Products Solutions Templates Pricing Resources

Why your metrics changed: the diagnostic analytics approach

Home / Everything About / Everything About Analytics / Why your metrics changed: the diagnostic analytics approach

Your website traffic was steady at 500 visitors a week. Last week it dropped to 350. You know what happened (fewer visitors). You have no idea why.

Descriptive analytics tells you the number changed. Diagnostic analytics is how you figure out the reason. This article covers the techniques to investigate what actually caused the change, and how to avoid chasing false leads.

Diagnostic analytics is the practice of investigating why something happened. You describe the symptom (traffic dropped 30%), then work backward to find the cause.

Here's what makes it hard: almost everything can seem like a cause if you're not careful. Your bounce rate increased the same week traffic dropped. Did bounce rate cause the drop, or did they both happen to move together but have nothing to do with each other? Your navigation changed. Did that cause the drop, or did Google rankings shift at the same time? Diagnostic analytics is about answering this precisely, not guessing.

How to investigate like you are solving a mystery

The diagnostic process works in steps. Start specific, then expand.

Step 1: Define the symptom precisely

Not "traffic is bad." Something like "organic search traffic dropped 35% between January 15 and January 20 compared to the same period last month."

The specificity matters. "Traffic dropped" is vague. "Organic search traffic specifically dropped 35% in a five-day window" tells you exactly what to investigate and narrows the possible causes dramatically. You're not trying to explain all traffic — just the specific metric that moved.

Step 2: Isolate what might have caused it

What changed around the time your metric moved? This is your hypothesis list.

Did you publish new content? Change your page titles? Update your site structure? Have a marketing campaign? Get mentioned on another site? Was there a Google algorithm update? Did a competitor launch something? Did you change your analytics settings?

Write down every plausible cause. Don't worry if the list is long. Your job next is to eliminate possibilities.

Step 3: Test the hypotheses

For each hypothesis, ask: if this was the cause, what would I expect to see in the data?

Say your hypothesis is "a competitor launched a campaign and stole our search traffic." If that's true, you'd expect to see: (a) certain keywords you used to rank for are now ranking for competitors, (b) your rankings dropped for those specific keywords, (c) the traffic drop is localized to those keywords only, not across all search traffic.

Test that. Do your rankings actually drop for those keywords? If yes, that's evidence for your hypothesis. If no, that hypothesis is wrong — move to the next one.

Same process for the other hypotheses. Did you publish new content? If that caused the traffic drop, why would new content reduce existing traffic? It wouldn't — unless the new content is cannibalizing clicks from your top pages (they now compete for the same search results). If that's not happening, that's not the cause.

Step 4: Check for causation, not just correlation

This step catches most people. Two metrics can move together without one causing the other.

Your bounce rate increased the same week traffic dropped. That's a correlation — they moved together. But correlation is not causation. Bounce rate didn't cause traffic to drop. Both could be symptoms of the same problem (maybe Google demoted your site; fewer visitors arrived, and the ones that did found your content less relevant).

To confirm causation, ask: is there a logical mechanism? If bounce rate caused traffic to drop, how would that work? It doesn't. Bounce rate is how visitors behave after they arrive. It can't affect how many visitors arrive in the first place. So this is probably correlation, not causation.

Step 5: Confirm with the simplest explanation

You've narrowed down the possibilities. Your traffic dropped in organic search specifically. Your rankings dropped. No new site changes happened. Conclusion: Google changed search results (algorithm update or rank shift). That's the simplest explanation.

The simplest explanation that fits the data is usually correct. Don't invent complicated theories if a simple one explains what you see.

The techniques that actually work

Diagnostic analytics uses specific techniques to investigate. Here are the ones that matter for website owners.

Breakdown analysis (segment your data)

Instead of looking at "all traffic," break it down by source. How much came from organic search? How much from direct? How much from referrals?

If overall traffic dropped 30% but organic search only dropped 10%, you know the problem is elsewhere. By breaking down the numbers, you've narrowed the investigation.

Same with pages. If overall traffic dropped but your top pages are actually up, you know the problem is smaller pages tanking. If your top pages are down, that's different. The breakdown tells you what specifically is broken.

Trend analysis (compare periods)

Look at the same metric across different time periods. Your traffic this month vs. last month vs. last year. Did it drop suddenly or gradually? Did it drop on a specific day or over several days?

A sudden drop on March 15 is different from a gradual decline over two weeks. One suggests an event (Google update, site change, competitor move). The other suggests a trend (seasonal decrease, gradual rank loss).

The shape of the trend tells you where to investigate.

Anomaly detection (spot what is different)

Your traffic usually averages 500 visitors per day. Last Thursday it was 650. Last Friday it was 300. Those are anomalies — they're different from normal.

Anomalies deserve investigation. What happened on Thursday? A spike suggests something positive (a popular social share, a mention, good Google rankings). What happened on Friday? A crash suggests something negative (a site error, something changed).

By spotting anomalies, you know which days to investigate deeply.

Cohort analysis (compare groups)

A cohort is a group of visitors that share something in common (they arrived this month, they came from Google, they viewed a specific page).

Say your conversion rate dropped. Instead of just looking at overall conversions, compare cohorts. Did conversions drop for people who came from Google? From ads? From referrals? Did it drop for new visitors but not repeat visitors?

By comparing cohorts, you see the problem is localized to specific sources or visitor types — which is a much better clue than a general trend.

The mistakes that send you down false paths

Diagnostic analytics looks simple but there are common ways to get it wrong.

Confusing correlation with causation

Two things moved together. You assume one caused the other. Usually wrong.

Your conversion rate and your bounce rate both increased. Coincidence? Maybe. Does higher bounce rate cause higher conversion? No — bounce rate and conversion measure different visitor groups. This correlation is probably meaningless.

Always ask: is there a logical mechanism? Could X actually cause Y?

Ignoring external factors

Your metrics changed. Before you blame your site, ask if something outside your site changed.

Did Google release an algorithm update? Did a competitor launch a campaign? Did your industry have a big news event? If your traffic dropped the same week Google announced a core update, that's probably the cause — not something you did.

Look outside your website before looking inside.

Overthinking simple causes

You notice traffic dropped. You investigate for weeks, uncovering complex issues. Then you discover the real cause was simple: your site went down for 2 hours one day.

Simple causes are more common than complex ones. Check the obvious first: did your site error out? Did you accidentally delete a page? Did you change something technical? Did search engine traffic change dramatically? These are easy to check and usually they're the answer.

Changing too many things at once

You noticed a problem. You updated your homepage, changed your navigation, updated your product pages, and rewrote your copy — all in one day.

Now conversions changed. Which change caused it? All of them? One? None — maybe it's seasonal? You can't tell because you changed too much at once.

When diagnosing problems, change one thing and measure the impact before changing the next thing.

When you need help diagnosing

Sometimes diagnostic analytics is straightforward (you made a change, metrics changed, the connection is obvious). Sometimes it's complex and messy.

When you're stuck:

Talk to the people who know the context. If you're an ecommerce store and conversions dropped, ask your payment processor if there was an issue. Ask your team if they changed anything. Ask your developer if there was a site update. The data analyst might find the pattern, but the team provides the context that solves the mystery.

Zoom in on the smallest change. Don't try to diagnose all metrics at once. Pick the one metric that matters most and diagnose that. Once you understand the cause of one change, the others often become clear.

Remember that some changes are just noise. Your traffic goes up and down every week due to random variation. Not every change needs diagnosis. If your average is 500 visitors and you got 490 one week, that's within normal variation. Don't investigate every micro-change — focus on significant deviations.