How insights are found

You don't need to know any of this to use Insights. But it helps to know what an insight means, and what it doesn't, before you act on one.

In short: at the end of every week and every month, GrowPanel does the drilldown a careful analyst would do, across every filter and grouping. Then it asks of everything it finds: is this unusual for this business? Is it big enough to matter? Could it just be chance? Only the few things that pass all of that are told.

What's watched

At every close, GrowPanel looks at the period that just ended:

• New business: leads, trials, new customers and new MRR. Leads and trials only if your funnel uses them, so a business that sells straight to paying customers never hears about trials.
• MRR movements: expansion, contraction, churn and reactivation, and net new MRR.
• Churned customers.
• Payments: failed payments and one-time revenue.
• Stripe fees: the share of the money that came in that went to Stripe, for accounts with Stripe payouts, at month-closes.

Each of these is also cut by the ways your customers can be grouped, as far as the grouping makes sense for it: where they are (continent, region, country), plan, billing interval, type of movement, acquisition channel, how they pay, and whether they started with a trial or a discount. On top of that GrowPanel watches the reasons customers give when they cancel, and what's already scheduled for the months ahead.

Finding the cause

This is the part you'd otherwise do by hand: open the report, filter by region, group by plan, try billing interval, and try again. For every change, GrowPanel tries every grouping, at every level, and looks for what drove it:

• Where: continent, region or country.
• What they're on: the plan (or plan group) and the billing interval.
• What kind of movement: for expansion, contraction and churn, things like discounts ending, upgrades, seat changes or failed payments.
• How they came in: the channel, for new customers.

A group is only named as the driver when it explains most of the change, and when it's clearly over-represented. A group that is 70% of your business and 70% of the change isn't a driver, it's just big. For places, you get the most specific one that tells the story: "the UK" rather than "Europe" when the UK is what changed.

For churn, the reasons customers gave when they cancelled are included, and the insight says so when one reason is clearly more common than usual. And when a single customer is behind most of it, you're told, because one account leaving is a conversation, not a trend.

For long-term differences it goes a step further and tries pairs of groupings too, such as customers in Western Europe who pay by card.

Changes in a group

Sometimes the total barely moves while one group changes a lot: new customers in one country doubling while everything else is flat, or churn jumping on one plan. GrowPanel compares every region, country, plan and billing interval with its own usual level, and with how the rest of the business moved at the same time.

The chart shows both, each against its own usual level, so you can see the group leave the rest behind. Because there are many more groups than metrics, the bar is higher here: the group must have moved a lot on its own, clearly differently from the rest, and the change must be worth even more before it's told. Only a couple are told per close.

What's normal for you

Every number is compared with your own recent history: the weeks before a week, the months before a month. There's no industry benchmark involved. A churn month that would be alarming for one business is perfectly ordinary for another, and Insights only cares about yours.

The usual level is set in a way that one odd month in the history doesn't drag it along. And a metric that happened to be very steady recently doesn't turn every small wobble into news.

Good news that has become normal isn't news

If new MRR has grown steadily for half a year, a seventh strong month isn't surprising. When a metric has been improving steadily, or has stepped up to a better level and stayed there, the bar rises with it. Good news has to beat its own trend to be told.

This only works one way. Bad news is always measured against the plain usual level, so a decline is never explained away as "the new normal". Slow declines even have their own check (see below).

Could it be chance?

Numbers move around by themselves. Ten more cancellations than usual can be a real change, or just a random month.

Every change is tested statistically, with a test suited to what's measured: counts of customers, amounts of MRR, or rates like failed payments. And because GrowPanel checks many numbers at every close, some would look unusual by pure luck, so the results are corrected for that too. Only changes that are very unlikely to be chance get through.

Some numbers are too small to judge fairly. A business with a handful of new customers a week hears about new customers at the month-close, when there's enough to go on, rather than every week.

Is it big enough to matter?

A change can be real and still not worth your time. So every change is valued in MRR, and it has to be worth a meaningful share of your MRR to be told. How big a share is up to you, in Insights settings. A $40 blip never becomes an insight, however unusual it is.

For MRR movements that value is simply the difference. For customer counts, it's the extra or missing customers at what a new customer is usually worth to you. For leads, it's the extra or missing leads at your usual conversion rate.

One story per cause

Changes rarely come alone. If leads fell, new customers and new MRR often fell with them, and net new MRR too. Instead of four insights, you get one story that starts where the change started, and follows it down.

The story also tells you what didn't move when that matters. "Leads fell, but new customers held at 212" means conversion improved. "New customers fell while trials were normal" means fewer trials converted, not that fewer came in.

Churned MRR and churned customers are one story. Net new MRR becomes part of whichever movement explains most of it, and is only told on its own when several movements add up to it.

What you already know

When you press We know why on an insight, GrowPanel treats it as known, and it moves down the list. When you also put what caused it on your timeline, the suggestions and GrowPanel AI name your explanation instead of guessing at a cause.

Slow declines

A number can slide a little every month without any single month standing out. At every month-close, GrowPanel looks back over the last half year or so for a steady slide in the wrong direction that adds up to something real.

Slow improvements aren't told. A business getting steadily better is what you're working for, not news.

A few, not all

Each close has room for only a few insights. When more changes pass every test, the most certain ones with the biggest impact go first. Weeks only flag the bigger changes; months tell the full story. When something big is coming up, you're always told.

Insights that are both very unlikely to be chance and worth a lot of MRR are marked Important.

How much you hear is up to you: the sensitivity setting goes from only the biggest changes to everything that stands out.

Ranked by what's at stake

On the Situation room, insights are ordered by the money they're about, weighed by how sure GrowPanel is:

• For a change, it's how far the number is from its usual level: churn $4.9k above usual is $4.9k at stake.
• For unpaid customers, it's what's likely to be lost, using how often invoices that late were never paid in your account.
• For a group that churns more, it's roughly the extra cancellations a month, valued at your average customer.
• For something coming up, it's the MRR it adds or takes away.

Money that's scheduled to arrive anyway, like discounts ending, doesn't push a problem out of your top three. It has its own card at the top of the page.

Money at risk

The insights about money at risk right now (unpaid customers, customers unpaid for over a year who still count in your MRR, busy renewal months and recovered payments) don't wait for a close. They're worked out from your billing data as it is now, refreshed during the day, and follow the same rule as every other insight: only what's unusual for you is told. Unpaid MRR, for example, is compared with its own usual share of your MRR over the last year, and the insight says how far above usual it is. The everyday level is always on the Revenue at risk tab.

How a story unfolds

A change that continues into the next period isn't a new insight: the same insight continues and counts the periods ("Leads below normal for 3 months"), with a timeline of each one.

When a period passes without it, the story ends in one of two ways:

• Back to normal: the number returned to its usual level.
• No longer flagged: the number is still high (or low), but it has been that way long enough to be the new usual. It's no longer unusual, even though it hasn't gone back.

Stories that lasted, or were important, stay under Recently ended for a while so you can see how they ended.

Weeks and months. Week-closes catch changes early; month-closes tell the whole story. When a weekly change is part of the same change the month tells, the month carries it and the week steps aside, so you don't read the same thing twice.

History is kept up to date. Every run re-reads your history with today's data. If an old invoice is corrected in your billing system, the insights from back then are corrected too. What you marked (Working on it, We know why, Not now, Not relevant) stays with its insights.

Cancellation reasons

Cancellation reasons have their own check: you're told when your customers start giving reasons when they cancel (from then on, churn insights include them), and when one reason becomes clearly more or less common than before.

See Kinds of insights for each kind, with examples.