A new call source begins with a practical problem: the buyer has no buyer-specific history.

The source may have performed well elsewhere. The publisher may have historical numbers. The operator may understand the traffic path, vertical, expected volume, and likely fit. Sample calls, landing pages, creatives, scripts, and source documentation may all look reasonable.

But none of that tells the buyer exactly how the source will perform with its campaigns, agents, qualification rules, schedules, routing setup, and conversion process.

That uncertainty is normal.

A source benchmark can help by giving the buyer a reasonable starting frame. It can describe what the source has historically looked like, which metrics matter, how the figures were produced, and what kind of controlled test makes sense.

A benchmark should not tell the buyer, “This source is proven for you.”

It should tell the buyer:

Here is the evidence available before your own history exists. Here is how much confidence that evidence deserves. Here is the test needed to replace general expectations with buyer-specific results.

That distinction matters.

Used well, benchmarks can help a buyer decide whether to consider a source, where to route it, how tightly to cap it, which outcomes to monitor, and when to review the test.

Used poorly, benchmarks become sales claims dressed up as data.

This guide explains how serious buyers should read pay-per-call source benchmarks, what a useful benchmark should include, which limitations need to stay visible, and how benchmark data should give way to the buyer’s own performance history.

A source benchmark is a starting point, not a verdict

A benchmark is a reference point.

It may summarize a source’s historical performance, expected operating range, or typical characteristics. It can help a buyer compare an unfamiliar source with the requirements of a campaign before the buyer has enough calls to form its own view.

A useful benchmark might help answer questions such as:

Does the source generally produce enough volume to support a test?
Is the traffic type consistent with the buyer’s call-handling model?
What billable talk-time pattern has the source shown?
What conversion definition supports the reported conversion rate?
How often do accepted pings become real calls?
How current are the figures?
Are the numbers self-reported or computed from live records?
How many calls support the live figures?
Is the benchmark specific to the same vertical, traffic type, and campaign context?
What should the buyer watch during the first test?

Those answers can reduce uncertainty.

They cannot eliminate it.

A benchmark does not prove:

The buyer will receive the same call mix.
The buyer’s agents will handle the calls well.
The source will convert at the same rate.
The buyer’s qualification rule matches the historical rule.
The source will remain stable after volume increases.
Every call will be compliant, qualified, billable, or payable.
The buyer’s economics will work.
The source is ready for unrestricted volume.

A benchmark is most valuable when everyone understands what it does not prove.

Why new sources face a buyer-confidence gap

Established sources can build buyer-specific history.

A buyer may already know:

How many calls it received.
When those calls arrived.
How often agents answered.
Average billable talk time.
Which calls met the qualification rule.
Which calls converted.
How disputes were resolved.
Whether the invoice reconciled to the call records.
Whether performance changed as caps increased.
How the source compared across campaigns and targets.

A new source has none of that history with the buyer.

The buyer therefore has to make an initial decision with incomplete information.

That decision is harder when the source is represented only by broad assurances:

“The traffic performs.”
“These calls convert.”
“The source has quality.”
“Other buyers like it.”
“We can send a lot of volume.”
“The average call is long.”
“This is direct traffic.”
“The leads are exclusive.”
“The source is ready to scale.”

Those statements may be sincere. They are not enough to configure a campaign.

A buyer needs a more disciplined way to translate unfamiliar supply into an operating decision.

That is where a benchmark can help.

It gives the buyer a structured reference before buyer-specific data exists. It also gives the publisher and operator a clearer way to discuss the source without pretending that prior results guarantee future outcomes.

For the source to route at all, the buyer still needs the controls described in source enablement: the operator offers a reviewed source, the buyer enables it in the appropriate scope, and the live routing path applies the remaining eligibility rules.

The benchmark supports that decision. It does not replace it.

The three evidence tiers buyers should keep separate

One of the most important benchmark rules is to separate different kinds of evidence.

A buyer should not see one number labeled “performance” when the underlying figures come from different sources, date ranges, definitions, or levels of verification.

A practical model uses three tiers.

1. Self-reported source figures

These are figures supplied during source onboarding.

Examples may include:

Expected daily volume.
Historical average talk time.
Historical conversion rate.
Historical ping-to-call ratio.
Available geographies.
Common delivery hours.
Traffic-type description.

