Quick answer
Data segmentation is the practice of dividing a dataset into smaller groups by one or more conditions, such as who people are, where they are, what they did or how they arrived, so each group can be examined and compared on its own.
Formula: Share of a segment = segment size ÷ total. Rough 95% margin for a rate p on n sessions = 1.96 × √(p × (1 − p) ÷ n).
What is data segmentation?
Data segmentation is the practice of dividing a dataset into smaller groups by one or more conditions, so each group can be examined and compared on its own. The conditions can be who the people are (age, company size), where they are (country), what they did (bought, searched, cancelled), how they arrived (channel) or when (week, season). The point is that an average over everyone can hide patterns that are obvious inside each group.
Experian describes it as organising data into smaller, more manageable groups; Coursera adds that it is most closely tied to sales and marketing, where the groups are used to tailor messages and offers. In security and networking the same words mean something different: separating data or networks so that only some people or systems can reach each part. This page is about the analysis meaning.
Two neighbours are worth separating. A segment is one reusable group; data segmentation is the act of choosing the conditions that produce the groups. Audience segmentation is the marketing use, where the groups become ad or email audiences.
Common segmentation dimensions
Experian lists behavioural, demographic, firmographic, geographic, psychographic and technographic splits. In product and web analytics, these map onto a handful of columns you already have:
| Dimension | Question it answers | Example split | Where the data comes from |
|---|---|---|---|
| Demographic | Who is this person? | Age band, role | Sign-up form, profile |
| Firmographic | What kind of company? | 1–10 vs 200+ employees | Sign-up form, enrichment |
| Geographic | Where? | Country, time zone | IP-derived location, billing address |
| Technographic | What do they use? | Mobile vs desktop, browser, app version | Device and browser data |
| Acquisition | How did they arrive? | Channel, campaign, landing page | Source and UTM data |
| Behavioural | What did they do? | Used feature X 3+ times in 30 days | Event tracking |
| Time-based | When? | Signup week, month, season | Event timestamps |
| Value | How much are they worth? | Spend band, plan | Orders, billing |
Behavioural splits get a page of their own at behavioral segmentation; groups that share a starting moment and are followed over time are cohorts.
How to choose a split
The best split depends on the decision, so work backwards from it.
- Name the decision. "Where do we spend the next campaign budget?" points at acquisition; "Who gets the onboarding email?" points at behaviour or signup age.
- Pick the dimension that could change the answer. If every group would lead to the same action, the split adds work and no insight.
- Make groups mutually exclusive and complete. Each person lands in exactly one group, and the groups add up to the total. If they do not, you cannot compare them fairly.
- Write each group as a rule a machine can recompute. "Mobile visitors in the last 30 days" is a rule; "casual users" is not.
- Check group sizes before you read the rates. A group of 40 cannot support a rate comparison, as the example below shows.
Worked example: the average that fell while every group improved
Sample data: a site compares March with April. Conversion rate is conversions ÷ sessions.
| Month | Segment | Sessions | Conversions | Conversion rate |
|---|---|---|---|---|
| March | Desktop | 6,000 | 300 | 5.0% |
| March | Mobile | 4,000 | 80 | 2.0% |
| March | All | 10,000 | 380 | 3.8% |
| April | Desktop | 4,000 | 208 | 5.2% |
| April | Mobile | 6,000 | 132 | 2.2% |
| April | All | 10,000 | 340 | 3.4% |
Overall conversion fell from 3.8% to 3.4%, and the team may start hunting for a bug. Split by device and the story reverses: desktop rose from 5.0% to 5.2%, mobile from 2.0% to 2.2%. What changed is the mix. Mobile went from 40% of sessions to 60%, and mobile converts at less than half the desktop rate, so the blended number sinks even though each group improved. This reversal of a trend when groups are combined is known as Simpson's paradox.
The segmentation lesson is concrete: a single blended rate answers "what happened overall" but not "did anything get better". Compare each segment with the same segment in the earlier period, then explain the overall change by separating the rate effect from the mix effect.
How many segments are too many
Every split divides the sample, and small groups give noisy rates. A rough 95% margin for a rate p measured on n sessions is 1.96 × √(p × (1 − p) ÷ n).
| Group | Sessions | Observed rate | Margin (± points) | Reading |
|---|---|---|---|---|
| Large segment | 4,000 | 5.0% (200 conversions) | 0.68 | 5.0% means roughly 4.3% to 5.7% |
| Tiny segment | 40 | 7.5% (3 conversions) | 8.16 | 7.5% means anything from about 0% to 15.7% |
The tiny segment looks 50% better than the large one (7.5% against 5.0%) but is statistically indistinguishable from it. Two practical limits follow: avoid splits that leave fewer than a few hundred sessions or a few dozen conversions in a group, and avoid slicing by many dimensions at once, because three splits of five values each make 125 cells. If you test many segments, some will look special by chance; treat surprising results as leads to confirm, not findings.
Common mistakes
- Reading only the blended average. See the example above: the total can move opposite to every part.
- Overlapping or incomplete groups. If a user can be in both "mobile" and "returning", say so, and do not add the groups up as if they were exclusive.
- Splitting on an outcome. Segmenting "users who converted" and then reporting their conversion rate proves nothing. Segment on what happened before the outcome.
- Fixed segments that go stale. "Active last month" written as a static list stops being true next month; write it as a rule with a relative date window.
- Too many cells for the traffic. Stop splitting when groups get too small to compare.
- Treating a segment as a cause. Mobile users convert less, but the device may be a stand-in for page speed, intent or checkout design. Segmentation finds where to look; it does not say why.
Data segmentation in TapCub
Take the March / April example. Here is how you would check it in TapCub.
- Split the same metric by device. In All-platform analytics › Behavior › Behavior analysis, chart the conversion event and use the breakdown (up to three layers) for device, channel or country. Turn on the compare option to line April up against March.
- Save the groups you will reuse. In All-platform analytics › Users › Segments & tags, build a rule segment, for example "did purchase in the last 30 days and did not refund". Rule segments refresh daily or on demand and keep a 30-day size trend, so you can see whether the group is big enough to read.
- Apply a segment to an analysis. Use Analyze on the segment to bring it in as a cohort; several segments in one analysis intersect.
- Mind the limits. Anonymous visitors have no profile and never enter user segments; a breakdown is still available for them in aggregate analysis. The weekly or monthly mix of devices is yours to read, so check session counts next to every rate.
TapCub does not tell you which segments are significant or recommend a split; it computes the groups and counts, and the judgement stays with you.
Sources
- Experian, What is data segmentation?.
- Coursera, What Is Data Segmentation?.
Was this explanation clear?
We use your answers to fix and expand this page.
Couldn’t send. Please try again in a minute.
Thanks for the feedback. We’ll use it to improve this page.You’ve already sent feedback on this page. Thanks.

Technical Engineer
Focuses on tracking code, SDKs, and API integrations, with practical guidance on implementation checks and troubleshooting.