Build your own baseline before borrowing anyone else’s.

Decide what you want to understand: how many eligible attempts were made, how often a person answered, whether recorded calls produced a reviewable output, or what happened after an appointment was set. Those are different questions and may require different cohorts.

Choose the reporting window, time zone, campaign scope and data cutoff. Identify the record keys needed to join an attempt to its recording, processing output and follow-up. If the join is unavailable, report separate counts instead of manufacturing a funnel.

Keep a record of exclusions and unresolved cases. A report becomes less useful when difficult records silently disappear between its numerator and denominator.

• Question: [the operational question being answered].
• Population: [campaigns, operators, contacts or attempts included].
• Window: [start, end, time zone and event used to place a record in the window].
• Cutoff: [when delayed records were last included].
• Numerator: [exact qualifying event and count].
• Denominator: [eligible records for that same definition and count].
• Join: [verified record key or explanation that counts are not joined].
• Exceptions: [unknown, pending, excluded or corrected records].
• Owner: [person responsible for reproducing the measure].

Leave an unknown field unknown. Do not insert a plausible denominator merely because the result would look familiar.

For a defined cohort, a rate is the qualifying count divided by the eligible count. Display both counts alongside the percentage. When the denominator is zero, report that the rate is not applicable rather than displaying a misleading zero-percent performance result.

Human-answer rate: define which attempts were answered by a person and which eligible attempts form the denominator. Keep human answering separate from a business-defined “qualified conversation,” which may require additional criteria.

AI coverage: define the expected output and eligible source calls, then join the output to those calls. Summaries produced this week can relate to earlier calls; dividing that event count by this week’s connects can mix populations.

Appointment follow-through: separate requested, confirmed, held, canceled and other relevant states under your actual process. An appointment set is not a held meeting or closed business.

A contact may have more than one eligible attempt. Report both attempts and distinct contacts when the question requires that distinction. A higher attempt count does not necessarily mean a broader audience or more conversations.

Preserve the contact and attempt identifiers needed to understand repeats. Check whether an apparent new record is a duplicate import rather than a genuinely new person. Do not treat suppression or assignment conflicts as a reason to count the more convenient version.

When comparing two periods, describe changes in list source, eligibility policy, staffing, timing and configuration. A changed mix can affect results even when the software is unchanged.

Choose whether the report is about calls occurring in a window or outputs produced in a window. Both can be useful, but they answer different questions.

For a call-cohort coverage report, allow a documented processing cutoff and join the outputs back to those calls. At that cutoff, distinguish completed output, still-pending work, unavailable source audio and unresolved failures where the actual data supports those categories.

An empty field alone does not establish the reason. Keep the uncertainty visible until the underlying record is reviewed. Do not count a reprocessed or duplicate output as an additional completed conversation.

Use your own ordinary operating periods to build a baseline. Keep definitions stable enough that a change in the number means a change in the work, not a renamed status or a different join.

For a small cohort, show the counts and avoid a confident causal conclusion from a large percentage swing. For a configuration change, document what else changed and what remained comparable. A before-and-after table alone does not prove the dialer caused the result.

The published DialBreeze snapshot illustrates why cohort discipline matters: its activity totals, shorter-window rates and inventory are separate measures. This guide deliberately does not copy those values into your blank worksheet.

A completed measure should be explainable from its definition, records and cutoff. Another responsible person should be able to identify what was counted, what was not and why the denominator belongs with the numerator.

Keep the source records appropriately protected; a useful aggregate does not require exposing personal lead details. Report unresolved data gaps alongside the number rather than making the dashboard appear more complete than the evidence.

Use the result to choose a concrete next investigation: an import problem, a missed callback, unavailable audio or a change in the eligible list. A benchmark is useful when it improves the work, not merely when it makes an impressive headline.