Self-reported information is useful.

Publishers often know their traffic before the operator has enough live records to calculate stable metrics. Without a self-reported starting point, a new source profile may be blank.

The problem is not that the figures are self-reported.

The problem is when self-reported figures are displayed as though they were independently verified.

A buyer should see a clear label such as:

Self-reported

That label lets the buyer use the information appropriately. It is source-supplied context that can help design a test. It is not yet live proof within the current operation.

2. Live-computed source benchmarks

After a source has accumulated enough live activity, the operation may calculate aggregate source metrics from actual records.

These figures may include:

Average billable talk time.
Live call count.
Conversion rate under a defined outcome.
Ping-to-call ratio.
The date or window covered.
The number of calls supporting the figure.

Live-computed figures deserve more weight than an onboarding estimate because they come from actual operating records.

They still require context.

A global source benchmark can summarize the source across buyers or campaigns without exposing any buyer’s confidential breakdown. But a global figure can still hide important differences in buyer handling, qualification rules, geography, seasonality, campaign configuration, and target capacity.

A clear label might say:

Verified · live
Based on 240 calls
As of July 31, 2026

The exact wording can vary. The principle should not.

The buyer needs to know that the number came from live data, how much data supports it, and how fresh it is.

3. The buyer’s own performance

Buyer-specific history is the most relevant evidence for that buyer.

Once the buyer begins receiving the source, it can measure the actual interaction between:

This source.
This campaign.
This target.
This qualification rule.
This buyer’s agents.
This schedule.
This geography.
This conversion process.

That history should eventually take priority over the general benchmark.

The benchmark remains useful as a comparison point. But it should not override what the buyer’s own records show.

A buyer-facing source view should make the distinction obvious:

Evidence tier	What it tells the buyer	Appropriate use
Self-reported	What the source says it has historically produced or expects to produce	Decide whether the source deserves review and what to test
Live-computed benchmark	What the source has produced in the current operating environment at an aggregate level	Set an informed starting range and compare early behavior
Your performance	What the source has produced for this buyer, campaign, and target	Decide whether to scale, revise, pause, or disable

When these tiers are blended, confidence becomes false confidence.

When they are separated, the buyer can trust-weight the information.

What a useful source benchmark should include

A benchmark should be a small operating spec sheet, not a promotional scorecard.

The best fields are the ones that help the buyer make a routing, capacity, qualification, and test decision.

Provenance

The buyer should know how the figure was created.

Possible provenance labels include:

Self-reported.
Operator-published.
Live-computed.
Buyer-specific.
Modeled or estimated.

Avoid vague labels such as “verified” unless the verification method is clear.

A number without provenance asks the buyer to guess how much trust it deserves.

As-of date and measurement window

Performance changes.

A source may change:

Creative.
Landing page.
Media channel.
Transfer script.
Agent team.
Upstream supplier.
Geography.
Bid strategy.
Call screening.
Hours.
Volume.

A benchmark should show when it was calculated and, when practical, the window it covers.

“Conversion rate: 18%” is incomplete.

“Live-computed conversion rate for the most recent completed review window, as of August 1” is more useful.

The exact window should fit the source and review process. The important point is that the buyer can see whether the figure is current.

Sample count

The buyer should know how much live data supports the benchmark.

A rate based on a small number of calls can move sharply after only a few more outcomes. A rate based on a larger, stable sample may deserve more confidence, provided the source definition and operating conditions have remained consistent.

There is no universal call count that makes every source benchmark reliable.

The required sample depends on:

The metric being estimated.
Natural variability.
The size of the difference the buyer cares about.
The cost of being wrong.
The traffic mix.
Whether outcomes are delayed.
Whether calls are independent enough to compare.
How narrowly the source has been segmented.

Research on online experimentation reaches the same broad conclusion: sample-size planning is part of trustworthy decision-making, and an undersized sample can produce weak or misleading inference even when the analysis itself is correct. The paper All about sample-size calculations for A/B testing addresses online experiments rather than pay-per-call, but the operating lesson transfers: a decision threshold should not be treated as meaningful without considering how much data supports it.

