Call Centre Service Level: ISO 18295, BDDK, and EPDK

The same month of calls can produce service levels 17 points apart without anyone getting the arithmetic wrong. “The percentage of calls answered within X seconds” sounds like one number, but it hides 2 decisions that nobody writes down: which calls belong in the total, and when the stopwatch starts.

That’s why our customers ask us how the service level in their call centre software can differ from the figure their regulator expects. It usually isn’t a bug. It’s a definition.

This post walks through both decisions, shows what ISO 18295-1, BDDK, and EPDK each decide, and works a single month of calls through all of them.

• ISO 18295-1
• BDDK
• EPDK

The service level formula everyone agrees on

Every published standard defines service level the same way:

Calls answered by an agent within the threshold, divided by calls offered.

The disagreements are entirely about the words offered and within. Get those 2 right and the arithmetic takes care of itself.

Decision 1: which calls go in the total

There are 3 defensible populations, and they produce very different denominators.

Population	What it includes	Who uses it
All inbound calls	Every call that reached your number, including ones handled entirely in the IVR and ones that hung up in the menu	ISO 18295-1
Calls that entered an agent queue	Only calls where the caller asked for a person and joined the queue	BDDK, EPDK
Only calls an agent answered	Calls that were picked up	No standard

That third row is the one to watch. It’s a common default in call centre reporting, and it flatters the number badly, because a call nobody picked up disappears from the total entirely. Miss every call in a bad hour and your service level for that hour is undefined rather than 0%. It was the old Hipcall default too, which is exactly why we replaced it.

Decision 2: when the clock starts

The threshold is a stopwatch, and the standards disagree about when you press start.

• On arrival—the moment the call hits the platform. Greeting, menu, and hold music all count against you.
• On queue entry—after the caller has finished with the IVR and joined the queue for a person. Menu navigation doesn’t count.
• On platform answer—when the system picks up, before any human is involved. This isn’t a caller-experience measure at all, and no standard uses it.

BDDK and EPDK both start at queue entry, and both exclude greeting and IVR time, on the reasoning that a caller browsing a self-service menu isn’t waiting for anyone. ISO 18295-1 doesn’t fix the clock at all and leaves it to the agreement between the contact centre and its client.

In Hipcall, a queue is a team, so “entered an agent queue” means the call was placed in a team’s queue.

What each standard actually says

	ISO 18295-1	BDDK	EPDK
Applies to	Any customer contact centre	Bank call centres in Türkiye	Electricity distribution and supply companies in Türkiye
Formula reference	Annex A, metric 4	MADDE 8(2), %HS = E / C x 100	MADDE 6(1)(b), %SS = F / E x 100
Denominator	All interactions offered	Calls transferred to the agent queue	Calls entering the operator queue
Clock starts	Not fixed by the standard	Queue entry	Queue entry, stated explicitly
Threshold	Client-defined	30 seconds, or 20 seconds for lost, stolen, and suspicious transactions	20 seconds
Target	Client-defined	At least 80% monthly, at least 90% monthly on the lost or stolen line	Not set by the regulation; contractual
Reporting period	Agreed intervals, from 15 minutes to annually	Daily, monthly, annually	15-minute, daily, monthly, annually

Sources: ISO 18295-1:2017 Annex A · BDDK regulation on bank call centre service level and quality, Resmî Gazete 20 May 2020, issue 31132 · EPDK principles for electricity distribution and supply call centre service quality, in force since 1 January 2021. Check the current text of the regulation that applies to you before you report against it.

3 things are worth knowing beyond the headline formula.

Service level is not the only number either regulator wants. BDDK also requires an answer rate (MADDE 8(1), at least 95% monthly) and an accessibility level measuring line capacity (MADDE 5(2), at least 95% annually), plus a monthly quality score of at least 70 out of 100 built from a random sample of recorded calls. EPDK requires accessibility, service level, answer rate, and a satisfaction rate collected by calling customers back within 2 days of their original call.

BDDK excludes statistical outlier days. Under MADDE 8(5), any day whose total call volume exceeds the trailing 180-day mean plus 2 standard deviations is dropped from the monthly average. A single viral incident doesn’t sink your month.

EPDK requires certification, not just numbers. MADDE 5(1)(k) obliges electricity companies to document that the call centre operates in line with TS EN ISO 18295-1 and -2, TS EN ISO 9001, TS ISO 10002, and TS EN ISO/IEC 27001, certified by a TÜRKAK-accredited body, with the audited report submitted to EPDK by the end of March each year.

The same month, 4 different answers

Here’s a support number with 1,000 inbound calls in a month.

• 120 never reached an agent queue—self-served in the IVR, called out of hours, or hung up in the menu.
• 880 entered the agent queue.
• 800 were answered by an agent.
• 80 abandoned while queuing, 35 of them within 5 seconds.

