How The Veterinary Phone Line is measured

The panel

The panel is a selection of veterinary clinics on PupPilot in the United States, Canada and the United Kingdom. Demo and internal organizations are excluded, as are animal shelters, whose calls follow a different pattern (intake, adoption, lost-and-found) and are analyzed separately. The reporting window for the 2026 first look is March 1 to September 27, 2026 — more than 100,000 inbound calls and hundreds of thousands of minutes of conversation. The panel is held fixed within each reporting period; a comparison between periods is made only on clinics present in both, so that a change in the figures reflects a change inside clinics rather than a change in who is in the panel.

A clinic joins the panel only when its line is being used by the public: at least 50 inbound calls in the window, from at least 20 distinct callers, with no single number accounting for half of its calls. The rule removes onboarding accounts, which are not clinics being called by pet owners, and costs the panel well under one percent of its calls.

Some figures use a narrower cohort. "Share of a clinic's calls" measures — calls per weekday, the hour and weekday shape, the after-hours share — are computed only on full-line clinics: clinics that route every call through the assistant, during business hours and after them, with at least 100 business-hours calls and at least 10 after-hours calls in the month. Clinics that use the assistant only when the desk is closed cannot show what share of their calls arrive after hours, so they are left out of those measures and kept in the others.

What is in the data

The source is a nightly, de-identified export of PupPilot's production call records into an analytics store. It holds codes, flags, counts, durations, times and money — how a call ended, whether a transfer was attempted, how long the call lasted, which assistant answered, the categories a classifier assigned. It holds no names, phone numbers, email addresses, transcripts, summaries or free text, by design: a caller appears only as a keyed pseudonym that cannot be reversed and that differs between clinics, and a clinic appears only as an identifier. Times are stored in UTC and converted to each clinic's local time for every hour-of-day and weekday figure.

The constructs

Inbound call

A call placed to the clinic and answered by the assistant. Outbound calls (reminders, follow-ups) are excluded from every figure.

Business hours and after hours

Two constructs, published separately. Clock-based: the call's clinic-local time, with Monday–Friday 8 a.m.–5 p.m. as a fixed reporting window and "after 6 p.m. or weekend" as a second one. Assistant-based: whether the call was answered by the after-hours assistant, which also takes lunch-hour closures and daytime calls the desk did not answer. Neither is a measure of the clinic's actual opening hours.

Transfer answered by a person

A transfer from the assistant to clinic staff that completed on a call that did not end in the clinic's voicemail. The ring window is the clinic's configured transfer timeout, about 28 seconds on average. Pickup time is measured from the start of the dial to the moment the clinic's line joined, on transfers from July 2026 onward. "Not answered by a person" includes transfers that reached voicemail, timed out, or were cancelled or abandoned before pickup.

Call reason (category)

Labels assigned to the call's inbox item by PupPilot's categorizer. The pages report selected categories — the reasons that matter most for how a front desk plans its work — rather than every label, and a call can carry more than one label, so the shares do not sum to 100%.

Urgency

The urgency the caller stated, as classified (routine, urgent, emergency), on the calls where one was stated — 59% of business-hours and 42% of after-hours phone items. It is not a clinical assessment.

Caller-behavior figures

Call length and speaking behavior describe what callers do, so two further exclusions apply to them. Numbers that placed more than 30 calls to one clinic in the window are removed: those are clinic lines, PBXs and answering services, not clients. Calls that ended on a line fault rather than by the caller or the clinic — a transfer leg left open for hours, for example — are excluded from the caller-behavior figures. Both exclusions are recorded in the snapshot.

Call length

Recorded duration of the call in seconds, reported as medians and percentiles for three groups: business-hours calls the assistant resolved alone, business-hours calls involving a transfer, and after-hours calls. It is elapsed call time, not staff handling time.

Spoke to the assistant

The caller said anything at all during the call. A silent hang-up is a call with no caller speech.

Species, language

Species mentioned on the call, in order of mention, as extracted; languages detected on the call. Both are model outputs.

Repeat caller

The same pseudonymous number calling the same clinic again within the window. Numbers with more than 30 calls in the window — clinic lines and PBXs — are excluded.

How the figures are produced

A fixed battery of saved queries runs against the export and writes a dated snapshot; every published figure is computed from one snapshot and traces to the query that produced it. Approximate percentiles can move by one unit between runs, so a published figure is frozen at its snapshot and not recomputed until the next edition. The export log is checked before any figure is trusted; a day whose export failed is never reported, and never reported as zero. Before the first edition, the panel was audited call by call; the eligibility rule above is the result.

What these figures can and cannot support

They describe clinics with an assistant answering first, not the industry as a whole, and a customer mix skewed toward the United States and toward clinics large enough to route every call. They describe calls, not clients: a caller can appear on many calls. They are not trends unless the comparison is made on the same panel. Nothing here identifies a clinic, a person or a state. Each finding states, beside its figure, the specific limits that apply to it.

Corrections and citation

Figures are reviewed at each refresh; corrections are recorded in the changelog on the data page with the date and the reason. Data on these pages is published under CC BY 4.0; a suggested citation appears at the foot of every finding.