Meanwhile, in San Francisco…We’ve been in our new office, aka “The Sunroom”, for four days now. These have been our observations/experiences thus far:
You know what, San Francisco? You’re OK. I think we’ll stay.*I’m know, I’m justifying. And yes, I also have special commuter socks that I tuck away into my shoes. I’m becoming my own worst nightmare.
- Despite it’s name, The Sunroom is not, in fact, warm or sunny. It’s freezing. Cory and I are displeased, while Gooley seems to be thriving in the frigid temperatures.
- We have discovered a worthwhile contender for the Taco Tuesday tradition: The Iron Cactus.
- I have become victim to the “commuter sneaker”. Yes, that’s right; I have special commuter shoes. I feel like this is more acceptable in San Francisco than LA, as the closest BART station is a mile away.*
We are slowly, but surely, making friends with the other people in our co-working space.Scratch that—just accidentally cut someone in line for the coffee machine this morning. But, it is really nice to be around other humans.
- Gooley just signed us up for EAT Club, which is basically the most awesome perk ever. Their ‘get started’ email just says: “Sign up and order your lunch by 10:30AM. Your lunch arrives no later than 12:30PM”. No thrills, no frills, just tell it like it is. That’s my kind of lunch.
Cyclical Metrics and Account Health
A large part of determining account or customer health is building adaptive models of what “normal” behavior looks like and then reacting to behaviors that deviate from these models. In many cases the question can be put simply: “How often does a healthy account or customer engage with feature X?” and “How healthy is an account or customer given their engagement level with feature X?”
One common case is when an account or customer pays for a monthly subscription to a product or service. We might expect a healthy customer to come back each month and engage with the product or service at a level that meets or exceeds their previous engagement. But how can we know when a customer starts to fall below their typical engagement level? Is there a way to be certain (in a statistical sense) that what you are looking at is abnormal?
Our approach to solving this problem is to build a model for each account or customer for signals that are anticipated to repeat themselves month to month (i.e. engagement levels, counts of actions performed, etc). The model adapts to the historical trend and tells you for every hour in the month what you should expect to see and with how much statistical confidence.
The image above is one such model. The red circles are the recorded values for a certain customer action count (i.e. sending emails) and the blue line is what we predict to happen based on this history. The model provides not only a prediction but a confidence interval for each point in time and plotted is the 95% confidence interval.
A simpler approach would be to make predictions from a linear model. The linear model is built on the following logic: we know that by the end of the month 200 actions are usually performed, if it’s 50% of the way through the month we expect that 100 actions will have been performed by now, etc. This is the model reflected by the red diagonal dotted line. As you can see, the linear model neglects a large amount of predictive power of this signal. Namely, we can see from the plot that this customer tends to over-engage at the start of the month and taper off toward the end. This is captured by the more flexible statistical model that we created.
The flexible statistical model we developed is capable of fitting a wide array of behavior patterns and adapts automatically when these trends change. In the image above, you can see how the model accounts for the deviation in customer behavior in the middle of the month and as the newest historical data runs out (the red dots in the upper right) the prediction smoothly converges with the historical trend.
This periodic modeling is one of many components rolled into a single customer/account health metric summarizing the overall health of a customer or account. By quantifying what is normal in a flexible and probabilistic way we can create meaningful indicators that are adaptive and responsive to sudden changes in behavior yet not overly sensitive to slight variations.
The Power of Recurring Revenue
OK, great. You are a salesman for XYZ SaaS Company, and you just sold a customer your premium package. Your boss gives you a nice Commission check so you go home and treat yourself to a nice fancy dinner. The deal is done, the money is collected, and you return to work the next day in search for the next big sale. The previous customer is lost in a list of subscribers as you wait until their contract ends and hope to close them on a new one. The time between “contract signed” and “contract expired” is quickly transforming from a period of satisfaction to a span of opportunity.
“What do you mean by opportunity, the deal is done?”, somebody might ask. Recent studies have shown that it can be 6-7 times more expensive to find a new customer than to up-sell a current one. By no means am I saying that you should stop looking for new accounts, but more company time and resources should be allocated in attempt to increase recurring revenue through current ones. Whether it be a mobile gaming company with in-app purchases, a package upgrade of any SaaS subscription, etc…, there is always the opportunity to boost revenue through constant customer engagement and interaction.
You will soon come to realize, however, that the process of keeping a customer happy is a battle that is not easily won. Success agents are constantly walking a fine line between constructive and possibly destructive communication. You want to let each user know that they are valued as an individual, while simultaneously keeping your distance. Now the easiest way to enhance a relationship with a customer is to be available and quick to resolve any issues that may arise. But how do you stay engaged with a customer who does not send in a single support ticket? Do you assume that they are fully content with your service and seek no additional assistance? As a matter of fact, support teams are not aware of most issues because their users don’t take the time to report them. Here is where the opportunity is born! You simply need the right tools to capitalize on that opportunity.
The ability to track individual/account churn likelihood and engagement patterns can turn your customer support/success teams into an up-sell powerhouse. By giving them detailed user activity, agents have the necessary information to proactively reach out to customers regarding any actionable behavior. For example, if an account’s usage pattern begins to decrease, a support agent can be alerted to contact the account before it churns. Furthermore, onboard monitoring allows success agents to nudge accounts towards experiencing new features so that they get all of the value out of the product. When the time comes to renew a contract, company XYZ can be confident that they took all the appropriate steps to retain that customer. At Preact, this is exactly what we aim to do!