The numbers, with their windows attached.

The published snapshot combines period activity, separately supplied shorter-window rates and inventory. Those measures answer different questions. Activity describes what was counted during a period. A rate needs a defined numerator and denominator. Inventory describes what exists at a point in time.

This article is the canonical home for the dataset. The team and methodology page provides the attribution, while the measurement guide provides a blank method for your own operation.

No customer testimonial, controlled comparison or causal performance claim is being inferred from these counts.

• 37,411 dials: calling attempts, not distinct people reached.
• 17,645 recordings: counted audio records, not qualified conversations by definition.
• 9,367 AI summaries: counted processing outputs, not appointments or closed business.
• 35 appointments set: set appointments, not a claim that they were held, signed or closed.

Scope: our own operation, aggregate and anonymized; reported 90-day activity as of September 26, 2026. These counts are not being presented as successive stages of one matched funnel.

A contact can receive more than one eligible attempt. A recording can capture something other than a qualified conversation. An output may be created after its underlying call. Those distinctions prevent a row of declining counts from being treated as automatic conversion math.

The separately reported 30-day activity totals are 11,266 dials, 5,044 recordings, 4,706 transcripts, 4,197 summaries and 9 appointments set. Scope: our own operation, aggregate and anonymized, as of September 26, 2026.

These are shorter-window output and activity totals. The published extract does not provide the matched records needed to assert that every transcript and summary belongs to the same call cohort as the dial count. Do not read the differences between totals as a verified processing failure rate.

Nor should a 30-day figure be used as the denominator for a 90-day numerator. A shared snapshot date does not make differently defined populations interchangeable.

• 17.8% person-connect: reported completed human-answer attempts divided by attempts in the reported 30-day cohort.
• 26.7% voicemail: reported voicemail outcomes divided by dials in that 30-day cohort.
• 78% AI coverage of connects: human connects with a summary and opportunity flag divided by human connects in the reported 30-day cohort.

Scope: our own operation, aggregate and anonymized, as of September 26, 2026. These are separately supplied rates. The published extract does not include the call-level joins needed to independently reconstruct their qualifying counts from the activity totals above.

In particular, do not divide the summary-event total by an estimated connect count and treat a mismatch as either proof of a broken engine or proof that all connects were covered. The populations and timing must be established first. The rates also should not be applied to the 90-day dial count.

A matched analysis would identify the eligible calls, their outcomes, the available recordings, the outputs linked to those calls and the relevant processing cutoff. It would document repeat attempts, missing joins and any reprocessed outputs.

For an appointment analysis, it would separately establish the appointment records, their relationship to the calls and the later held or business outcome. Those links are not supplied merely because the activity counts appear on the same dashboard.

Until those relationships are verified, the defensible result is the separate published measures with their limits. This page does not calculate an end-to-end conversion rate, customer-acquisition cost or return on investment from them.

15,694 leads describes all-time lead inventory in the system. 33 active lists describes the active list inventory at the snapshot. Scope: our own operation, aggregate and anonymized, as of September 26, 2026.

These are not leads acquired or lists created during the last 90 days. They are not a count of distinct people contacted during either reporting window. Keep inventory separate when assessing the size of the database versus the amount of recent calling work.

For your operation, begin with the decisions that a measure should support. Do you need to understand available operator time, the handling of human answers, incomplete after-call records or follow-up completion? Each question calls for a different definition and source.

Use the matched-cohort worksheet to record population, period, cutoff, numerator, denominator and join. Show the underlying counts alongside a percentage, and preserve unknowns instead of turning them into convenient zeros.

Then investigate a concrete issue in the workflow. A metric is useful when the team can connect it to a record and a responsible action, not only when it fits a prominent proof block.

Ask whose operation is described, when the work occurred, how the events were counted and what the denominator includes. Distinguish a vendor’s own activity from multi-customer research and a set appointment from a later business result.

A useful comparison should preserve differences in list source, eligibility, staffing, timing and configuration. A large sample does not repair an undefined cohort, and a precise percentage does not establish a causal explanation.

Our snapshot establishes reported workload in the Milner Team’s operation. It does not promise the same outcomes for another team or claim that a sandbox trial can reproduce real prospecting performance.