Of the 800 answered calls: 660 were picked up within 20 seconds of joining the queue, 720 within 30 seconds, and 700 within 60 seconds of the call first arriving.

Same calls, 4 definitions:

Definition	Answered in time	Total	Service level
ISO 18295-1, 60 s from arrival, all inbound	700	1,000	70.0%
EPDK, 20 s from queue entry, queued calls	660	880	75.0%
BDDK, 30 s from queue entry, queued calls	720	880	81.8%
Old default, 60 s from arrival, answered calls only	700	800	87.5%

Look at the first and last rows. Identical numerator, 17.5 points apart, purely because one keeps the calls nobody answered in the total and the other doesn’t.

The exclusion trap

At some point someone will suggest removing short abandons—the caller who joins the queue and hangs up 4 seconds later, before any agent could realistically have reached them. It feels fair. It’s also a common way a reported service level quietly stops matching its standard.

ISO 18295-1 rules it out in the definition itself:

No exclusions to be factored in such as abandoned contacts under threshold.

Neither BDDK nor EPDK defines an exclusion either. In the worked example above, dropping the 35 short abandons moves the ISO figure from 70.0% to 72.5%—small enough to look harmless, large enough to matter when your target is 80%.

There are legitimate reasons to do it anyway: a client contract may specifically call for it. Just know that the moment you do, the number you’re publishing is your own metric, not an ISO 18295-1, BDDK, or EPDK one.

Classifying short abandons is a different thing from excluding them, and it’s worth doing. In Hipcall the short abandoned threshold (8 seconds by default) gives those calls their own column in the reports, so you can see how many there were, without touching the denominator.

Which standard applies to you

• A bank in Türkiye—BDDK. It isn’t a choice.
• An electricity distribution or supply company in Türkiye—EPDK, along with the certification requirement.
• A BPO or outsourced contact centre—whatever your client’s SLA specifies. ISO 18295-1 is the usual reference frame, and part 2 puts the obligation on the client to agree the service level with the contact centre, taking customer wait tolerance into account.
• Everyone else—ISO 18295-1, then set your own threshold and target. There’s nothing sacred about 20 seconds, 30 seconds, or 80%. What matters is that the definition is written down and stays fixed long enough to mean something.

How this works in Hipcall

Service level is configurable rather than hard-coded, under Settings → Communication → Phone → General → Call metrics.

Pick a preset—ISO 18295, BDDK, EPDK, or Custom—and it writes the population, the clock, and the threshold together. Choosing BDDK sets calls-that-entered-a-queue, the queue clock, and 30 seconds in one step, so the 3 axes can’t drift out of alignment with each other.

Custom exposes the same axes individually, with the warnings attached where they belong. Tick “short abandoned calls” under Remove from the total and the form tells you ISO 18295 forbids it before you save.

5 details matter for anyone who reports these numbers to a regulator or a client:

• Presets store resolved values, not a label. Your account keeps the actual axes. If we later refine a preset definition, your historical figures don’t move—the page shows you the difference and lets you apply it deliberately.
• Changing the setting never rewrites history. We record when the definition changed, and a report spanning that date says so. An auditor asking “what was March?” gets the number March was reported with.
• “Within 30 seconds” means a 30-second wait counts. The threshold you enter is the number your contract or regulator states, and the form shows you the exact comparison it will run.
• Missing data shows as an em dash, never 0%. If you measure on the queue clock over a range with no queue data, the report says so rather than rendering a catastrophic-looking zero.
• Nothing moved when we shipped this. Existing accounts were migrated to Custom carrying their previous definition exactly, so no reported figure changed on deploy. New accounts start on ISO 18295.

The number report in our call centre software also splits inbound calls into agent answered, system answered, and never answered—a partition that sums to the total—with the reason each unanswered call was lost. In a typical account, a large share of inbound calls are resolved in the IVR without ever reaching an agent, and that used to look identical to a missed call. Now it doesn’t.

If you’re measuring on queued calls, the report adds a queue answer rate card next to the service level: answered divided by queued, which is what BDDK calls %KO and EPDK calls %CO. Same ratio, 2 names, and both regulators want it alongside the service level.

3 rules that survive every standard

Whichever definition you land on:

1. Keep unanswered calls in the total. A denominator of answered calls can’t drag your figure down no matter how badly the day goes, which is precisely the problem.
2. Don’t start the clock when the agent’s phone rings. Measure what the caller experienced, from arrival or from queue entry.
3. Don’t delete the short abandons. All 3 standards keep them, and the gain from removing them is exactly the amount by which your number stops being comparable.

Then write down which definition you use and the date you last changed it. Most service level arguments turn out to be arguments about an undocumented definition, and they end the moment somebody produces the definition.

If you’re setting this up for the first time, start on the call metrics page, pick the standard that applies to you, and check the preview before you commit—it shows what the last 30 days would have reported under each standard. And if you’re still deciding whether you need queue-level measurement at all, call centre software vs a business phone system is the better place to begin.