Ecommerce Data Analysis: Core Metrics and Dashboard Design

Summary: In ecommerce data analysis the metric system comes before the data: organize metrics into the three layers of traffic, conversion and value, use a north star metric to anchor the direction of the breakdown, and only then talk about dashboards; dashboards are segmented by user, each layer answering a different question, so that no single giant table has to serve everyone.

Ecommerce operators face far more data every day than any person can process: browsing, search, add-to-cart, ordering, refunds, repeat purchase, every step of the chain producing numbers. Analysis without a system shows two classic symptoms. Either there is a pile of dashboards that still cannot answer the basic question of whether the business is actually doing well, or every analysis starts again from raw data, duplicating work with definitions that differ each time. This article gives a practical method for ecommerce data analysis: build a three-layer metric system first, fix the north star metric and its breakdown chain, then design dashboards segmented by user, and finally pair them with funnel, cohort and other common analysis methods so that data truly enters the daily rhythm of operational decisions.

Key Takeaways

The metric system comes before the dashboard: define the calculation rule and the owner of every metric first, then decide what goes on the dashboard.
Core ecommerce metrics can be organized into three layers: the traffic side asks how many people arrive, the conversion side asks how many of them buy, the value side asks what the orders are worth.
The north star metric solves the question of where to look: the whole company shares one core number that represents both the value users receive and the scale of the business.
Dashboards are segmented into three layers by user: the strategic layer watches trends, the tactical layer watches attribution, the execution layer watches detail; mixing layers is the main reason dashboards fail.
Funnel, cohort and comparison analysis cover most daily scenarios, and are more practical than chasing advanced models.
The metric system needs a quarterly checkup: cut the metrics nobody uses and add metrics for newly emerged business questions.

1. The three-layer metric structure of ecommerce data

What makes a metric system operable is one key property: every layer answers one clear question.

Traffic side: who is actually arriving? Core metrics include unique visitors (UV), sessions, channel mix, the share of new versus returning visitors, and the bounce rate. This layer answers where traffic comes from and how good it is. The supporting metrics for judging traffic quality are the above-the-fold arrival rate and the product-page arrival rate broken down by channel: with the same volume of visitors, a channel that sends people straight to a product page is worth more than one that drops them on the homepage and loses them there.

Conversion side: how many of them are buying? The core metrics are the full-chain funnel: product-page arrival rate, add-to-cart rate, checkout initiation rate, payment success rate, and the overall conversion rate. Note that the overall conversion rate must always be read together with the per-stage funnel rates: the overall rate is the result, while the stage rates are the levers you can actually intervene on.

Value side: what are the orders worth? Core metrics include average order value, items per buyer, gross margin, return rate, repurchase rate and LTV (customer lifetime value). This layer is the easiest one to overlook, yet it is the basis for judging growth quality: if a rising conversion rate arrives together with a rising return rate, the color of that incremental volume deserves a discount.

Beyond the three layers there should also be a supplementary product side dimension: sell-through rate, inventory turnover and cross-purchase rate. These metrics support category decisions, and they are the entry point through which data analysis moves from watching results to finding opportunities.

2. The north star metric and the breakdown chain

Once the metric system is in place, it needs an anchor that ties all the metrics together, and that anchor is the north star metric.

The selection criteria for a north star metric are that it reflects both user value and business scale, and that it can be broken down. Common ecommerce candidates are GMV, the number of purchasing users, and monthly active buyers. The north star framework proposed by Amplitude stresses that this metric should connect the value a product creates for users with business growth, and be driven by a set of input metrics: choose GMV as the north star and the input metrics become the drill-down metrics of the traffic, conversion-rate and average-order-value groups.

The point of a breakdown chain is that every team can find its own position on it: the acquisition team owns traffic and CAC (customer acquisition cost), the operations team owns conversion at each funnel stage, and the merchandise team owns average order value and cross-purchase. A north star without a breakdown chain is only a slogan; with the chain, it becomes the coordinate system of daily work.

3. Dashboard design: the right thing for the right person

A dashboard is the presentation layer of the metric system. There is only one design principle: segment by the decision frequency and the decision type of the user.

Layer	User	Update frequency	Question it answers	Metric count
Strategic	Management	Weekly / monthly	Is the direction right; trend and structure	≤10
Tactical	Business owners	Daily / weekly	Where is the problem; attribution and comparison	15–30
Execution	Front-line operators	Near real time	What needs handling today; detail and anomalies	No limit, on demand

The discipline of the strategic dashboard is restraint: show only the trends of the north star metric, revenue, the core conversion rate and acquisition cost. A strategic dashboard carrying more than ten metrics is one that nobody reads seriously.

The tactical layer is the main engine of attribution: drill-down comparisons by channel, by category, by new versus returning customers, plus before-and-after comparisons around campaigns. Dashboards at this layer should let users adjust dimensions themselves instead of freezing the view.

The execution layer serves daily action: zero-result search terms, out-of-stock alerts, add-to-cart-without-payment detail. Its value lies not in how polished the visualization is but in being able to generate to-dos directly: seeing an anomaly should let you locate the specific product or the specific channel behind it.

4. Four common analysis methods

With metrics and dashboards in place, four methods are enough to cover most daily scenarios.

