Measure response time end to end—not with a “fastest” headline.

Three clocks, measured separately

"Response time" hides three different experiences. Time each one on its own and never add them together into a single headline figure.

• Pickup time: from the moment you press dial to the first audible sample of the greeting. If calls reach the receptionist by forwarding, this includes your own phone's ringing.
• Turn latency: from the end of your sentence to the first audible sample of the reply. Do not substitute speech-to-text completion, model first-token time or any server timestamp that leaves out the phone network.
• Time to summary: from hanging up to the call summary landing in your inbox. This is the handoff the business actually receives.

Sample call flow: where each clock starts and stops

Sample flow, not a recorded call. The marks below are the five moments to write down on every test call.

Repeat T2 and T3 for every turn in the script below. One call gives one pickup time, nine or more turn latencies and one time to summary.

Test script: nine turns that exercise each tool

Run it from a mobile phone on the network your customers use. The right-hand column says what an Anna account is built to do at each turn; where a feature is switched off for the account, the correct result is a polite refusal, and that turn still counts.

Place the calls on different days and at different hours, and do not run turn 8 against a real emergency line without telling the owner first.

Results template: copy into a spreadsheet

vendor,scenario,call_id,metric,start_mark,end_mark,milliseconds,valid,notes
VendorA,no-cooling,001,pickup,T0,T1,,yes,
VendorA,no-cooling,001,turn-2,T2,T3,,yes,
VendorA,no-cooling,001,summary,T4,T5,,yes,

Record every attempted turn, including timeouts, interruptions, wrong turns and unusable audio, and mark each one valid or not with a reason. Compute the median, the 75th and 95th percentiles, the minimum, the maximum and the count from the valid rows, and quote the count of attempted turns beside them.

What published sources say

• Speed to lead: 37% of 2,241 U.S. companies replied to a web test lead within an hour, 24% took more than 24 hours, and the average was 42 hours among those that replied within 30 days. Method: the authors submitted a web-generated test lead to each company and timed the first reply. These were web forms, not phone calls. Checked 2026-10-18. Source: Harvard Business Review, March 2011.
• Network delay: ITU-T G.114 recommends that one-way delay not exceed 400 ms for general network planning and notes that highly interactive tasks such as voice calls can be affected by much lower delays. Method: this is a planning limit for the network path; a caller-side stopwatch test includes that path, which is why a result belongs to one phone, network and time window. Checked 2026-10-18. Source: ITU-T G.114, approved 2003-05-07, in force.
• Forwarding delay: Verizon describes conditional forwarding as sending a call on when the line is busy or unanswered after 3-6 rings. Method: read from Verizon's support page, so time the forward on your own line rather than assume it. Checked 2026-10-18. Source: Verizon call forwarding support.

Questions about timing a receptionist

How do I measure how fast an AI receptionist responds?

Time two things separately with a stopwatch or an audio editor. Pickup time runs from pressing dial to the first audible greeting. Turn latency runs from the moment you stop speaking to the first audible sample of the reply. Repeat each at least 30 times, at different hours, and report the median and the slowest 5%.

What makes a response-time test fair?

Use the same script, phone, network, time window and retry rules for every system, and write the plan down before the first call. Keep every failed or interrupted turn in the results. A test that drops the bad calls, or sets one vendor's best call beside another's average, shows a preference, not a measurement.

Does VantaWeb claim to be the fastest AI receptionist?

No. VantaWeb publishes no measured time to answer and no latency target for Anna, and it has no comparative test to support a fastest claim. Run the method on this page against any vendor you are considering, including VantaWeb, and keep the raw timings. Treat a vendor's own number as a claim until you reproduce it.

Does call forwarding add time before the receptionist picks up?

Yes, when you use no-answer forwarding. Your own phone rings first, and the carrier forwards the call only after its ring count. Verizon's support page describes forwarding after three to six unanswered rings. Time the full path from dialing your public number, not only the forwarded leg, because that is what a caller experiences.

How many calls do I need before the numbers mean anything?

Use at least 30 valid turns per scenario and say how many you attempted. A handful of calls cannot show normal variation, and the slowest turns matter most to callers, so report the 95th percentile next to the median. With fewer than 30 turns, describe what you saw as anecdotes, not a benchmark.

Pick someone else if

Facts as read on 2026-10-18. A timing method will not fix a mismatch in what the product does:

• You need a written speed guarantee. VantaWeb publishes none, so ask any vendor for the figure and the test behind it, and put it in the contract.
• Your calls start inside Jobber, Housecall Pro or ServiceTitan. Jobber Receptionist, Housecall Pro's CSR AI and ServiceTitan's AI Virtual Agent each work from the data in their own account, per Jobber, Housecall Pro and ServiceTitan, checked 2026-10-18.
• You want a person on every call. No humans sit behind Anna, so a human receptionist service is the better choice when every caller must reach a person.

Limits that apply to any test of Anna: VantaWeb does not write into ServiceTitan, Jobber, Housecall Pro or any CRM. Anna does not take Spanish-language calls. VantaWeb is not offered for healthcare practices and Anna does not handle PHI. VantaWeb has no missed-call text-back: a call nobody answers gets no text.