Agent Highlight.
Needs coaching or a customer callback: assign the call to the right Team Lead or Manager using the assign icon next to the task, with a note explaining why.

We suggest assigning every call that still fails compliance or scores poorly after grading to the agent's Team Lead.

📝 Note: The person you assign to must be a member of the Channel.

Step 5: Report On Your Reviewed Calls

This is where the process really earns its keep. Your checked calls now tell you exactly where your team and each agent need work, and give you a report you can stand behind for coaching or bonuses.

See where the team needs work

Go to Dashboard and set your Date range and Channel.
Select + Add filter, then Grading Status: Graded.
Open the Scores By Channel tab and select View next to your Channel.

Under Scorecard Scores, every Scorecard Group has its own Group Score, so you can see at a glance which parts of the call your team nails and which ones need work. Open a Group to see the adherence for each Statement, then select the eye icon next to Mentioned or Not Mentioned to open the conversations behind the number. For example, a Data Protection Group at 33% tells you the reminder is being missed on most calls, so that's your coaching focus for the week.

To learn more, read How To View Scores By Channel.

Drill into a specific agent

Open the Agent Insights tab on the Dashboard, keep the Grading Status: Graded filter on and choose the agent and Channel. You'll see where that agent is strong, where they need to improve and the exact conversations behind each score, which is everything you need for a focused coaching session.

To learn more, read How To Use The Agent Insights Dashboard.

Generate the report for coaching or bonuses

When it's time to share results with a Team Lead, or with whoever approves bonuses, generate an Agent Overview report. Filtering it by your Reviewed tag (and date range) means it only includes calls from that period that a person has checked.

On the Channel's scores page, select Generate Report (or go to Reports and select Generate).
Choose Agent Overview, then select your Channel, Agent(s) and Reporting Period.
Under Filter By Tags, choose Reviewed.
Check the summary and select Generate Report. You'll find it in the Reports Library when it's ready.

Weekly: run it for the past week so issues are spotted early and coached before they snowball.

Monthly: run it for the full month and use it for bonus calculations.

💡 Tip: Give yourself a buffer between the end of the month and the bonus decision, for example a week or two, so any late grading or disputes can be cleared first.

Keep Fine-Tuning As You Go

The calls you grade are also your best source of Scorecard improvements. If the same Statement keeps getting corrected, that's a sign it needs adjusting: open your Scorecard in Configurations and tweak its criteria or keywords. Each round of changes narrows the gap between Voyc's score and your team's, so scores become steadier and you grade fewer calls over time.

Confirm your scores with a monthly spot check

Your weekly review covers the calls you graded. Once a month, a quick random spot check measures the accuracy of the rest of your scores against your team's judgement. It shows, with 95% confidence, the lowest accuracy Voyc could have, and where fine-tuning will help most. To be 100% confident, you'd need to review every call.

The idea: a spoonful of soup

To check soup, you taste a spoonful, not the whole pot. A spoonful tells you about a small pot and a huge pot equally well. That's why the number of calls you need barely grows with volume: about 70 calls whether you have 1,000 calls a month or 5,000.

The catch is luck. Check 70 calls and find 2 that needed changing, and Voyc's accuracy looks like 97%. Check a different 70 and you might find 1, or 5. So rather than take 97% at face value, you work out the lowest accuracy you can claim with 95% confidence, and make sure that still meets your accuracy target.

Calls checked	Calls changed	What you saw	Lowest accuracy at 95% confidence	Meets 90% accuracy?
29	0	100%	90%	Just
70	2	97%	91%	Yes
70	3	96%	89%	Not yet

95% confidence means there's a 5% chance the real accuracy is lower than this number.

How to run the check

Go to your Channel's Conversations list, filter Call Date to last month and sort by Call Date, not by score.
Pick 70 calls spread evenly through the list, for example every 10th call if you had 700. Don't look at the score before you choose: the check only works if the calls are random. If you had fewer than 70 calls, check them all.
Grade each one the same way as Step 2 and add your Reviewed tag.
Count how many calls needed their score changed when you graded them. A call counts as changed if you changed any Statement on it, so this is a strict test.

