Shopify Growth
The case against dashboards
The average Shopify merchant has access to more analytics than a mid-size retailer had a decade ago, and acts on almost none of it. The problem is not data literacy. It is that a dashboard answers a question nobody asked.
August 4, 2026 · 10 min read
Dashboards fail merchants because they report state rather than change, leave the diagnosis and the decision entirely to the reader, and demand a daily check to be useful at all. What replaces them is a small number of findings that each state what changed, why, what it costs, and what to do — delivered when something is actually worth knowing.
Open a typical ecommerce analytics setup and count the numbers on screen. Twenty is common; forty is not unusual. Now ask the question that matters: which of them changed this week in a way that should alter what you do tomorrow?
Answering that requires knowing each metric's normal range, its seasonality, its relationship to the others, and what a given deviation implies. That is a analyst's job, and it has been quietly delegated to a store owner who also handles supplier calls, customer complaints and this month's campaign. The tooling did not fail to deliver data. It failed to do the part that was hard.
What a dashboard actually asks of you
Consider a single tile: conversion rate, 1.9%, down from 2.4%. To act on it you must perform, unaided, roughly the following:
- Decide whether the move exceeds normal variation for your volume — which requires knowing your normal variation.
- Check whether it is a weekday artefact, a seasonal effect, or a genuine change.
- Determine whether traffic mix changed, because more low-intent sessions lower the rate without anything getting worse.
- Segment by device, country, channel and landing page to find where the drop lives.
- Identify what changed at the onset — a deploy, an app, a price, a stock-out.
- Estimate what it costs in money, so you can decide whether it outranks the other four things on your list.
- Decide what to do, and how to tell later whether it worked.
Seven analytical steps, per tile, per day. Multiplied by forty tiles, this is not a workload anyone completes. So the rational response is the one merchants actually have: glance at revenue, ignore the rest, and open the dashboard properly only when something already hurts.
Four specific failures
1. State instead of change
"Revenue this month: 84,200" is state. It is only interesting relative to something — the same period last year, the trailing baseline for this weekday, what you expected. Dashboards do show comparisons, usually against the prior period, which is the weakest available baseline because it ignores weekday effects and seasonality entirely.
A comparison worth acting on is against the same weekday over a trailing window, computed with a median rather than a mean so one bulk order does not distort it, and partitioned by calendar context so a promotional period is not silently compared with a normal one.
2. Equal weight for everything
A tile showing a metric that moved 40% and costs you 8,000 a month looks identical to a tile showing a metric that moved 3% and costs nothing. The layout communicates that all forty numbers deserve equal attention, which is false and expensively so.
Prioritisation is the scarce resource in a small business, and it is exactly the thing a dashboard declines to provide.
3. Pull, not push
A dashboard is only useful when you open it. The weeks you most need to know that something broke are the weeks you are dealing with a supplier problem and have not opened it in nine days. By the time you look, the drop has been running for a week and the cause is buried under everything that happened since.
4. Silence is unrepresentable
If nothing is wrong, a dashboard still shows forty numbers, and you still have to read them to establish that nothing is wrong. The state "you are fine, do nothing" — which is the correct answer most days — is the one state the format cannot express clearly.
What replaces it
Not fewer charts. The replacement is a different output: a small number of findings, each of which has done the analytical work before reaching you.
| Element | The question it answers | Why it is non-negotiable |
|---|---|---|
| What changed | What is actually different? | Stated against a seasonality-aware baseline, not last week |
| Where | Which product, channel, segment or page? | A store-wide number cannot be acted on |
| Why, or what coincided | What explains it? | Labelled as correlation unless causation is proven |
| What it costs | How much money, over what window? | This is what makes prioritisation possible |
| What to do | What is the specific next action? | A finding with no action is trivia |
| The math | How was this calculated? | A number you cannot audit is a number you cannot trust |
The last row carries more weight than it appears to. Once a system tells you a problem costs 3,400 a month, you are being asked to make a decision on its arithmetic. If you cannot inspect the formula, the inputs and the date window, you are trusting a black box with your merchandising. Every impact figure should be able to show its own derivation on demand.
The honesty requirement
A system that produces findings has a failure mode a dashboard does not: it can manufacture them. A dashboard showing forty flat numbers is boring but honest. A findings system with nothing to report faces commercial pressure to report something anyway, and a stream of low-quality findings trains you to ignore all of them within a month.
Which makes a few constraints structural rather than stylistic:
- Below a minimum sample, produce nothing. A pattern in eleven sessions is not a pattern, and reporting it as one is worse than silence.
- Say "started the same day as" rather than "caused by", unless causation was actually established. Most detected relationships are correlations and should be labelled as such.
- Show the range, not a false point estimate. An impact between 1,800 and 4,100 should be stated that way.
- State plainly when nothing needs attention. "Your store is healthy today" is a complete and valuable answer, and a system unwilling to give it will invent problems.
What dashboards are still good for
This is an argument about the default interface, not a claim that charts are useless. Exploration is real work and it needs a surface: investigating a specific hypothesis, examining an unusual period, or answering an ad-hoc question benefits from being able to slice data freely.
The distinction is between a tool you go to with a question and a tool that arrives with an answer. Exploration is the former and should be available. The daily default should be the latter, because the daily reality is a merchant with fifteen minutes who needs to know whether anything requires them today.
Frequently asked questions
Are you saying I shouldn't look at my analytics?
No — the argument is that the analytics should do the analytical work before reaching you. Reviewing data when you have a specific question is valuable. Scanning forty tiles daily hoping something jumps out is a poor use of the scarcest resource in a small business.
How is this different from alerts?
An alert fires on a threshold and tells you a number crossed a line. A finding states what changed against a seasonality-aware baseline, where it is concentrated, what it costs, what to do, and how the figure was computed. The difference is whether the diagnosis was done before or after the notification.
What if I want to see the underlying data?
You should be able to, always. Every impact figure needs an inspectable derivation — formula, named inputs, date window, assumptions. A system that cannot show its arithmetic is asking for trust it has not earned.
Doesn't this just move the judgement to whoever built the detectors?
Yes, and that is the point of insisting on published thresholds and visible math. The judgement is now made once, deliberately, by someone who calibrated it — and it is auditable. The alternative is the same judgement made ad hoc by a tired merchant at 11pm, forty times a day.
What about KPI tiles — are those always bad?
A bare number is close to useless on its own. The same number accompanied by why it moved, what it costs and what to do is worth showing. The tile is not the problem; the tile presented as the finished product is.
How many findings a day is reasonable?
Few, and often zero. A system producing several every day on a stable store is generating noise, and the honest output on most days is that nothing needs attention.
The verdict
The analytics problem in ecommerce was solved a decade ago. Merchants have access to more data than they can act on, and the marginal chart adds nothing. The unsolved problem is the layer above it: deciding what matters, quantifying it in money, and saying what to do.
That layer has to be honest about its own limits — refusing to report below a usable sample, labelling correlation as correlation, showing ranges instead of false precision, and saying nothing on the days when there is nothing to say. A tool that gets that right is worth more than the forty tiles it replaces. A tool that gets it wrong is a dashboard that also interrupts you.
More in Shopify Growth