Right now, your existing data and analytics is full of opportunities. If the right stakeholders are made aware at the right time, the business will...
Catch and fix ops issues sooner. Win more expansion or upsell opportunities. Shutdown fraud faster. Patch leaky user acquisition or onboarding funnels.
And your data team will look omnipresent and all-knowing for making it possible. So what’s the hold up?
Dashboards. Relying on dashboards and reports to capitalize on these opportunities is like trying to catch fish with a net full of holes. You’ll catch a few, but a whole lot more will slip through unnoticed. That’s a frustrating way to fish.
Dashboards aren’t designed with proactivity in mind. Between data latency, aggregation, and information density, dashboards are more suited for exploring lagging indicators and learning from mistakes.
But if you want to get proactive and avoid the mistakes in the first place, data-driven alerts—that deliver the right data to the right place at the right time—are the way to go.
Goodbye net full of holes, hello shooting fish in a barrel.
No fish were harmed in the writing of this piece ;-)
It's inefficient to rely on dashboards for proactively identifying issues or seizing opportunities in data
There are two foundational problems with using dashboards to be proactive:
They introduce too much latency.
How often is the data in dashboards refreshed? When and how often do stakeholders look at the dashboards? How fast does your actual business move?
If your dashboards are going to avoid the latency-trap, you need the answers to all these questions to consistently align in your favor. That’s asking a lot.
More likely, asking your stakeholders to react to dashboards means making decisions and taking action too late or too slowly.
If a high-value customer account is within 10% of its allocated usage or seat licenses, what’s the simplest way to get their account manager involved before they hit a limit (and have their work potentially blocked)?
You can hope the account manager scans the relevant dashboard (and finds the interesting data point) before the customer account has issues. But why take that risk?
Ultimately this approach isn’t reliable or scalable. It leads to subpar business outcomes: higher costs, lost or missed revenue, etc.
They aren’t focused enough
Dashboards aren’t designed to surface clear insights or drive specific action. A typical dashboard–even one following best practices–will contain 5-10 different widgets or charts (and often as many filters).
So it's hard for dashboard consumers to consistently extract relevant, actionable data. They get overwhelmed or sidetracked by dashboards that have a lot going on.
The dashboard above avoids many of the common dashboards mistakes. But it's still overwhelming and hard to quickly extract actionable insights from.
This dashboard, like most, is much better suited for exploratory analysis, after a broad question or area of investigation has already been identified.
In addition to being high-latency and unfocused, relying on dashboards is simply too manual. Even if your team reliability completes dashboards sweeps or reviews, they'll grow to resent it.
Data-driven alerts let your business users capitalize on opportunities in your data
You want to grow a more data-driven culture and business, right? Here’s why alerts triggered from the data warehouse and sent directly to stakeholders are the simplest path to that outcome.
Capitalizing on opportunities in your existing data and analytics isn't rocket science. Fancy algorithms are rarely required. It comes down to awareness and distribution—AKA enabling proactivity.
Stakeholders are busy and don't have eyes in the back of their head—and they can't capitalize on things they're unaware of. Alerts can put issues or opportunities on their radar that they would've missed otherwise.
Alerts also avoid the latency issues of dashboards To carpe diem and make the most of opportunities (or mitigate and minimize the impact of issues) business users need to act at the right moment. Data-driven alerts show up at the right time—and skip unnecessary intermediate steps like data extracts or dashboards.
Lastly, alerts also reduce the friction for stakeholders to take action. They reach stakeholders at the right place (e.g. Slack) and provide focused, actionable insights that have clear next steps.
What makes a good data-driven alert? Context
Alerts are typically short and to the point. But like most things, a lot goes into achieving that brevity and clarity. For data-driven alerts, context is the top consideration.
Any piece of data is only as valuable as the context it's in. One person's trash is another's treasure. Here are 4 types of context you should always keep in mind so you’re always delivering treasure.
A large portion of data's value comes from time. For many kinds of data, its value decays as time goes on. High latency dashboards are especially poor in this situation.
The data may have changed, so that using it means you’re acting on old info. Or the window of opportunity may have passed so it’s pointless or impossible to use the data to take action.
So how can you make it easy for your stakeholders to strike while the iron’s hot?
Imagine a model from my data science team suggests fraudulent activity by a particular user of my product. But my customer support team doesn't review the dashboard on schedule.
They don’t see and act on that insight until next week. The user's fraud has already cost us significantly.
Instead, a timely data-driven alert will push the customer support team to take action as soon as potential fraud is detected, without anyone having to manually review a dashboard.
Data is most valuable to stakeholders when it's easy to find, understand, and use. Instead of burying insights in databases or BI platforms that most business users don't spend time in and aren't comfortable with, good alerts meet stakeholders where they are—communications tools like Slack.
But it's not enough to push a dashboard link or image to Slack. Good alerts should deliver a “data package” to stakeholders' front doors, not ask them to go visit the post office.
If my customer success team is busy juggling 100 accounts per CSM, and our data clearly shows a spike in churn risk for one of them, why ask them to open and parse a dashboard in Tableau or Looker first?
Instead, the alert itself should contain the full insight.
We're all wired to be selfish. When it comes to getting the most out of data, analysts should always keep this in mind.
Stakeholders are far more likely to engage with and use data that's tailored to them—their accounts, their region, their products, etc.
Any analytics product that's too generic or broad will eventually be tuned out by most stakeholders. Consistent engagement comes from targeted content.
Your customer success team is a good example. Because we care deeply about our own stuff, it's much more effective to send "risk of churn" alerts that are personalized per CSM, covering only their accounts than it is to send all CSMs a lengthy list of all accounts at risk of churn.
Alerts traditionally contain a single metric or datapoint. But this is often only a partial insight.
The next step in the recipient workflow is always to “go get more data” to put the alert in context or understand what action should be taken.
If the next steps are consistent (i.e. go pull more data), why not include that data within the alert in the first place? Alerts should be focused and concise, but that doesn't mean they have to exclude obviously useful information.
In fraud prevention use-cases, why should an alert stop at: "Customer ID 123 appears suspicious" The alerts could also include explanatory details: relevant audit logs, associated risky IP address, last N transactions, etc.
Avoid noisy alerts (and prevent alert fatigue) by only sending relevant, personalized, actionable alerts
Alerts are great, and we strongly believe most teams are underutilizing them. But they're not without their risks.
The most common issue data teams face with alerts? Sending alerts that create noise for recipients. Noisy alerts will eventually be tuned out by your stakeholders. And cause stakeholders to lose trust in your data team.
In the worst cases, you'll be the boy who cried wolf and your stakeholders will tune out all your alerts, even the ones that do surface valuable, actionable issues or opportunities.
What makes an alert noisy?
An alert is noisy if recipients consistently don't engage with it. Stakeholders get to decide what's noise, not data teams. This is tricky, because you won't always know ahead of time what stakeholder engagement will look like.
Improve your odds of sending alerts that get engagement by creating data-driven alerts that:
- arrive at the right place and right time
- are personalized to recipients
- contain a complete insight
- have clear next steps
Even with improved odds, you should verify that your alerts are hitting the mark. You need a feedback loop that measures engagement and tracks outcomes—so you too can be data-driven and take iterative action to optimize your data team’s work.
Avoiding alert fatigue by sending only high-quality alerts can be complicated and labor intensive without the right tools. Rupert fills this gap and makes preventing noisy alerts easy for data teams and data analysts.