Dead Air, Talk Ratio and Other Call Metrics That Predict a Bad Call

The call metrics that best predict a bad call are the ones you can measure from the audio itself: unexplained dead air, a talk ratio far outside the normal range for that call type, slow responses after the customer stops talking, frequent interruptions, a pace that does not match the customer's, and a slow pickup after the customer says hello. None of them proves a call went badly on its own. Together, they point QA to the calls worth hearing first.

This guide defines each metric, shows how to calculate it, and explains how to set targets from your own calls instead of borrowed benchmarks.

Why measured call metrics matter

Most QA scorecards judge delivery by ear: "good pace," "no awkward silences," "let the customer talk." Two reviewers will hear the same call differently. Speech metrics are measurements taken from the timing of words in a recording, such as who spoke when and for how long. They give the same answer every time, they can be taken on every call, and they are good at spotting calls where something went wrong even when the words on the transcript look fine.

Use them as signals, not verdicts. A long silence might be a broken CRM, not a careless agent. The value is in spotting patterns across many calls and pulling the right ones for review.

What the research says about how a call feels

None of the studies below measured dead air or talk ratio directly. They do show that the moments these metrics catch, waiting, being talked over and not being helped, matter to customers and to revenue:

• Speed and friendliness count. PwC's Experience is everything research (2018) found speed, convenience, helpful employees and friendly service each rated above 70% in importance to consumers, and that 32% would walk away from a brand they love after just one bad experience.
• Customers expect someone right away. In Salesforce's 2023 State of the Connected Customer, 77% of customers said they expect to interact with someone immediately when they contact a company. A long pickup delay or an unexplained silence runs straight into that expectation.
• Ease is slipping. Forrester's 2024 US Customer Experience Index, based on more than 98,000 US customers, found customer experience quality at an all-time low, with average ease down to 66% and declines in effectiveness, ease and emotion.
• Good experiences pay. In a Qualtrics XM Institute study of 33,093 consumers published in 2023, people who rated a company's experience 5 stars were 2.3 times as likely to purchase more as those who rated it 1 or 2 stars.

Dead air

Dead air is silence on a live call where neither side is speaking and the customer has not been told why. It is different from a hold the agent announced.

How to measure dead air

1. Take the word level timestamps from the transcript.
2. Find gaps between the end of one person's speech and the start of the next of at least your threshold, for example 5 seconds.
3. Exclude gaps the agent explained right before, such as "give me one moment while I pull up your account."
4. Decide who owns the gap. A customer thinking before they answer is not the agent's dead air.

Count of unexplained gaps per call, and total dead air seconds per call, are the two numbers worth tracking.

What dead air predicts

Unexplained silence makes customers feel ignored and often signals a deeper problem: the agent does not know the answer, cannot find it in the system, or is doing something else. On sales calls it breaks momentum. On support calls, check whether calls with the most dead air are also the ones that end unresolved. On most floors it is worth looking.

How to fix it

• Teach agents to narrate: "I am checking that now, it will take about thirty seconds."
• Look at where dead air happens in the call. If it clusters at the same step, the issue is usually a tool or a process, not the agent.

Talk ratio

Talk ratio, also called talk to listen ratio or talk share, is the share of the conversation the agent spends speaking.

Talk ratio = agent talk time divided by (agent talk time plus customer talk time), times 100.

Worked example with hypothetical numbers: on a 10 minute call the agent spoke for 6 minutes and the customer for 3 minutes, with 1 minute of silence. Talk ratio = 6 divided by 9, times 100, or about 67 percent.

What talk ratio predicts

A very high agent talk ratio on a sales or discovery call often means the agent pitched without asking questions or did not let the customer explain. A very low talk ratio can mean the agent lost control of the call or could not answer. The right range depends on the call type: a tech support walkthrough and a cold call should not have the same target.

How to set a target

Pull the talk ratio on calls that ended well (booked, resolved, paid) and calls that did not, for the same campaign. Set the target range where the good calls cluster. Recheck it every quarter.

Response time

Response time, sometimes called reaction time, is how long the agent takes to start speaking after the customer finishes. Measure it on every agent turn and take the median per call so one slow moment does not dominate.

Slow responses over a whole call suggest the agent is unsure, distracted or searching for information. They also make customers repeat themselves and raise frustration.

Interruptions

An interruption is when the agent starts talking while the customer is still speaking, or cuts in mid sentence. Measure it from overlapping speech and normalize long calls, for example interruptions per 10 minutes.

Frequent interruptions are a strong sign a call is going badly. Customers who keep getting cut off feel unheard, and those calls are worth hearing before they turn into an escalation. Interruptions also show up in agents who are rushing to hit handle time targets.

Pace match

Pace is speaking speed in words per minute. Pace match compares the agent's pace with the customer's.

Pace difference = (agent words per minute minus customer words per minute) divided by customer words per minute, times 100.

An agent talking much faster than a customer, especially an older customer or someone hearing complex terms, loses them. That matters a lot on Medicare, insurance and collections calls where disclosures must be understood, not just read.

Pickup delay

Pickup delay is the time between the customer's first "hello" and the agent's first words. On outbound floors, a long gap here is a common reason people hang up in the first seconds of a call.

Pickup delay is not always the agent's fault. Predictive dialers can take time to connect a live answer to an available agent. Compare pickup delay across agents on the same campaign: if everyone is slow, look at the dialer settings; if one agent is slow, look at the agent.

Comparison: call metrics that predict a bad call

Metrics that mislead on their own

• Average handle time (AHT). Short calls can be great or rushed. Long calls can be thorough or stuck. Read it next to the scorecard, not instead of it.
• Hold time without context. A hold the agent announced and kept short is good service. Unexplained silence is the problem.
• Average score alone. A floor average hides the few calls with critical failures. Track auto-fails separately.

How to use these metrics in QA

1. Add them to the scorecard as a small delivery section with modest weights, as in our QA scorecard template. They should inform the score, not dominate it.
2. Set thresholds per campaign from your own good calls.
3. Use them to rank the review queue. A call with several dead air events, a high interruption count and a failed critical item should be heard before a clean call.
4. Coach with the moment, not the number. "At 3:12 there were 14 seconds of silence after the customer asked about the fee" beats "your dead air is high."
5. Separate agent problems from system problems. If a metric is bad for the whole team at the same step, fix the process.

Support and sales teams use these differently. See our pages on customer support and solar and home services for examples of what each floor flags.

How Sillana helps

Sillana measures these metrics from the word timings of every recorded call: response time, dead air (silences of 5 seconds or more that the agent did not explain, with customer pauses not counted against the agent), talk ratio against a target range you set, interruptions, pace compared with the customer, and, optionally, pickup delay after the first hello. They appear as delivery criteria on your scorecard next to the items the AI judges, so you can sort and review calls by them. Defaults are a starting point, and you can adjust ranges per scorecard. Plans are on the pricing page, and you can compare the approach with manual QA.

Key takeaways

• Dead air, talk ratio, response time, interruptions, pace match and pickup delay can be measured from the audio on every call.
• No single metric proves a bad call. Combined, they are strong signals for what to review first.
• Set targets from your own good calls per campaign, not from generic benchmarks.
• Separate agent behavior from dialer and system problems before coaching.
• Coach with the specific moment in the call, not just the number.

Questions people ask