Why your helpdesk shows fewer calls than you get

What we found

These figures come from a Morsegrid case study published on this site, last updated 14 October 2026: a US automotive parts retailer with a Shopify storefront backed by Odoo. The baseline is a verified 30-day window before this engagement.

• 3,067 inbound calls in 30 days, and 155 were answered live. That is 94.9% missed. It is a raw count of calls, so a caller who redials is counted each time.
• 76.7% of the calls came from numbers that had already called.
• In an earlier three-week count, 21 tickets carried a phone tag against roughly 2,161 inbound calls.
• The voice agent now answers 69.5% of closed-hours calls. That figure covers closed hours only, so it is not a before and after of the 94.9%.

The calls that never become tickets

A helpdesk is a list of what the phone system handed it. A call becomes a ticket only when the phone is wired to the helpdesk and the call is a type the integration logs. Gorgias Voice creates a ticket when a call connects, is missed or someone leaves a voicemail. Aircall's Gorgias integration creates one when a call is answered or missed, and only for the numbers you selected during setup.

That leaves three ways to undercount: the phone isn't connected at all, only some of your numbers are connected, or callers hang up before anything rings. Gorgias counts a caller who hangs up in under 10 seconds, before any agent's phone rings, as cancelled, a separate bucket from missed, and its Voice guide does not say cancelled calls become tickets. Check what your own setup does with each one.

At that retailer the helpdesk couldn't show the gap. The case study says the missed calls never entered it, so the helpdesk data said everything was fine while the phone said the opposite. In an earlier three-week count, 21 phone-tagged tickets against roughly 2,161 calls is what that looks like in numbers.

Measure your real answer rate over 30 days

Start from the phone system, not the helpdesk. The count takes one export and a spreadsheet.

• Export every inbound call for the last 30 days: start time, caller number, who answered and, for unanswered calls, the reason.
• Count the calls a person answered and treat everything else as unanswered. Gorgias defines unanswered as missed, abandoned, cancelled and callback-requested calls added together, so a report that shows only "missed" leaves out three buckets.
• Split the unanswered calls by reason. Aircall's missed-call reasons include no available agent, agents did not answer, short abandoned (the caller hangs up in the first 10 seconds), out of opening hours and abandoned in the phone menu. The biggest bucket tells you which fix to build first.
• Count distinct caller numbers next to the raw total. The raw rate sizes the load on your phones. The distinct count sizes the people who went without an answer.
• Divide phone tickets by inbound calls. That is your ticket coverage. A report built from tickets alone can't show it.

Why repeat dialing hides demand

A caller who gets no answer may call again. Each unanswered redial is a new missed call, so one person who tries four times adds four missed calls to the raw count, and may add nothing to the helpdesk. In that retailer's baseline, 76.7% of calls came from numbers that had already called.

Some of those callers are buyers who give up. The case study estimates that roughly 413 buyers a month were walking away, and that fewer than 5% of them were ever reached. The helpdesk didn't see them. The trace was in the phone log.

This is why the obvious fix can be the wrong one. A drowning phone line suggests hiring more people to resolve calls. The case study's finding was that the team resolved the work it received fine. The failure was intake: most calls never reached a person. Every build in that case study started from that.

A callback worklist for every missed caller

Counting is the diagnosis. One fix for the retailer was a ranked, self-cleaning callback worklist that every missed call and voicemail flows onto. It also got three deduplicated sales-recovery lists: lost sales callers, abandoned carts and open quotes, and after-hours leads. The three lists together put more than 1,000 reachable buyers in front of the team, and a dedicated sales seat was approved against the backlog on the strength of the data.

If you build your own, these rules keep it from turning into a graveyard.

• One row per caller, not per call. A redial updates the row instead of adding one.
• Attach what you know: the open order, cart or quote, and why they called if a menu or an agent captured it.
• Rank by buying signal, not by who called first. A caller with an open quote goes above a caller with a general question.
• Close rows automatically. A row clears when the caller is reached, or calls back and is answered. If someone has to clear rows by hand, the list grows until nobody trusts it.
• Give it one owner and one fixed block of time every day.

After-hours coverage and its limits

The retailer put an AI voice agent on its after-hours and holiday branches, where calls used to end in voicemail. It now answers 69.5% of closed-hours calls. It works from a locked knowledge base of facts the client confirmed.

Voice coverage has limits. Three of them matter before you buy any.

• It fixes only the hours it covers. Check the reason split from your count. If most of your misses happen while the shop is open, after-hours coverage won't move your answer rate much.
• It should answer only what is confirmed. The retailer's agent is built never to quote dynamic repair prices or invent warranty terms. On anything uncertain it captures a callback for a person.
• Recording has rules. Washington State requires the consent of all participants before a private phone call is recorded, with narrow exceptions, and treats consent as given when one party announces the recording and the announcement is itself recorded (RCW 9.73.030). Other states and countries differ. If your phone system or agent records calls, get legal advice before you go live. This isn't legal advice.

Check this in your own store

Do this with your own numbers before you change anything. You need your phone system's call history and your helpdesk's ticket list for the same 30 days.

• Answer rate = calls answered by a person ÷ all inbound calls.
• Unserved callers = distinct numbers with no answered call in the 30 days.
• Repeat share = calls from numbers that had already called earlier in the 30 days ÷ all inbound calls.
• Ticket coverage = tickets tagged phone ÷ all inbound calls.

How to read the result

Start with the wiring. Open your helpdesk's phone integration and list which numbers are connected. Aircall's Gorgias documentation says you only receive tickets and tags for the numbers you selected, so a number that isn't selected stays invisible.

Read the result like this. A high answer rate with low ticket coverage means logging is the gap. A low answer rate means coverage or capacity is the gap. A high repeat share can mean callers are redialing after a failed first try.

Sources

Outside sources, read on 17 October 2026.

• Gorgias, Guide to Gorgias Voice: https://docs.gorgias.com/en-US/guide-to-gorgias-voice-190713
• Gorgias, Measure channel performance: Voice: https://docs.gorgias.com/en-US/voice-overview-report-419525
• Aircall, Gorgias v2 integration features and limitations: https://support.aircall.io/en-gb/articles/20378827278493
• Aircall, Missed call reasons in call flows: https://support.aircall.io/en-gb/articles/14936017803549
• Washington State Legislature, RCW 9.73.030: https://app.leg.wa.gov/RCW/default.aspx?cite=9.73.030

Frequently asked questions

How do I measure missed calls in my ecommerce store?

Use the phone system's call history, not the helpdesk. Export 30 days of inbound calls, count the ones a person answered and treat everything else as unanswered. Split those by reason, and count distinct caller numbers next to the raw total. The helpdesk only shows the calls it was given.

Can an AI voice agent answer an ecommerce phone line?

For a defined job, yes. At a US automotive parts retailer we work with, an agent answers from a locked set of client-confirmed facts and captures a callback when it is unsure. It now answers 69.5% of closed-hours calls. It picks up the call and hands the rest to the people who resolve it.

What should an AI voice agent never say?

Anything it can't confirm. That agent is built never to quote dynamic repair prices or invent warranty terms. On anything uncertain it takes a callback instead of guessing. Recording rules vary by place. Washington State requires all-party consent (RCW 9.73.030), so tell callers up front.