Funnel analysis. Chain the stages of the conversion path together to find where the biggest drop-off happens. When you run funnel analysis, define the attribution window carefully: whether add-to-cart to payment is attributed same-day or across seven days leads to completely different conclusions.

Cohort analysis. Group users by their first-visit or first-order month and track the retention and repurchase curves of each group. It separates two situations that look similar on the surface but call for opposite responses: the value of existing users is rising versus the quality of new users is falling.

Comparison analysis. No metric means anything except in comparison: against your own history (year over year, month over month), against the target (completion rate), against a control group (A/B test). A single number presented in isolation carries no decision value.

Path analysis. Look at how users actually move between key pages. It is often used to surface the gap between design intent and real behavior: for example, the checkout entry clearly sits in the navigation bar, yet most users enter from a badge on the product page.

5. Rollout path and common pitfalls

Roll the work out in three steps: first spend two to four weeks completing the metric dictionary (definition, source, owner), then launch the two minimum dashboards for the strategic and execution layers, and finally add the tactical layer step by step as real analysis needs appear. This ordering puts the alignment of definitions, the step most often skipped and the most expensive to redo, right at the front.

There are four common pitfalls:

Metric inflation: laying out 200 metrics from the start when fewer than 20 of them ever get used, with the maintenance cost eroding the team's trust.
Definition drift: the same metric computed differently across dashboards, turning meetings into arguments over whose number is right; the root cause is the dictionary not being treated as a binding constraint.
Dashboard silos: dashboards built by the data team go unread by the business, while the spreadsheets the business builds itself conflict with dashboard definitions; the metric dictionary has to close this centrally.
Averages only: metrics such as average order value and dwell time get skewed by a handful of extreme values; looking at the distribution exposes the truth far better than looking at the mean.

FAQ

Q: How many metrics should an ecommerce data analysis metric system have?

A: Use "every metric has a clearly identified user" as the upper bound rather than the more the better. In practice, 40–60 metrics under a three-layer structure is enough to cover the core ecommerce scenarios, and anything beyond that is mostly redundancy that nobody references. Run a quarterly checkup: dashboard modules with no visits and metrics with no usage for two consecutive months should be demoted or deleted.

Q: Should the north star metric be GMV or the number of purchasing users?

A: It depends on the business stage and model. If growth comes mainly from acquisition, the number of purchasing users better represents health; if growth comes from raising the value of each customer, GMV or LTV fits better. The key test is whether the metric can be broken down into input metrics the teams can act on, and whether it also reflects the value users receive; a metric that only reflects revenue scale is easily distorted by short-term promotions.

Q: The add-to-cart rate is high but payment conversion is low. How should I analyze it?

A: Break the checkout step open and look at each segment: checkout initiation rate, form completion rate, payment success rate. Common causes include shipping cost only being disclosed at checkout, forced registration, and insufficient payment method coverage. Research from Baymard Institute shows that shipping cost and forced account registration have long ranked among the top reasons for cart abandonment, so you can order your investigation by these known factors.

Q: How often should a dashboard be updated?

A: Set it by the decision frequency of the user: weekly is enough for the strategic layer, daily for the tactical layer, and near real time for the execution layer. A higher update frequency is not automatically better; updating faster than the decision rhythm only burns compute resources and dilutes attention away from the key numbers. Compared with data latency, the far more common problem is a team watching high-frequency data with no matching action attached.

Q: How do I stop analysis conclusions from sounding reasonable but being unactionable?

A: Attach an action and an expectation to every conclusion: what to change, which metric is expected to move by how much, and when to re-check. For example: the cohort view shows the June (month 6) group has a 30-day repurchase rate 8 points lower, concentrated among users who got no guidance on their first order; the suggestion is to launch a repurchase reminder for first-order users, expecting the 30-day repurchase rate to rise 2 points, with a re-check in two weeks. Only conclusions carrying an action can enter execution.

Q: A small team is just starting out. Which dashboard should it build first?

A: Start with one table for the execution layer plus one chart for the strategic layer. The execution layer serves daily action directly and pays off fastest, while the strategic layer aligns definitions. The tactical attribution dashboard can be added gradually after the first two are running smoothly, following actual analysis needs, which avoids falling into the trap of building one large all-in-one platform from the start.

About Tengence

Tengence Search serves the ecommerce data analysis scenario described in this article, providing metric dashboards for the search and recommendation chain that cover key dimensions such as the zero-result rate, search conversion rate and query-term analysis, and can be plugged directly into the daily decision flow of operations. Tengence (TENGENCE) is the enterprise growth engine of the AI era, focused on public-traffic acquisition and private-domain conversion; its core products are Tengence GEO (dual-engine GEO + SEO) and Tengence Search.

Data Sources

Amplitude, The North Star Framework: proposes using a north star metric to connect user value with business growth, driven by a set of input metrics; it is the methodological source of the metric breakdown chain.
Amplitude, North Star Metrics: Examples and Definition: gives north star metric selection examples from multiple products together with a discussion of the selection criteria.
Baymard Institute, Cart Abandonment Rate Statistics: benchmark data putting the average cart abandonment rate at about 70%, plus the main abandonment reasons such as shipping cost disclosure and forced account registration.
Nielsen Norman Group, Conversion Rates: How to Measure and Improve: conversion rates must be interpreted together with segment dimensions and behavioral analysis; the aggregate number alone cannot guide optimization.