I learnt recently that even when all indications are that your business (or life. or city. or whatever) is running fine, something could be wrong – in plain view – that those indicators can’t tell you.

The only way to discover these problems is to tinker around with data. Ask questions of it.

Here’s what happened.

Ours is a prepaid subscriptions business. You sign up, put money into your account and pick your subscriptions.

Our signups, payments and new subscriptions – the three primary indicators of our health [1] – were growing at expected rates relative to each other. Nothing seemed to be wrong.

Until one day, I discovered that many new signups didn’t have any subscriptions. That was unexpected. It meant that most of our new subscriptions were via older subscribers.

That meant – and this was quickly confirmed – that most of our payments were also made by older subscribers. This is a problem, and we commissioned a quick survey to find out what was wrong with our new signups.

But then we tinkered further. We plotted a histogram of (normalized) new subscriptions started (all of this is excluding renewals) versus how long ago the subscribers had signed up, and we found this:

Click for a larger image - bet you can't read the tiny text

Astonishing. The older the subscriber, the more the number of new subscriptions they started recently [2]. We had a larger problem than we expected; our older subscribers were so active, they’d hidden how un-engaged our newer subscribers were for several months.

While we took immediate steps to fix this, we also realized that it’s hard to build a dashboard for stuff like this. You can – and should – track primary measures of success, results of specific campaigns, and suchlike. But under-the-surface stuff like this – we’d never have figured it out if we hadn’t tinkered with data.

–
[1] There’s also ARPU and churn, but they aren’t material to this discussion.

[2] The data for months 8 and 9 is skewed by a small set of people with a lot of subscriptions each, but they’re still much higher than any other month, and the trend is the same

(Cross posted from the MyToday blog.)

A post on Fred Wilson’s excellent blog about cost of acquiring each customer versus the lifetime value of that customer. And it’s pretty simple: “LTV has to be greater than CPA or you won’t be able to scale – or, for that matter, survive.”

This seems obvious. But when you’re preparing a revenues-versus-costs estimate for a business plan, you often overlook how much you earn versus spend per customer over time. Here’s a slight variation of that from a few weeks ago:

The CEO of our firm shot down a recent plan I presented, one that involved both the mobile web and SMS working in tandem. The product was different, compelling, and the estimates said we’d be profitable in a year on the gross. But our SMS costs were 80% of the revenue we would have earned from advertising.

“Keep SMS out; figure out the mobile web part. If you’re spending 80% of your revenue on acquisition and retention, you won’t have enough to spend on content and infra and operations and product innovation – and that’s not even counting people.”

And this was true not just in the month we acquired the customer. Month on month, the SMS costs kept pace with the ad revenue per customer [1], so we’d never have enough money to spare. In other words, the CPA was lower than the LTV. But not nearly low enough [2].

–
[1] It was also likely, I realised later, that over time the customer would yield less ad revenue as he/she tired of the service, but the SMS costs would be the same. So we would have to evaluate the customer’s worth and adjust SMS quality of service constantly, making things rather complicated.

[2] As an aside, these costs also grew linearly not just over the lifetime of each customer, but also with the addition of every new customer – there were no economies of scale to be had. If there were, the total lifetime value of all customers would have grown faster than the total (SMS) cost of acquisition and retention, and it would’ve been viable after a point of time.