What if you had a conversation with your metrics about how to improve your business? Or more accurately, what if your voice-enabled agent asked you every day how you were doing?
A simple idea with with big implications
You might start with a simple question, like:
Hey Siri, how many widgets did we sell yesterday?
Ok Google, Which Customers Do I Need to Talk To Today?
And you might proceed to a more complicated idea, like:
Help me write a white paper to encourage more people to try my software today who match the “Small Business” segment.
In a few simple questions that you answer about your business you could determine what to do next, understand how things are going, and earn valuable insights you might not have anticipated.
Meh – you say – not possible. Objectives and key results are based on multiple variables that are difficult to correlate effectively. The metrics they are based on do not collect themselves. In addition, how would you isolate the behavior that drives these metrics?
This is a human-centric way of thinking. Because we’re used to the idea that machines are dumb calculators or not yet capable of building models to make the kinds of decisions we make every day, we discount the future that might happen if we create a model for decision-making that we train every day.
So let’s conduct an experiment. Think about your business and identify one or two metrics for acquisition, behavior, and revenue that you want to track on a daily (or weekly or monthly basis). Create a simple spreadsheet to track the data series putting the name of the metric as the column, the date as the row, and the value in the intersection of the row and column.
The Challenge in Learning About Your Data
What now? The biggest problem in capturing in calculating data about everyday metrics is to answer the “so, what” question. That is, what about the everyday (or slightly less often) metric creates an objective or a key result that you care about?
The output metrics here are contingent on your business type, and probably look familiar, e.g.
- Period for learning -> how often do things happen, e.g. daily, weekly, monthly, quarterly?
- Sales -> how much is sold, how many people started a subscription, how big was the deal?
- Retention -> how many people left in the period out of how many total?, how much revenue does an average customer represent?
- Activity -> how often does an average customer login or use your software?
The hard part is understanding the inputs into those metrics and finding the ones that matter. For example, you might find it easy to understand when someone puts their credit card into your system or signs an annual contract to commit to your business for the next year, while at the same time it is not easy to predict whether a given customer will be more likely to stick around based on the way they use your product.
The perfect “mad lib” for this hypothesis might look like the following:
where the acquisition type describes a method for finding a certain type of user who is more likely to engage in a repeatable behavior that creates a signal identifying that person from that channel as more likely to convert.
Computers Match Patterns More Effectively Than You Do
People don’t match this data very well in more than simple results. It’s not hard to know that a particular cohort of trial users is more or less likely to end up as your paid customer. It’s very hard to find out why.
When you think more like a computer you will start building a model that is easier to feed to a machine learning process.
Imagine this set of steps to evaluate business metrics:
- Find a metric you want to affect, like “number of days to close”
- Identify the potential inputs for that metric as a hypothesis, like “number of logins in first 7 days”
- Find a group of users in a cohort (if you have a 14 day trial, the cohort would last 14 days) and measure the number of logins they have, and whether there is any change or difference between the people who engage with your organization a lot, and the ones who engage and end up buying
- Store the data in an easy-to-review place like a Google Spreadsheet.
Watch the data for a few cohorts in a row, and use the Google Explore function to see what you learn.
What will you learn? Maybe nothing at first, but having the data in a format you can ask questions of will help you to start to see patterns.
In our fictional example you might find that most people who buy your product login at least 4 times during the 14 day trial period, and this observation could lead you to design an experiment to increase the number of logins during the trial period to see whether it has a positive correlation to the sales rate.
Using Reminders As a Hack
The next step (today) might be setting a daily reminder in your voice agent to review the results you’re seeing in each of your cohorts to determine whether there is any movement in the data.
Siri won’t be able to analyze your data (yet), but will help remind you to focus on the same part of your business daily and improve things a little bit at a time.