Is 65% the resolution rate I should expect?

No. It is a vendor reported figure for routine calls, published by Ringg and repeated in OpenAI's customer story. Treat it as a ceiling, apply it only to the routine slice of your own call log, and plan lower for the first month of tuning.

Does a voice agent only handle simple FAQs?

The published claim is broader than FAQ: bookings, status checks and similar requests resolved without a human. What decides it in practice is whether the answer lives in a system the agent can read and write, such as your calendar or CRM. If it does not, the call becomes a transfer.

How do I test this without committing?

Take one week of call reasons, sort them into routine and judgement, and route only one routine category to the agent while staff keep the rest. Measure the escalation rate and the transfer quality before you widen it.

A voice agent claims 65% resolution: the math to run before you buy

A vendor percentage is only useful once you have turned it into your own number, and most coverage of this announcement stops before that step.

What was published

OpenAI's customer story and Ringg's own blog both say Ringg's agents resolve up to 65% of customer calls without human involvement and that the platform handles more than 7 million connected calls each month. Ringg's docs add that the platform supports 10,000+ simultaneous calls, 20+ languages and claims 99.9% script adherence.

There is also a cost line: Ringg says selected real time workloads cost approximately 90% less after moving from GPT-4.1 to a newer OpenAI model. That is Ringg's model bill, not yours. Your bill is whatever the platform charges per minute.

I have not tested Ringg. Everything above is from the primary sources.

What it means for a small multi-location operator

If you run two or more locations and a shared phone number, a voice agent can plausibly absorb a large share of your repetitive inbound calls, but the share depends on your call mix, not on the vendor's average.

The 65% is described as routine calls. Bookings, reschedules, hours, order or job status, simple price questions. It is not a claim about calls where someone is upset, negotiating, or describing a problem nobody has written a script for.

The worksheet

No numbers here are mine. Fill them from your own phone system, in about forty minutes.

Line	Where it comes from
A: calls received in a typical week	phone system or call tracking report
B: share of A that are booking, reschedule, hours, status or simple price	listen to or read a week of call reasons
C: share of B where the answer already sits in a system an agent can read	your calendar, CRM or order table
Routine answerable calls = A x B x C	multiply
Planning figure	cap it at the published 65% of that result, then halve it for month one while you tune

Then the cost side: contained calls x average handle time x the vendor's per minute price, against the same minutes of staff time. If the gap is small, the win is coverage after hours, not payroll.

Related reading on wiring this kind of thing into a CRM sits on the blog index.

What to ask before you sign

Which call types does the 65% cover, and what is the escalation rate on the rest.
What happens on transfer: does the human get the transcript and the caller's record, or start cold.
Which languages and accents are supported for your actual customer base, not the marketing list.
Does it read and write your scheduler and CRM, or only read a knowledge base.
What the per minute price is at your volume, including failed and abandoned calls.

Limitations

The 65% is self reported by the vendor and its model provider. No independent audit is published that I can find.

The 90% cost reduction is a model cost, measured on selected workloads, and does not translate into your invoice.

Script adherence is not the same as being correct. An agent can follow the script perfectly and still book the wrong slot if the calendar integration is wrong.

My guess, not a fact: the first month of any deployment runs well below the published ceiling because the routing rules and the escalation wording are wrong until real calls correct them.

My take

Use the number as a benchmark for the conversation, not as a business case. Pull one week of call reasons, sort them into routine and judgement, and see what is left. If the routine pile is thin, keep the humans on the phone and spend the money on the booking flow instead.