Choose the comparison for the question
Different comparisons answer different questions, and a report that mixes them without saying so leaves readers drawing conclusions the data does not support. Start from the question the reader is asking.
| Question | Comparison | What it hides or adds |
|---|---|---|
| Are we on track this month? | The plan or budget, phased by day or week | Depends on how well the plan was phased; a flat phasing makes every seasonal month look early or late |
| Is performance getting better or worse right now? | The previous period of the same length | Mostly measures the season in a seasonal business; weekday and holiday mix differ |
| Are we growing? | The same period last year, aligned | Includes every change made during the year, in prices, range, spend and tracking |
| Is this a trend or noise? | A rolling window, such as the trailing weeks or a rolling year | Responds slowly; a real change takes time to show |
| Is this campaign or market doing better than another? | A comparison group in the same period | Not a period comparison at all; differences between groups are not caused by time |
Two habits help. First, decide the comparison for each KPIA metric tied directly to an objective. before looking at the numbers, and keep it the same from report to report; switching to whichever comparison looks best is a quiet way of misleading. Second, where the plan exists, put it first. A plan that was phased with the season already contains the seasonal expectation, so actual against plan is often the fairest single comparison in the report. Tracking spend and results against that phasing is what Budget PacingHow fast a campaign spends its budget across the flight compared with the plan. means, and it is a comparison like any other.
Calendars: weekdays, holidays and events
Most marketing metrics have a weekly rhythm. Businesses that sell to consumers often peak on particular days, and business-to-business enquiries usually fall away at weekends. A period with one more of the busy days than its comparison will look better without anything having changed.
Align by weekday. Compare periods that contain the same number of each weekday. Week against week does this automatically. Year on yearComparing a period with the same period a year earlier, aligned by weekday and by holidays that move, so that the seasonal pattern drops out of the change., compare a week with the week containing the same weekdays a year earlier, not with the same dates, because a given date falls on a different weekday each year. For months, either compare daily averages per weekday, or use a retail-style calendar of whole weeks so that every period contains complete weeks.
Align holidays and events that move. Easter, school holidays, religious festivals that follow a lunar calendar, sale events tied to a weekday, sporting finals and the business's own promotions all move between years. Comparing the same dates puts the holiday in one period and not the other, producing a large swing in both directions. Align by event instead: compare the holiday week with last year's holiday week, and the weeks either side with the weeks either side. Keep a calendar of these events for every year of history the business uses, because every later comparison depends on it.
Months are not equal. Months differ in length and in their mix of weekdays and weekends. A month with more trading days will look better in totals; daily averages, or totals per trading day, remove most of the difference.
Mark the one-offs. Weather, outages, a competitor's closure, a viral moment: events that will not repeat belong on the calendar too, so that next year's comparison does not treat them as normal. The same alignment is the first check in the diagnosis for Business Profile calls and directions dropping, where a fall in a local listing's actions often turns out to be the calendar rather than the profile.
Partial periods and late conversions
Month to date. A month-to-date figure compared with the whole of last month always looks behind. Compare it with the same number of days in the comparison month, matched by weekday, or with the plan phased to the same day. The phased plan is better because it already knows that some weeks of the month are busier than others.
Late conversions. The most recent days of any report are incomplete. Some people convert days or weeks after the click that the platform credits, leads count only once a salesperson qualifies them, and conversions uploaded from a CRM arrive when the import runs. The conversion lag entry explains the mechanism. A comparison of last week against the week before is therefore biased against last week, and a report read on the first morning of a month understates that month's final figure.
Three practices handle this:
- Mark recent days as provisional in dashboards and reports, for as long as the business's own lag history says conversions are still arriving.
- Compare at equal age. Measure each period's results after the same number of days have passed since it ended, or count only conversions within the same number of days of the click for both periods. This is a CohortA group of customers who share a starting event and period, tracked together over time. comparison: the group of clicks or sign-ups from each period is followed for the same length of time.
- Restate and say so. When a period's figure is updated as late conversions arrive, the report that quoted the provisional figure is footnoted rather than left to disagree with the dashboard.
Breaks in the data
A comparison assumes the same thing was measured in the same way in both periods. Every one of these breaks that assumption:
- A new or reconfigured tag, a migration of the analytics setup, or a consent banner change.
- A metric definition change: fees added to cost, refunds removed from revenue, a lead redefined.
- A change of attribution setting or channel grouping, which moves results between channels without changing the total.
- A site migration, a new checkout or a new domain.
- A new market, product line or sales channel added to the totals.
There are three honest ways through a break. Restate the earlier period under the new rules, where the raw data allows it, and say that it was restated; the metrics layer guide covers versioning definitions so that this is possible. Compare within one side of the break only, accepting a shorter history. Or change the measure to one that did not break, such as orders in the shop system or leads in the CRM, which a tag change does not touch. Where both old and new tracking ran together for a while, the overlap shows how far the measures differ and is worth keeping. In every case, mark the date on every chart that crosses it, and keep the record of measurement changes alongside the data quality routine so that the next analyst does not rediscover it.
Separating growth from seasonality
Year on year takes the season out but adds a year of change. Month on month keeps the season in. To compare consecutive periods in a seasonal business, or to see whether growth is speeding up or slowing down, the seasonal shape has to be measured and removed. The seasonality entry explains how to read the pattern from the business's own records; the steps below turn it into an index. External seasonality figures from other businesses are a poor substitute, because the shape depends on the business's own mix of products, customers and markets.