The buyer does not need a statistics lecture in the catalog.

It does need the sample count.

Typical billable talk time

Talk time can help the buyer understand whether calls commonly remain connected long enough to pass a duration-based qualification rule.

But the metric needs an exact definition.

Ask:

Is this raw connected duration or billable transformed duration?
Are unanswered calls included?
Are abandoned calls included?
Are transfers timed from the consumer’s first leg or the buyer connection?
Is the mean being used, or the median?
Which qualification threshold applied?
Are duplicate or disputed calls included?

A number called “average call duration” can mean several different things.

For settlement and buyer evaluation, the operation should be clear about whether it is measuring connected duration, qualified duration, or billable duration.

The distinctions in routed, qualified, and billable calls should remain visible.

Typical daily volume and arrival pattern

Daily volume helps the buyer understand whether the source is large enough to test and whether the buyer has enough capacity to receive it.

A daily average alone can hide burstiness.

Twenty calls spread across ten hours are operationally different from twenty calls arriving within forty minutes.

Useful volume context may include:

Typical daily range.
Common delivery hours.
Day-of-week pattern.
Geographic concentration.
Peak arrival periods.
Whether volume is steady or bursty.
Whether the source can honor a cap.
Whether a source sends pings that may not become calls.

The purpose is not to promise volume.

It is to help the buyer set realistic caps, schedules, and concurrency. Those controls are explained further in how caps, schedules, and concurrency shape call flow.

Conversion rate with a named conversion

“Conversion rate” is meaningless until the conversion is defined.

Depending on the vertical and commercial model, conversion may mean:

A sold policy.
A completed enrollment.
A retained client.
A booked appointment.
A qualified intake.
A signed agreement.
A completed sale.
Another buyer-confirmed outcome.

The denominator also matters.

Is conversion rate calculated from:

All pings?
Routed calls?
Connected calls?
Qualified calls?
Billable calls?
Calls with a returned disposition?
Calls that reached an eligible agent?

Two sources can display the same conversion percentage while measuring different events.

A benchmark should name the numerator and denominator or link to the governing definition.

Ping-to-call ratio

In a ping-based workflow, not every ping becomes a real call.

The ping-to-call ratio can help the buyer and operator understand how often source opportunities turn into connected call activity.

This metric can expose operational differences such as:

A source sending many opportunities that rarely progress.
A source sending fewer pings with a high rate of real calls.
Bid or routing conditions that screen out most opportunities.
Source behavior that changes after bids are returned.
Campaign configuration that creates avoidable rejection.

Like every rate, it needs a defined numerator, denominator, and window.

It also should not be confused with conversion rate.

Ping-to-call describes movement into call activity.

Conversion describes a downstream buyer outcome.

Source scope

The benchmark must apply to a defined source.

If materially different traffic is blended, the benchmark becomes harder to trust.

A benchmark may be unreliable when one source label includes:

Consumer-initiated inbound calls and transfers.
Paid search and social traffic.
Direct publisher traffic and network traffic.
Several verticals.
Several landing pages with different consumer expectations.
Multiple upstream call centers.
Different geographies with different buyer eligibility.
New creatives mixed into an old source identity.

A source should be narrow enough that its history still describes the traffic the buyer is considering.

That is why source-level reporting matters. A strong source identity allows performance to accumulate without blending unrelated traffic into one number.

Definitions matter more than polished dashboards

A benchmark dashboard can look precise while the underlying definitions remain weak.

That creates a dangerous kind of confidence.

Suppose two source cards show:

Average talk time: 175 seconds.
Conversion rate: 14%.
Daily volume: 35 calls.

The cards appear comparable.

They may not be.

For Source A:

Talk time may mean buyer-connected duration.
Conversion may mean sold outcomes divided by billable calls.
Daily volume may mean routed calls.
The data may cover the last thirty days.

For Source B:

Talk time may include the transfer leg.
Conversion may mean appointments divided by connected calls.
Daily volume may mean all pings.
The data may come from an old onboarding form.

The numbers look identical because the labels are broad.

The operating reality is different because the definitions are different.

Before comparing sources, buyers should confirm:

The same event is being measured.
The same denominator is being used.
The time window is comparable.
The traffic type is comparable.
The vertical and qualification rule are comparable.
The source identity has not materially changed.
The provenance is visible.

