GA4 custom insights: setup, limits, and what they miss
How to set up GA4 custom insights step by step, the documented limits (50 per property, hourly is web only), a starter set for agency clients, and the gaps.
GA4 custom insights are rules you set on a property: when a metric crosses a threshold, or when Google’s model flags it as an anomaly, the insight triggers on the Insights dashboard and can send an email. You create them from Home > Insights > View all insights > Create, you need the Editor or Administrator role, and you get up to 50 per property (Google’s documentation).
They are free and worth setting up on every client property, but they stop at “this number moved”. Below: the exact setup, the documented limits, a starter set for an agency client, and what you still have to cover yourself.
What GA4 custom insights are (and how they differ from automated insights)
GA4 has two kinds of insights. Automated insights are whatever Google detects as unusual on its own. Custom insights are the ones you define: metric, segment, condition and how often it is checked. For an agency, that makes them a contract (“tell me when purchases hit zero on this client’s site”), and only custom insights can email you.
How to set up a custom insight in GA4
- On the GA4 Home page, scroll to the Insights card and click View all insights.
- Click Create.
- Either tick one of the suggested custom insights and click Create selected, or click Review and create to use a suggestion as a template. To build your own, click Create new under Start from scratch.
- Set the evaluation frequency: Hourly, Daily, Weekly or Monthly. Hourly is only available for web data.
- Set the segment. All Users is the default. Click Change to pick a dimension and value (a channel, a country, a landing page), and choose whether to include or exclude it.
- Set the metric, condition and value. Google’s own example is “30-day active users, % decrease more than, 20”. If you pick Has anomaly as the condition, you don’t enter a value.
- Give the insight a name.
- Under Manage notifications, add the email addresses that should receive it. Without emails, the insight only shows on the dashboard.
- Click Create in the upper right.
To pause emails or delete an insight later, go back to View all insights, click Manage, and use the Email Notifications switch or the delete option on each row. All steps above follow Google’s help page as of October 2026.
Has anomaly vs a threshold: which condition to pick
| Has anomaly | Threshold (% change or fixed value) | |
|---|---|---|
| You set | Nothing, GA4 decides | The metric value or percentage |
| Good for | Metrics with a stable history (sessions, revenue) | Hard rules (a metric at zero, a known floor) |
| Needs history | Yes: 2 weeks for hourly, 90 days for daily, 32 weeks for weekly | No |
| Typical failure | Fires on changes nobody cares about, or misses a slow decline | Too tight fires every week, too loose never fires |
Has anomaly uses GA4’s anomaly detection: a Bayesian state-space time series model forecasts the metric from its recent history, and the insight triggers when the actual value falls outside the credible interval. The training window depends on the frequency: 2 weeks of history for hourly, 90 days for daily, 32 weeks for weekly (Google’s documentation). On a property set up last month, a daily anomaly rule is working from less history than that window, so start a new client on thresholds.
Thresholds are predictable, which is their strength. “Key events less than or equal to 0” will never fire on a normal day and always fires when a tag breaks. Percentage rules are harder to tune: a 20% drop can be noise on a small site and a crisis on a large one.
GA4 custom insights limits
| Limit | What Google documents |
|---|---|
| Number of custom insights | Up to 50 per property |
| Hourly evaluation | Web data only; hourly insights for app events are unavailable because delayed app events would trigger false notifications |
| Anomaly training windows | 2 weeks (hourly), 90 days (daily), 32 weeks (weekly) |
| Who can create or edit | Editor or Administrator role |
| Where they show up | The Insights dashboard in GA4, plus optional email |
| How long they are kept | One year after they are generated |
Sources: Analytics Insights and anomaly detection, checked October 2026.
Two consequences for agencies. First, the 50 slots are per property, so the cap rarely bites; the real cost is that every set is built property by property, and Google’s documentation describes no way to copy a set to another property. Second, if a client only granted you Viewer access, you cannot create anything until they upgrade you to Editor.
A starter set for an agency client
Six insights per client cover the failures a client will blame you for missing. Treat the percentages as starting points to tune after the first month, not as benchmarks.