- Gather the history. Weekly or monthly figures for as many full years as are available and comparable, aligned by weekday and with moving holidays matched to their week in each year.
- Index each year. Divide each week's value by that year's average week. A value above one marks a busier than average week; below one, a quieter one.
- Average across years. For each week of the year, take the average of its index across the years, or the median if one year contains a one-off shock. This is the seasonal index.
- Adjust. Divide each actual weekly figure by that week's index. The result is the figure with the season taken out, and adjusted figures can be compared between any two periods, consecutive ones included.
- Read the trend. A rising line of adjusted figures is growth; a flat one means the movements in the raw numbers were the season.
The index assumes the shape of the year is stable. Check it each year against the new data, and rebuild it after anything that changes the shape, such as a new product range with a different season. With less than a full year of history, the season cannot be separated from growth at all, and the report should say so rather than present a month-on-month change as a trend.
Rolling windows are the simpler companion. A trailing average of recent weeks smooths weekly noise, and a rolling total of the last full year contains every season exactly once, so its movement is growth alone. Both respond slowly, which is the price of their stability.
Frostlantern Ski Hire and the half-term that moved
Frostlantern Ski Hire is an invented ski and snowboard hire business that takes most bookings online. The February school half-term fell in calendar week 8 last season and in week 9 this season. At the end of week 8 the monthly report compares weeks 6 to 8 with the same calendar weeks last season:
| Calendar week | Last season | This season |
|---|---|---|
| Week 6 | 380 | 395 |
| Week 7 | 410 | 455 |
| Week 8 | 1,180 (half-term) | 470 |
| Week 9 | 520 | 1,240 (half-term) |
| Week 10 | 400 | 560 |
Aligned by date, weeks 6 to 8 show 1,320 bookings against 1,970, a fall of 33%, and the owner drafts an emergency discount. The analyst realigns by holiday week instead. The two weeks before half-term show 925 bookings this season (weeks 7 and 8) against 790 last season (weeks 6 and 7), up 17%. Once half-term has passed, the three weeks up to and including the holiday week give 2,165 against 1,970, up 10%, and the five calendar weeks 6 to 10, which contain both holidays, give 3,120 against 2,890, up 8%. Every holiday-aligned comparison shows growth; the date-aligned one showed a fall that was only the calendar. The discount is not run, and the reporting calendar gains a row for each season's half-term dates.
Writing comparisons into reports
The reader of a report cannot see how the comparison was built, so the report has to say. A comparison is fully stated when a reader could rebuild it:
- Name both periods with their dates, not as "this month" and "last year".
- Name the basis: aligned by weekday, aligned by holiday, against the plan, seasonally adjusted.
- Give the absolute and the relative change together. A large relative change on a small base and a small relative change on a large base read very differently once the counts are shown, and the guide to reading small numbers covers putting a range around a figure built on a small count.
- Flag what is provisional, and when the figure will be final.
- Footnote the breaks and the one-offs that sit inside either period.
- Use one basis per table. A table whose rows compare against different periods cannot be read across.
In a dashboard, the comparison period is a setting and it should be visible next to the figures, with the default chosen deliberately; the dashboard design guide covers where comparisons sit in the layout. The reporting cadence playbook decides which comparison each pack uses, and the monthly marketing report template puts results against the plan first, which is the comparison the rest of the report explains. When a comparison shows a change that needs an explanation, the method in explaining a change in a marketing metric takes over, and the BI hub shows where period logic belongs in a reporting stack so it is built once rather than in every chart.
A comparison checklist
Run through these before any comparison reaches a report.
| Check | Question | If not |
|---|---|---|
| Question | Does the comparison answer the reader's question? | Choose the comparison from the table above |
| Weekdays | Do both periods contain the same number of each weekday? | Align by week, or compare daily averages per weekday |
| Moving events | Is every holiday and event in the same position in both periods? | Align by event week and say so |
| Length | Are the periods the same length? | Use daily averages or equal-length windows |
| Completeness | Is either period still collecting late conversions? | Mark as provisional or compare at equal age |
| Breaks | Did tracking, definitions, attribution or scope change between them? | Restate, compare within one side, or switch measure |
| Season | Could the season explain the change? | Compare year on year aligned, or use the seasonal index |
| One-offs | Does either period contain an event that will not repeat? | Footnote it, or exclude it from the baseline |
| Volume | Are the counts large enough to read a change? | Pool periods or show a range around the figure |
| Statement | Does the report name both periods and the basis? | Add the dates and basis to the table header |
Common mistakes
- Comparing month on month in a seasonal business. The change mostly measures the calendar.
- Comparing the same dates year on year. Weekdays shift and holidays move; align by weekday and event.
- Judging month to date against the whole of last month. Compare the same days, or the plan phased by day.
- Reading the latest days as final. Conversions are still arriving; mark them provisional.
- Ignoring a tracking or definition break. A step in the data is not a step in the business.
- Choosing the comparison after seeing the numbers. Fix the basis in advance and keep it.
- Using another business's seasonality. Build the index from your own history, or say the season cannot yet be separated.
- Mixing bases in one table. Every row in a table should compare against the same thing.
- Reporting relative change without the counts. Show both, so that a change on a small base is not mistaken for a big one.
