Understand your website analytics before you automate your next decision
By Plain Reckoner · 1,280 words · about 6 min · Updated
A useful website analytics tool makes its inputs, arithmetic and limits clear enough to check. Start with matching units and dates, then compare the cost of acting with the uncertainty in the result. Our free tools explain the first step; you keep control of the decision.
- How it was measured
- An editorial decision checklist, illustrated with explicitly hypothetical calculations. RPM definitions and reporting distinctions checked against primary Google documentation on 2026-10-21. This is not a test of AI products, a representative publisher survey or a promise of traffic or earnings.
- Last checked
Understand first. Decide for yourself. That is the promise we want a website analytics tool to make. You should leave with a clearer picture of your traffic, ad revenue and choices, and enough working to challenge the answer.
Automation can save time. It can collect a report, calculate a rate and help organize an investigation. The difficult part is deciding whether the report answers your question. A confident recommendation cannot supply a missing denominator, repair mismatched dates or establish why your visitors behaved differently.
This guide gives small publishers a practical starting point: three checks before acting on an analytics dashboard, an AI-generated recommendation or a website revenue calculator. You can use the free measurement toolkit to do the arithmetic. The decision remains yours.
1. Check what the number actually measures
Start with one question you could answer from a report. “How is my website doing?” is too broad to test. “Did ad earnings per pageview fall between these two complete months?” has a unit, a period and a comparison.
Write down the names of both the money and the traffic measure. Estimated ad earnings, finalized payable revenue, gross revenue and net profit describe different things. Pageviews, sessions, users and monetized sessions also describe different things. A calculator cannot make them interchangeable.
Google defines RPM through earnings divided by the relevant count, multiplied by 1,000. For page RPM, that count is pageviews. Google’s RPM definition is the starting point; the label in a provider dashboard determines which rate you are actually looking at.
| Question | Inputs that belong together | What the answer cannot establish |
|---|---|---|
| What was my page RPM? | Ad revenue and pageviews for matching dates | Why the rate changed |
| What was my session RPM? | Ad revenue and sessions for matching dates | A directly comparable page RPM without pageviews |
| What might a listing benchmark imply? | Net profit and the stated listing multiple | A buyer’s offer or a completed sale price |
| Is a new network worth investigating? | Relevant traffic, revenue and eligibility rules | Acceptance or an earnings improvement |
The traffic unit matters even when both dashboards use familiar words. GA4’s traffic acquisition documentation describes a session as a group of interactions within its session rules. Counting those sessions is a different operation from counting pageviews. Our session RPM versus page RPM guide works through that difference.
A useful tool makes the denominator visible next to its output. If the unit is unclear, resolve it before comparing the result with someone else’s screenshot.
2. Match the reporting periods before explaining a change
A 15-day period and a 30-day period can produce a dramatic-looking total change without showing a comparable daily change. The same problem appears when one month is incomplete, the revenue report is delayed, or the two exports use different timezones.
Check the date range, complete days, traffic unit, revenue definition and reporting timezone together. Keep an original copy of each source export. If you reconcile adjustments, record what you changed and why.
Consider this hypothetical example, using two complete periods of equal length:
Swipe or scroll the table horizontally to see every column.
| Measure | Earlier period | Later period | Change |
|---|---|---|---|
| Pageviews | 20,000 | 30,000 | +50% |
| Ad revenue | $150 | $180 | +20% |
| Page RPM | $7.50 | $6.00 | −20% |
Both headlines are true: revenue increased, and revenue per thousand pageviews decreased. The arithmetic is 150 ÷ 20,000 × 1,000 = 7.50, then 180 ÷ 30,000 × 1,000 = 6.00. Equivalently, 1.50 × 0.80 = 1.20: 50% more traffic at a 20% lower rate produces 20% more revenue.
That does not prove that the ad network performed worse. Additional visitors might have arrived from different countries, used different devices, read different pages or encountered different consent conditions. Those are questions to investigate in appropriate reports, not conclusions hidden inside the multiplication.
Use the free revenue comparison to compare totals or preview a daily CSV locally. It separates traffic and RPM components using midpoint arithmetic. That split describes the change; it does not identify its cause. See the traffic and revenue troubleshooting guide for the next reporting checks.
3. Compare the decision cost with the strength of the evidence
A benchmark can suggest a question worth investigating. Its usefulness depends on the sample, the unit and the decision you plan to make.
Our historical page-RPM sample contains 123 broker-checked, sale-selected businesses. It is a view of those records, with the selection limits that implies. A median from that group cannot establish what a new site, a different audience or a different ad layout will earn. Read the published data and methods before using the comparison.
The same principle applies to a large external data collection. More country-year observations do not turn connectivity statistics into measurements of your website’s visitors. A relevant, matched observation can answer a narrower question than a much larger file of unrelated values.
Before changing something expensive, write down:
- The observation: which value changed, by how much, and over which dates.
- The uncertainty: what you cannot separate with the available reports.
- The action: the specific change you are considering.
- The cost: money, implementation time and possible disruption.
- The check afterward: which comparable measure would help assess the result.
For an ad-network move, a larger quoted RPM needs to be considered alongside payable revenue, fees, migration costs and traffic assumptions. The network-switching planner makes those inputs explicit. Its result is a scenario under your assumptions; it is not an endorsement or a prediction.
For a possible search decline, the traffic-loss stress test can show how a chosen loss would affect revenue and cash profit. It does not assign a probability to that loss. That distinction lets you use a scenario for planning without treating it as a forecast.
Use automation where you can inspect its work
A sensible starting point is to automate repeatable collection and transparent calculation, while reviewing the choice of question, source, unit and action. The right boundary depends on the consequence of a mistake and the quality of the evidence available.
Ask any analytics product to show its inputs, source dates, definitions and assumptions. Check whether you can export the working. Look for an explanation of what would change the conclusion. A score without those details is harder to use responsibly, whether a person or a model produced it.
For content decisions, a score is only a starting signal. Read the page, examine the question it answers and check whether it helps its intended audience. Google’s people-first content guidance emphasizes useful, reliable content for readers. Repeating keywords or obeying a universal word-count target does not substitute for answering their question.
Your first step is free
Start with a public-site check, a comparison of your own revenue periods, or a clearly labelled historical benchmark. Follow the formulas and sources. Keep the parts that help you understand your situation, and challenge the parts that do not fit.
You can use the free toolkit without an account. If the available evidence does not answer your question, request a free manual review. Explain what you want to understand, and a researcher will look at what can be established and where the limits remain. We offer a reply by email in less than 24 hours; keep sensitive records out of the initial form.
Understanding your website gives you a stronger basis for choosing what happens next. That is the point of the tool.
Tracked as RPM-UNIT-COMPARISON. If a number here looks wrong, tell us and we will check it against the source.