Know what the number means.

Write a metric contract.

Before producing a dashboard, agree on which fields are included, how each measure is calculated and what decision it supports. Preserve the source label alongside your reporting label. Views, impressions and reach should not silently become the same measure because they all describe attention.

For each reporting period, record platform, account, format, publication time, observation time and any paid distribution you can establish. If the actual observation time is unavailable, say so. A refresh time only tells the reader when the application requested or received a response.

Give each measure one job.

The following definitions are a reporting framework, not a promise that every platform supplies every field. Check the current source documentation and the fields available for your account before applying a benchmark.

Make the arithmetic inspectable.

Using fictional data: a post has 240 likes, 12 comments and 8 shares. Supplied interactions total 260. If recorded views are 10,000, the interactions-per-view ratio is 260 ÷ 10,000 × 100 = 2.6%. Comments account for 12 ÷ 260 × 100, or approximately 4.6% of supplied interactions.

If shares are missing, the 260 total cannot be reproduced. Report the 252 known likes and comments as a partial total, identify the missing category and avoid comparing it directly with a complete three-category total. A zero is an observed value; unavailable is a different state.

When a denominator is zero, report the ratio as unavailable rather than dividing by zero. Keep enough precision for the decision, but do not let extra decimal places imply greater certainty than the data supports.

Compare moments fairly.

An older post has had longer to collect views and actions. Use a consistent post age where possible and separate formats and platforms. Record how the sample was selected before examining the results; otherwise the selection itself can favor the story you want to tell.

A snapshot with 1,200 views after two hours has a lifetime pace of 600 views per hour. A later observation of 1,500 views after three hours shows a 300-view increase in that interval. It does not establish acceleration without further comparable intervals. Corrections and provider delays can also affect the difference.

Keep uncertainty visible.

Compare peers with matching metric categories. Show sample size and exclusions, and use a median when a single large result would distort the average. Label repeated posts and do not assume overlapping audiences can be deduplicated from public counts.

A useful report separates observed facts, calculated measures and modeled interpretations. Sentiment, sponsor fit and intent need their own definitions and validation; likes and comments alone do not establish any of them. Ask what a score means before treating it as a decision rule.

• Name the numerator and denominator.
• Preserve missing data; never silently fill it with zero.
• Show the source and relevant timestamps.
• State the sample and the limits of the comparison.

Use the framework in Social Intel.

The public Signal Room provides a bounded sample with source context and explanations beside its measures. The measures can change with the available fields—for example, a views-only video sample needs view-based comparisons. Read the calculation before interpreting the result.

For a customer workflow, agree on coverage, history and refresh expectations during evaluation. Use the downloadable scorecard to review whether your intended report has the necessary evidence, rather than assuming every public preview represents a complete reporting service.