Designing an Auto-Attendant Callers Don't Hate

The auto-attendant is the first thing most callers experience, and it is usually designed by someone who already knows the answer to every question it asks. That is why so many are bad in the same ways.

None of the fixes below need new software. They are design decisions available on any competent platform.

Rule 1: Options before explanations

Compare:

"Thank you for calling Acme. Your call is important to us. Our business hours are Monday to Friday, 8am to 6pm. For sales, press 1…"

"Thank you for calling Acme. For sales, press 1. For support, press 2. For accounts, press 3."

The caller in the first version waits through eleven seconds of material they did not ask for before the menu starts. They cannot act on anything until the options arrive, so everything before the options is dead air that feels longer than it is.

Hours, website and promotions belong after the options, or in a sub-menu, or nowhere.

Rule 2: Four options, maximum

People cannot hold more than a few spoken options in working memory. Past four, callers stop listening and press whatever they heard last — or zero — which is why overloaded menus produce more misroutes than no menu at all.

If you genuinely have eight destinations, structure them two deep with four at each level. But first check whether you actually have eight destinations or whether you have an org chart. Callers do not know your departments; they know what they want.

Rule 3: Name the outcome, not the department

"Press 3 for Client Services" assumes the caller knows what Client Services does. "Press 3 to check on an existing order" does not.

Write each option as the thing the caller is trying to achieve:

• ✅ "to book or change an appointment"
• ✅ "about an invoice or a payment"
• ✅ "if you're having a problem with a service you already have"
• ❌ "for the Operations team"
• ❌ "for Tier 2"

If two options could plausibly describe the same intent, callers will split between them unpredictably — and every misroute costs two people's time plus a transfer.

Rule 4: The number comes after the action

"Press 1 for sales" makes the caller hold "1" in memory while they decide whether they want sales. "For sales, press 1" lets them decide first and act immediately.

It is a small thing that measurably reduces repeats and misdials, and it costs nothing to reorder.

Rule 5: Always provide a way to a human

Every menu should accept 0 for an operator, whether or not it is announced. Callers who are confused, in a hurry, or using a rotary-style interface will press it anyway.

Equally important: handle the caller who presses nothing. Many callers are driving, or on a handset where DTMF is unreliable, or simply waiting. A menu that loops forever on no input is a dead end. Repeat once, then route to a default destination.

Rule 6: Design the out-of-hours path deliberately

Out-of-hours is where most auto-attendants fall apart, because it is configured last and tested never.

Decide explicitly:

• What plays outside business hours, and does it state when you reopen?
• Is there an emergency path, and does it reach someone who is actually on call?
• Does voicemail reach a monitored destination, or an inbox nobody opens until Monday?
• What happens on public holidays — and is the holiday calendar maintained for next year?

A holiday schedule that lapsed last December is the most common broken thing in any phone system.

Rule 7: Record it properly

Professionally recorded prompts in a consistent voice are a small cost and a disproportionate signal of competence. Mixed recordings — half professional, half recorded on a handset in a noisy office — are worse than all-amateur, because the inconsistency is what people notice.

If you use text-to-speech, use one voice throughout and listen to it on a phone, not on laptop speakers. Numbers, addresses and unusual names are where synthetic voices fail; check those specifically.

Rule 8: Measure it

The platform knows what callers do. Review monthly:

• Abandonment within the menu — callers hanging up before choosing. High numbers mean the menu is too long or the options do not match what people want.
• Option distribution. One option taking 70% of traffic means it should be first, or the others are badly described.
• Zero-outs. Callers escaping to the operator rather than choosing. A high rate means the menu is not answering their question.
• Transfers after routing. Callers reaching one destination and being moved — the clearest evidence of a misleading option.

These numbers tell you what to change far better than internal opinion, and they take ten minutes to read.

What good looks like

A caller hears a one-line greeting, three or four outcome-shaped options in the order matching how often they are chosen, and reaches the right place on the first attempt. Total time from answer to routing: under fifteen seconds.

Anything longer is not giving them information. It is asking them to wait while you describe your organisation to someone who called about an invoice.