85 calls. 8 days.
Zero missed.

What exactly was measured?

Every inbound call to the client's business line across eight consecutive days, from 25 February to 4 March 2026. No sampling, no selected window, no excluded calls. The agent answered the business's main published number; nothing was routed away from it during the period.

The client is an antique and estate buyer serving northern New Jersey. Before the deployment the owner answered the phone himself, including while driving to house calls and while standing in front of customers.

What does the agent actually do on a call?

Three things, in one conversation, without a human involved:

• Voice intake. What the caller has, its approximate age and condition, whether it is a full estate or a single piece, and their contact details.
• Mid-call photo collection. While the caller is still on the line, the agent sends a text asking for photographs of the item. Not after the call — during it, while intent is highest.
• Location confirmation. Where the item is, checked against the buyer's service area, before anyone commits to a visit.

What counts as a qualified lead here?

A call is counted as a qualified selling lead when the caller intended to sell an item and the agent captured their contact details together with a description of what they had. Of the 85 calls, 74 met that bar. The remaining 11 were calls that were not selling enquiries at all — wrong numbers, existing-customer questions, and general enquiries.

The 74 leads came from 54 unique phone numbers, which means some callers rang more than once. Counting unique numbers rather than calls, the contact database grew by roughly seven new contacts per day across the period.

How much did each lead cost?

Here is the whole calculation, at list price, so it can be checked:

This figure covers AI usage only. It excludes the $300 monthly retainer and the one-time setup fee, which are fixed and do not vary with call volume. A business handling this call volume pays roughly $313 per month all-in.

How does that compare to buying leads elsewhere?

Against published 2026 cost-per-lead ranges for local service businesses:

The comparison deserves a caveat that most vendor case studies leave out. Paid channels generate demand — they create calls that would not otherwise have happened. This agent captures demand that already existed and was previously going to voicemail. They are not substitutes, and the honest claim is narrower than the table suggests: the cheapest lead available to this business was the one already ringing its phone.

What would the same coverage cost in staff?

Receptionist figures are wage cost only and exclude payroll taxes, benefits, holiday and sick cover, and cover 40 hours out of 168.

The difference is not only the monthly number, it is what arrives: message-taking delivers a name and a number, and this delivers a described item, a location, and photographs.

What happened outside business hours?

Eleven of the 85 calls — 13% — arrived outside normal working hours. Under the previous arrangement those calls reached voicemail. All eleven were answered, qualified and photographed on arrival.

For a business where the seller is often clearing a house on a weekend or an evening, this is the segment with the least competition for the call. Nobody else picks up either.

What are the limits of this data?

Stated plainly, because a case study that only reports what flatters it is not evidence:

• One business, one vertical, eight days. 85 calls is enough to show the system works end to end. It is not enough to establish a seasonal pattern or a conversion rate.
• No revenue figure is claimed. We measured calls, leads and cost. We did not track how many of those 74 leads became purchases, so no revenue impact is asserted here.
• Lead quality is not compared to a human baseline. There is no matched period of human-answered calls to compare against, so the improvement over the previous arrangement is not quantified.
• Cost per lead excludes fixed costs — retainer and setup — as labelled above.
• Photo submission rate was not recorded for this period. The agent requested photos on every selling call; the share of callers who sent them is not in this dataset.

Has it worked anywhere else?

A second agent is live at Clar Fine Violins in Fort Lee, New Jersey, handling intake for instrument repair and appraisal enquiries. The owner reports being fully booked following deployment. That is the owner's account rather than a measurement, and no figures are claimed for it here — when a comparable measured period exists it will be published on this page with the same method.

Method and reproducibility

Call counts, timestamps and outcomes come from the call platform's own logs for the client's line across the eight-day window. Usage cost is computed from total call minutes at the published rate of $0.18 per minute. Cost per lead is total AI usage cost divided by the number of qualified selling leads. Comparison figures for other channels and for answering services are published 2026 list prices, retrieved August 2026.

Deployment: NJ Antique Buyers, Rochelle Park, NJ. Measurement window: 25 February – 4 March 2026. Stack: Vapi (voice), Twilio (telephony and SMS), n8n (automation).