Benchmark quality begins with definitions, not design.

Why there should not be one universal “good” benchmark

Buyers often want a simple answer:

What is a good conversion rate for this kind of call?

That question is understandable. It is usually too broad.

The same source can perform differently because of:

Vertical.
Geography.
Consumer intent.
Traffic type.
Qualification threshold.
Buyer price.
Publisher payout.
Product availability.
Licensing.
Seasonality.
Agent skill.
Answer speed.
Hold time.
Script.
Disposition discipline.
Call recording and QA process.
Duplicate policy.
Conversion window.
Campaign and target configuration.

A benchmark should therefore be contextual.

A final-expense transfer should not be compared casually with a consumer-initiated auto-insurance inbound.

A national call center should not assume a local home-services source will behave the same way across every service area.

A buyer with a strict qualification threshold should not compare its results with a source benchmark based on a looser rule.

A source benchmark is most useful when it narrows the comparison to the closest operational context available.

The buyer should ask:

Is this the same vertical?
Is this the same traffic type?
Is the source direct or aggregated?
Is the buyer receiving the same consumer journey?
Are the qualification and conversion definitions similar?
Are the delivery hours and geographies similar?
Is the source being evaluated at a similar volume level?

The answer does not have to be perfect.

It has to be honest.

Early results should not be promoted into proof too quickly

New source tests create pressure to reach a conclusion.

A buyer may see a good first day and want more volume.

A publisher may see a poor first day and argue that the sample is unfair.

An operator may want to show progress.

All three pressures can lead to premature declarations.

Small samples are noisy. Outcomes can also arrive late. Conversion feedback may be incomplete. A handful of strong or weak calls can dominate the average.

Another common problem is continuous peeking.

The team checks the dashboard after every few calls and stops the test as soon as the result supports the decision someone already wanted to make.

Sequential analysis research has shown why conventional fixed-sample inference can become unreliable when people repeatedly monitor results and choose when to stop. The paper Always Valid Inference: Bringing Sequential Analysis to A/B Testing concerns online A/B testing, not call buying, but the practical warning applies: decide how the test will be reviewed before the result starts influencing the stopping rule.

A pay-per-call source test does not need to become an academic experiment.

It should still define:

The initial cap.
The review window.
The minimum evidence expected before a scale decision.
Which outcomes may arrive after the call.
Which failure conditions justify an immediate pause.
Which metrics are directional rather than conclusive.
Who can approve a cap change.
How changes in buyer handling will be documented.

This keeps the team from turning every early fluctuation into a new policy.

Buyer handling can make a good source look bad

Source benchmarks describe source behavior.

Buyer-specific performance describes the source and buyer operating together.

That difference is critical.

A source may underperform because:

The destination was closed.
Calls waited too long.
Agents did not answer.
The wrong team received the call.
Agents were not trained for the vertical.
The script did not match the consumer’s expectation.
The buyer rejected geographies that were accidentally left open.
The buyer returned dispositions late or inconsistently.
The target reached concurrency.
The source routed during understaffed hours.
The conversion window had not completed.
Duplicate rules differed from the source’s historical environment.

A buyer should not excuse weak traffic by blaming itself for everything.

It also should not treat every poor outcome as a source defect.

The test needs handling metrics and routing records alongside source metrics.

Before a buyer decides that a benchmark was wrong, it should ask:

Did eligible calls reach the intended target?
Was the target open and staffed?
What was answer speed?
How many calls connected?
Which calls failed before an agent conversation?
Were dispositions returned consistently?
Did the buyer apply the agreed qualification rule?
Were calls judged before delayed conversions arrived?
Did source volume arrive in the expected pattern?
Did the buyer change campaign settings during the test?

A benchmark gives the buyer a reference.

The operating record explains the difference.

How to turn a benchmark into a controlled source test

A good benchmark should lead to a better test design.

It should not lead directly to unrestricted traffic.

Step 1: Confirm source fit

Review:

Vertical.
Traffic type.
Consumer journey.
Supply type.
Geography.
Delivery hours.
Expected volume.
Available decision materials.
Benchmark provenance.
Benchmark sample count.
Known limitations.