What your result means

Calls changed (out of 70)	What it means	What to do
0 to 2	With 95% confidence, Voyc's scoring accuracy is at least 90%	Keep going and check again next month
3 or more	You can't yet confirm 90% scoring accuracy. Your Scorecard has room to improve	Look at which Statements you changed, fine-tune them and check again next month

How many calls to check

For a 90% accuracy target at 95% confidence, check at least this many calls:

Calls changed allowed	Calls to check
0	29
1	46
2	61
3	76
4	89

It works the other way too. For example, if you check 100 calls, up to 4 changed calls still confirms 90% accuracy at 95% confidence.

Why 70? You don't know up front how many changes you'll find, so you need a starting number. If your team already agrees with Voyc's scores most of the time, 70 calls gives you room for 2 changed calls.

For the curious

There are two formulas. One plans how many calls to check. The other tells you if Voyc's accuracy meets your target, with 95% confidence.

1. How many calls to check (Cochran's formula, 95% confidence)

Calls needed = 3.84 × p × (1 − p) ÷ gap²

p is the accuracy you expect from Voyc's scores, for example 0.95 (95% accuracy)
gap is how far that accuracy is above your accuracy target, for example 0.95 − 0.90 = 0.05
3.84 sets 95% confidence and never changes

So 3.84 × 0.95 × (1 − 0.95) ÷ 0.05² ≈ 73 calls. That's where 70 comes from. If you have fewer calls in the month, you need slightly fewer:

Calls that month	Calls needed
5,000	72
1,000	69

This is a planning number, so it is a little cautious. The exact test below shows 61 calls is enough if you find 2 or fewer changed calls.

2. Have you confirmed your accuracy? (Clopper-Pearson method, one-sided)

Accuracy = BETA.INV(0.05, calls right, calls changed + 1)

calls right is how many calls kept Voyc's score
calls changed + 1 is how many calls needed changing, plus 1
0.05 sets 95% confidence

This gives the lowest accuracy you can claim with 95% confidence. We only care how low the accuracy could be, not how high, so we use the one-sided version. For 70 calls with 2 changed, =BETA.INV(0.05, 68, 3) gives 91% accuracy, which meets the 90% accuracy target.

The big lesson: the closer Voyc's accuracy is to your accuracy target, the more calls you need to confirm it. Halve the gap and you need four times as many calls. If Voyc's accuracy is 90% or lower, checking more calls won't help. Fine-tune the Scorecard first, then check again.

📝 Note: This check is about how well Voyc scores calls against your Scorecard, not compliance. Every compliance fail is already checked by a person in Step 2.

FAQ

Do I need to check every call?

No. You check every call below your threshold, every call that fails compliance and a sample of the rest. Your spot checks are what give you confidence in the calls you don't check.

How many calls do I need to check?

Every call below your threshold, every compliance fail and a few high scorers in each review. Once a month, add a random spot check of 70 calls. If 2 or fewer need changing, you can say with 95% confidence that Voyc's scoring accuracy is at least 90%. See Confirm your scores with a monthly spot check above.

What's the difference between my weekly review and the monthly spot check?

Your weekly review is about people: it checks the calls behind an agent's or Channel's scores so you can coach and report on them. The monthly spot check is about your Scorecard: a random sample across all your calls that measures the accuracy of the scores nobody checked and shows whether your fine-tuning is working. Weekly reviews pick calls on purpose, mostly low scorers, so they can't tell you how good the unchecked scores are. Only a random check can.

Shouldn't the automatic scores just be right?

Voyc gets most calls right, but no automatic (or human) scoring is perfect, just as two people grading the same call won't always agree. Mistranscribed words, callbacks and calls that don't follow the usual journey can all trip it up. This process means those calls are caught by a person before a score is used for anything that matters.

Why tag calls as well as grading them?

The tag is what lets you pull a report on those tagged calls only. It also shows at a glance on the Conversations list which calls your team has already looked at.

What threshold should I use?

Start with ambers and reds. If your team is grading more calls than they can manage, look at why so many are scoring low first. It's often a Statement worth adjusting rather than a threshold worth moving.

Can I really use these scores for bonuses?

That's your call, and your bonus policy should say how scores are used. What this process gives you is a set of scores that a person has checked, with a clear record of who graded each call and why.

The Gist

You don't need perfect automatic scores to use them with confidence. Let Voyc score every call, grade everything below your threshold, spot check a few above it, tag what you've checked and report on those calls. Do that consistently and you'll have scores you can coach on, report on and, if you choose, pay bonuses on.

How To Get QA Scores You Can Trust Enough To Pay Bonuses On