Two Sets of Numbers That Disagree

Both are right

Two properties on the same site, over the same fortnight, will report different visits, different sessions, and different conversions.

Neither is broken; they are counting different things and calling them similar names.

A gap somewhere between ten and thirty per cent is ordinary and nothing needs doing about it.

A gap of eighty per cent is something else, and this is the last month in which you can tell the difference.

The differences that are just definitional

• A session ends differently.
• A user is identified differently.
• Bot filtering is applied differently.
• A pageview is one event among many.
• Timezone and reporting day boundaries.

The first accounts for most of it. The old version started a new session when the traffic source changed mid-visit and the new one does not, which means the same behaviour produces different session counts by design.

The differences you should investigate

Since a large gap points at something fixable.

A tag present on some pages and not others, which shows as a page missing entirely from one property.

Filters applied in one and not the other, particularly internal traffic exclusions set up years ago.

Consent handling differing between the two, so one records visitors the other does not.

And a conversion firing twice in one, which is common where a thank-you page is counted on refresh.

Each of those has a specific check and each is worth doing before the end of the month.

Work through them in that order, since a missing tag explains a large gap and a filter explains a small one.

Compare page by page, not in total

Which is what makes the cause visible.

Take the same fortnight in both and list the top twenty pages by visits in each.

A uniform gap across every page is definitional and needs no action.

A page appearing in one list and not the other is a tagging problem on that page.

And a page with a wildly different figure is usually a duplicate event, or a redirect being counted as two separate pages in one property and one in the other.

That comparison takes twenty minutes and it is the whole diagnosis.

A worked example

A business found their new property reporting about sixty per cent of the old one's visits and assumed the setup was wrong.

Comparing page by page, most pages showed a gap of about fifteen per cent, which was expected.

Three pages appeared in the old property and not in the new one at all, and those three accounted for the rest of the difference.

They had been built outside the main template and had never received the new tag.

Fixing that brought the gap to the ordinary range.

Comparing totals would have suggested a general problem and produced no answer.

Write down the ratio

Which is the thing worth keeping from this exercise.

Once the investigable causes are resolved, note what the remaining gap is: the new property reports roughly this proportion of the old.

That single figure lets you read your historical archive against future numbers approximately, which is otherwise impossible.

It is not precise and it is considerably better than nothing when somebody asks whether this year is better than two years ago.

Record it with the dates it was measured over and the caveat that it is a rough factor.

Nobody can produce it after the end of the month.

Conversions matter more than visits

So reconcile those specifically.

Take one week and count form submissions in the old property, in the new one, and in your own inbox.

Three numbers that roughly agree means the whole chain is working.

Where the tracked figures agree with each other and not with your inbox, the tracking is missing submissions or counting failures as successes.

Where they disagree with each other, one is misconfigured, and you still have the ability to find out which.

This is the single most valuable check available this month.

Do not average them

A tempting response worth naming.

Two figures that disagree do not produce a better one when combined, since they measure different things.

Pick the one you will be using going forward, report from it consistently, and use the other only for this reconciliation.

A report that switches sources between months is worse than one using a consistently imperfect figure.

Consistency matters considerably more than accuracy in a monthly comparison.

An imperfect figure measured the same way every month still shows you direction, which is the only thing a monthly series is for.

Tell whoever will read the reports

Since the discrepancy will otherwise surface as a question you cannot answer in six months.

Write two lines now: the new figures run about this proportion of the old, measured over this fortnight, for these reasons.

Keep it with the exported archive rather than in an email, because that is where somebody will look when the question arises.

Anybody comparing a chart across the boundary later will otherwise conclude that something went badly wrong in July.

Two lines now saves an afternoon of investigation into a step that was entirely expected.

The counter-case

This may not be worth an afternoon.

A business that looks at enquiries rather than visits can ignore the visit discrepancy entirely and check only the conversion count.

Very small sites also produce numbers too low for a percentage gap to be meaningful.

And a general reconciliation produces a figure nobody uses if nobody was comparing years anyway.

Do the conversion check regardless, since that one is about whether your enquiries are being counted at all rather than about historical comparison.

Compare a fortnight page by page, fix any page missing entirely, reconcile conversions against your inbox, and write down the ratio.

The afternoon

1. Take the same fortnight in both.
2. List the top twenty pages in each.
3. Look for pages missing entirely.
4. Check filters in both properties.
5. Count one week's forms three ways.
6. Write down the ratio.
7. Pick one to report from.

Step five is the one with lasting value, since a form count that agrees with your own inbox confirms the whole chain and nothing else does.

Why both were running is covered in running both analytics at once.

Frequently asked questions

West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.