The first question is not, “Is the number high?”

It is, “Does this source fit this buyer and campaign?”

Step 2: Choose the campaign and target deliberately

The buyer should decide where the source belongs.

A source may fit one campaign but not another.

It may fit the primary target but not an overflow team.

It may fit a licensed geography but not the buyer’s full footprint.

The source should be enabled only where the current evidence supports it.

Step 3: Set a controlled cap and schedule

Use the benchmark to estimate a safe starting range.

Consider:

Agent capacity.
Source arrival pattern.
Buyer concurrency.
Test budget.
Qualification cost.
Risk of delayed outcomes.
The number of calls needed for a useful review.
The consequences of a poor fit.

A small test should still be large enough to learn from.

A large test should still be small enough to stop without creating an avoidable financial or operational problem.

Step 4: Lock definitions before the test

Document:

Routed.
Connected.
Qualified.
Billable.
Payable.
Converted.
Duplicate.
Disputed.
Adjusted.

Also document:

The talk-time definition.
The conversion event.
The conversion denominator.
The disposition process.
The dispute window.
The review date.
The conditions for an immediate pause.

This prevents the parties from redefining success after seeing the results.

Step 5: Compare the benchmark with buyer-specific results

The buyer should not ask only whether it “beat the benchmark.”

It should examine the pattern:

Was volume within the expected range?
Did calls arrive at expected times?
Was billable talk time materially different?
Was conversion higher or lower?
Did answer speed or agent availability explain the difference?
Were disputes concentrated around one rule?
Did the traffic mix change during the test?
Did the result improve as agents learned the source?
Was the benchmark based on a genuinely comparable context?

The difference between the benchmark and buyer result is information.

It is not automatically a failure.

Step 6: Scale, revise, pause, or disable

A source test should end with a decision.

Possible outcomes include:

Increase the cap.
Expand hours.
Add another geography.
Enable another target.
Keep the source limited.
Revise qualification rules.
Improve buyer handling.
Request revised source materials.
Separate a blended sub-source.
Pause for more review.
Disable the source.
Withdraw the source from the buyer’s catalog.

For a broader framework, see how to evaluate a pay-per-call source before scaling it.

A hypothetical benchmark-to-test example

Consider a hypothetical home-services buyer evaluating a new consumer-initiated plumbing source.

The source profile includes:

A buyer-facing pseudonym.
Consumer-initiated inbound traffic.
Direct-publisher supply type.
Plumbing vertical.
Selected service areas.
Self-reported expected volume.
A live-computed aggregate talk-time benchmark.
A live sample count and as-of date.
Landing-page and creative examples.
A note that the source tends to deliver most calls during weekday mornings.

The buyer has one experienced plumbing target and one general overflow target.

The buyer does not enable the source everywhere.

It:

Enables the source for the plumbing campaign.
Enables it only on the experienced target.
Uses a limited weekday-morning schedule.
Sets a modest daily cap.
Confirms accepted service areas.
Defines the billable threshold and conversion event.
Schedules a review after enough calls and outcomes have accumulated.
Monitors answer speed, connected calls, billable talk time, conversions, and disputes separately.

Early results show talk time near the benchmark but lower conversion.

The buyer does not immediately declare the source weak.

The call records show that several calls reached agents who were unfamiliar with the buyer’s plumbing intake flow. The buyer corrects the queue assignment and continues the test under the original cap.

Later results improve.

This does not prove the source will always perform.

It shows why the benchmark and the operating record belong together.

The benchmark gave the buyer a reasonable expectation.

The buyer-specific history explained where the first result differed.

The controlled cap kept the learning process manageable.

Benchmark red flags buyers should notice

A source benchmark deserves less weight when it has one or more of these problems.

No provenance

The buyer cannot tell whether the number is self-reported, operator-entered, live-computed, or buyer-specific.

No as-of date

Old performance is presented as current.

No sample count

A rate based on a handful of calls looks as mature as a rate based on sustained activity.

No metric definition

“Conversion,” “qualified,” or “duration” is displayed without defining the event.

Blended traffic

Several traffic paths share one source label and one benchmark.

Cross-vertical comparison

A broad network average is used to describe a specific campaign.

