Every support team watches the same number climb at the same times of day. Calls come in faster than people can pick them up, the queue backs up, and a share of callers give up before anyone says hello. That share is your abandonment rate, and it is the cleanest signal you have that demand and capacity have come apart.
What makes the number harder in Ethiopia is that capacity is not just a headcount question. A caller who reaches an agent who does not speak Afaan Oromo has technically been answered, but the call still fails. Language coverage and staffing coverage are two separate ceilings, and the queue hits whichever one is lower.
Why callers actually hang up
Three causes account for most of what we see when teams share their call data with us.
- Hold time. Abandonment rises sharply somewhere between 60 and 90 seconds of hold. Past two minutes, most of the curve has already happened.
- Menu friction. Every IVR level costs callers. An eight-option menu read in a second language costs more than the hold time that follows it.
- Hours. Anything that arrives after closing is not abandoned so much as never offered. It shows up as a missed-call figure that nobody owns.
What adding people does, and does not, fix
More agents raise the ceiling, and that is genuinely useful. What they do not do is change the shape of the day. Volume in a support queue is spiky, and staffing for the spike means paying for idle capacity in the trough. Most teams settle somewhere in between and absorb the abandonment at peak as a cost of doing business.
A voice agent changes the shape rather than the ceiling. Because it answers every call the moment it arrives, hold time stops being a variable. The queue does not form, so there is nothing to abandon.
“We stopped losing calls after hours. That alone changed how our customers talk about us.”
Deflection is not resolution
This is where a lot of voice deployments stall. An agent that answers, collects a name and a reason for calling, and then hands every conversation to a person has moved the hold time, not removed it. The caller waits in the same queue, one step later.
The gain comes from resolution: the agent completing the routine request itself. Balance checks, branch hours, claim status, delivery windows, password resets, ticket creation. In the deployments we run, these are the majority of inbound volume, and they never need to touch a person.
| Call type | Share of volume | Handled end to end |
|---|---|---|
| Status and balance checks | 38% | Yes |
| Hours, locations, general info | 21% | Yes |
| Ticket creation and follow-up | 17% | Yes |
| Account changes | 12% | Partly, with verification |
| Complaints and exceptions | 12% | No, routed with context |
The last row matters as much as the first. When a call does need a person, the agent hands it over with the transcript, the caller's intent and any records it already pulled. The human starts at minute three of the conversation instead of minute zero.
Language is the other half of the problem
A queue that answers instantly in one language has solved half of its abandonment. Callers who open in Amharic, switch to English for a product name, and switch back mid-sentence are not an edge case here, they are the normal case. An agent that cannot follow that will lose the call just as surely as a two-minute hold will.
What to expect in the first month
The first week is usually spent watching transcripts and tightening the knowledge base. The abandonment number moves immediately because hold time goes to zero on day one, but resolution rate keeps climbing for a few weeks as gaps in the knowledge base get closed.
See it on your own call volume
Tell us where your queue backs up and we will show you what Lucy does with it, in Amharic, Afaan Oromo, Tigrinya or English.
Book a demoWritten by
The Lucy Team
Lucy AI