- Tracking break: key events. Daily, All Users, key events less than or equal to 0. The single most useful rule: it catches a broken tag or a consent change before the monthly report does.
- Tracking break: purchases (ecommerce only). Daily, purchases less than or equal to 0. Skip it for stores that can genuinely go a day without an order.
- Total traffic. Daily, sessions, Has anomaly. Switch to a weekly threshold for the first three months on a new property.
- Paid traffic. Daily, segment Session default channel group = Paid Search, sessions, Has anomaly. Paid budgets get paused or exhausted without anyone telling the agency.
- Organic trend. Weekly, segment Organic Search, users, % decrease more than 20. Weekly evaluation smooths out the daily swings that make organic rules noisy.
- Revenue. Weekly, purchase revenue, % change more than 25. Catches both a bad week and a good one worth reporting.
Send them to the account manager and a shared team inbox, not to the client. You want the chance to check what happened before the client reads “revenue dropped 31%” with no context.
What custom insights miss
They say what moved, never why
A triggered insight tells you sessions fell, not that the drop is entirely Paid Search, one landing page or one country. You still open GA4 and dig. (Our traffic drop diagnosis checklist covers that digging.)
Threshold rules have no weekday context
A threshold is one fixed number applied every time the rule runs. Google’s documentation describes the condition as metric, comparison and value, and nothing about matching the day of week. On a B2B site where Saturday traffic is half of Tuesday’s, a single daily percentage either fires every weekend or is too loose to catch a real Tuesday drop. Has anomaly is the better choice for daily rules, but the anomaly detection page says nothing about how it handles weekly cycles, so test it on your own properties before you trust it.
Email is noisy at agency scale
Six insights across fifteen clients is ninety rules, all landing in email. Soon someone builds an inbox filter, and the filter is where alerts go to die. Email also has no thread: when a teammate checks the drop, the finding stays in their head, not next to the alert.
No Slack, and no per-client channel
Custom insights deliver to the GA4 dashboard and to email, nothing else. If your team works in one Slack channel per client, the only native route is forwarding each email to that channel’s email address, which Slack offers on paid plans only. We cover what GA4 does and doesn’t offer for Slack in does GA4 integrate with Slack.
Where Dialytics fits
This is the gap we built Dialytics for, so read this section knowing it is our product. Dialytics checks your GA4 data every day and posts an alert in Slack only when yesterday is unusual compared with the same weekday over the previous weeks, so a quiet Saturday does not page anyone. When an alert lands, anyone in the channel can mention the bot in the thread and ask why it moved, and the answer draws on what happened in previous weeks. Every Monday it also posts a weekly digest of what moved and the one thing worth doing about it.
Should you still set up custom insights?
Yes. The tracking-break rules (key events and purchases at zero) are free insurance, and they work on day one with no history. Keep them even if you add another alerting tool.
If you manage one site and live in email, the starter set above may be all you need. If you manage many clients and work in Slack, treat custom insights as a safety net, not your alerting system.
To see how alerts fit with weekly reports and follow-up questions in Slack, read GA4 in Slack: the complete guide.
Frequently asked questions
How many custom insights can you create in GA4?
Up to 50 per property, according to Google's documentation. The limit is per property, not per account, so an agency managing fifteen client properties has fifty slots on each one. Most properties need far fewer: a set of five or six well-chosen insights covers tracking breaks, revenue and the main traffic channels.
Who can create custom insights in GA4?
You need the Editor or Administrator role on the property to create, edit and share insights. Viewers and Analysts can still see triggered insights on the Insights dashboard. If a client gave your agency Viewer access only, ask them to raise it to Editor before you build the set.
Can GA4 custom insights send alerts to Slack?
Not directly. A triggered custom insight appears on the Insights dashboard inside GA4 and, optionally, as an email. To get it into Slack you have to forward that email to a channel's email address, which Slack only offers on paid plans, or use a separate tool that posts to Slack.
What does Has anomaly mean in a GA4 custom insight?
It replaces a fixed threshold with Google's anomaly detection. Instead of a value you choose, GA4 compares the metric to a forecast built from recent history (90 days for daily evaluation) and triggers when the actual value falls outside the expected range. You don't enter a number.