Cherry-picked windows

Only the strongest period is shown, or the window changes without explanation.

Benchmark used as a guarantee

Historical or aggregate performance is presented as the result the buyer should expect.

Price and performance are collapsed

A high buyer price is treated as proof of quality, or a high publisher payout is treated as proof of buyer value.

Buyer price is what the buyer is charged.

Publisher payout is what the publisher is paid.

Neither number proves call quality by itself.

Material source changes are ignored

The source changes creative, traffic channel, transfer process, upstream supplier, geography, or caller experience while keeping the old benchmark.

Buyer-specific data is exposed broadly

One buyer’s performance is presented to another buyer without appropriate aggregation and privacy controls.

The benchmark survives after better evidence exists

The buyer’s own sustained performance contradicts the benchmark, but the operation keeps using the general number because it is more attractive.

A benchmark should become more accurate over time, not more politically convenient.

What publishers should do with benchmarks

Benchmarks can help serious publishers too.

A publisher introducing a new source should prepare:

A narrow source definition.
Consistent source labels.
Traffic-type classification.
Supply-type classification.
Vertical and geography.
Consumer journey.
Creative, landing page, script, or sample-call materials.
Historical figures with provenance.
Date ranges.
Sample counts.
Metric definitions.
Expected delivery pattern.
Known limitations.
Material-change process.

The publisher should be willing to say:

“This figure is self-reported.”
“This figure comes from a different buyer context.”
“This metric uses connected calls as the denominator.”
“This is an estimate, not a guarantee.”
“The source changed after this period.”
“We do not yet have enough live data.”
“The buyer should begin with a controlled test.”

That honesty can create more confidence than an aggressive claim.

A clean benchmark package helps a buyer understand what it is evaluating. It also gives the publisher a fairer chance to build source-specific history instead of being judged by a broad account reputation.

The buyer benchmark checklist

Before relying on a new-source benchmark, ask:

Source identity

Is the source defined narrowly?
Is materially different traffic separated?
Is the traffic inbound, transferred, direct, or aggregated?
Has the source changed since the benchmark window?

Metric integrity

What exactly is being measured?
What are the numerator and denominator?
Is talk time raw, connected, qualified, or billable?
Is conversion defined?
Are routed, connected, qualified, billable, payable, and converted statuses separate?

Evidence quality

Is the figure self-reported or live-computed?
What date range does it cover?
What is the sample count?
Is the benchmark current?
Is the benchmark relevant to this vertical and traffic type?
Are delayed outcomes complete?

Buyer fit

Which campaign should receive the source?
Which target is prepared to handle it?
Which geographies and schedules fit?
What cap and concurrency limit are safe?
Can buyer handling be measured separately?

Test governance

When will the test be reviewed?
What would justify scaling?
What would justify a pause?
Who can change the cap?
Will source or buyer configuration changes be recorded?
Can the source be disabled without affecting unrelated supply?

A buyer that can answer these questions has more than a benchmark.

It has a controlled decision process.

How Dependable Calls is approaching source benchmarks

Dependable Calls is being built around curated source enablement rather than unrestricted source discovery.

The current implementation supports separate buyer-facing tiers for source stats and buyer-specific performance. Source stats can carry provenance, an as-of date, sample information, typical billable talk time, typical daily volume, typical conversion, and ping-to-call context. The source catalog can distinguish self-reported figures from live-computed figures so an estimate does not masquerade as verified history.

The current implementation also supports operator-controlled source offers, buyer source pseudonyms, campaign-level source controls, target-level source controls, buyer-specific source metrics, and live routing gates.

That does not mean a benchmark proves a source is safe, compliant, profitable, or ready for unlimited scale.

It also does not mean every metric is equally mature or every source has enough live history. Some figures may begin as self-reported onboarding data. Some sources may remain below the sample level needed for stable live interpretation. Buyer-specific performance still has to be earned through controlled live use.

The operating principle is simple:

Give the buyer enough evidence to make a scoped decision, label the limits of that evidence, and replace general expectations with buyer-specific history as the relationship develops.

That is how benchmarks should build confidence.

Not by promising the result.

By making the next decision more explainable.

Want access to curated call sources? Apply to join the Dependable Calls buyer beta.