Why Ninja? It’s the approach you take when you don’t want to hire an army –you just want to get something done quickly without making a lot of noise, and without following the rules of war. Dressing in black is optional but I find it useful to get into the mood.
People talk about DAU/MAU being a stickiness ratio, or an engagement metric. The ninja approach here is to use it wisely.
By which I mean only use it when you have no choice. It’s free - but you get what you pay for.
People THINK they know what DAU/MAU is. They think it’s something to do with the relationship between daily use, and monthly use. Something to do with retention. Something to do with engagement.
No no no.
Here’s the catch. DAU/MAU doesn’t tell you anything for sure about whether an individual user ever comes back, ever. That’s something you’d want in a measure of retention, for sure.Let’s look at this example, just quickly. Let’s say you have a game that kills off anyone who plays it. We’ve all been there, yes? Player A comes, aaagh. Player B, eurgh. Player C, (come on, let’s have a sound effect). This is not a sticky application, except if you mean blood sticky. But if you look at DAU/MAU on day 28 – it’s got a 33% DAU/MAU. Hello? By all means look at it. But don’t take it seriously. So, if you’re looking at your own data, you can do better, for sure. If you want to know about retention, look at retention. If you want to know about frequency and duration of play, look at it.
Another piece of ninja advice. Have spies everywhere. Don’t rely on any one of them. (You know what spies are like.)
Now if you’re going into a project with a big budget and full stack data science skills in house, you probably want to have a standing army. OK. But if you are a ninja, you want irregular forces that you can co-opt at low cost, and use all of them to the limit of their ability, even if their ability is limited. I worked on a project where we duct-taped 3 types of info together. Google Analytics – which does a handy session timing measure (excluding the last page, as you never know what the last page is until the universe ends), Facebook Insights, which is particularly powerful in combination with buying facebook ADS, and tells you stuff you can’t get any other way, stuff that only Facebook knows for sure, and MixPanel.When I’m saying we used duct tape for integration, that’s overstating it. Manual integration, cut and paste. Still very useful. Appropriate technology. You could build a dashboard with an integrated realtime feed, and you probably should, probably, for projects with a long lifespan. But you can do a lot with gestures.MixPanel is cloud-based with a freemium pricing model, which makes it a very easy on-ramp, and easy to ditz around with. I’ve also, personally, had really good experiences with their support. And the cool thing about it is it’s totally flexible data model. You throw it stuff, it catches it, and then it does its best to let you query it in a sensible way. It’s up to you to log what you think is important, when you throw it an event. You can append whatever you like in the way of event properties, when you send it an event to log. So it you want to say that when victim dies, professor plum was in the library with a candlestick, then go ahead. Gofor it. It’s happy for you to add OnVictimDie(timestamp, professorloc, weaponloc). You can capture as much context as you like, and interrogate it later. At some point you might find yourself telling it everything. That’s when you probably want to look at bringing it inhouse.I’ll talk a bit later about HOW you can use information from multiple spies.
Here’s another Ninja trick.
Admiral LordNelson’s view on strategy was ‘just go straight at ‘em’. This isn’t stealthy, as it involves cannon, and large warships, but it’s ninja.Basically the right tools are the ones you can use right away. When you outgrow them, get new ones.
Don’t worry about how you kill Professor Plum. Just do it.
It’s very ninja to use your sword a lot, so do it.
The basic sword technique is to never just have one slice. Go back for 2nds, and thirds.
Here’s a sanitised example. You’ve got 2 types of ad campaigns. One costs double the other, per acquired user. If you don’t know anything more, you would think the cheaper one is better. Sensible.
But the point is you don’t need to stop here. Look at what happens to the different groups of users. The cheap ones don’t stay, and the pricey ones do.So if what you want are people who stick around, then the seemingly expensive campaign is much better value. Twice as good.
But this isn’t the end of the story. If you drag in the Facebook Insights perspective on things, you can see there are some very interesting interactions between the variables. The cheap ad converted lots of GUYS. The expensive ad converted lots of GALS.And guys hate the app.Well, I lied a bit. It doesn’t tell you the reason. It doesn’t tell you why guys like the first ad, but they don’t like the game. But it’s something you can think about once you know what’s going on.
This is not rocket science. You can use rocket science, but you can go a fair old way by using your eyes and your brain.
this is a biggie
Ask big questions. Experiment. A/B is best but doing something is better than nothing. Ninja’s think tactically. Do users do better with a tutorial? When is bundled pricing better? What level of reward, for whom? Which reminders.Ninja’s also need to think strategically. What INTERESTING things can you learn about how to improve your app, by changing it in a systematic way? Not just butt-physics.
As a good Ninja you should be into plots. The more dastardly the better, really.People usuallly think of plots as hiding secrets. Ninja plots are different. Plots can reveal secrets.
Let’s say you’ve got two design variations, people using one spends £12 with you, and the other £16.That’s pretty cool. 33% improvement, whey hey! Let’s put everyone into Design B.
But not so fast. If you look at the average, you miss a lot of really interesting info.Both these picture show two groups with and average spend of 12 and 16. What the plot shows, in addition to the average, is the range of values in each group, and how frequent values are within this range.For the first set of data you can be pretty confident that the groups were different. And for the second, you can’t. There are ways of doing this with statistics, but your eyes are actually pretty good at it. And very ninja.And for social games, there’s a lot of interesting info in the tail. Revenue tends to fat-tailed.
If someone leaves something potentially valuable lying around in public, like on Slideshare, it’s ninja to run off with it into a darkened room. But once you’ve run off with your borrowed benchmark, or mechanic, or whatever, don’t swallow it wholesale.People are impressed by stray facts, like so-and-so’s app has a k-factor of 3, or .3, or a day 2 retention of 30%.The point is that if you’ve stolen this number from someone who really knows what they are doing, the number they’ve picked out is one which is important for their app. And they’ve found that out by looking. Is Day 2 retention the best predictor of long term loyalty, successful invitations, and good monetisation? Maybe. Or maybe not. If you intervene and try to boost Day 2 retention, will that be the same as Day 2 retention which is organic? Maybe. Maybe not.
as a ninja you are violent and quick but don’t necessarily have a huge brain. so don’t clutter it up.
All the other stuff you look at is only important because it relates to lifetime customer value. And – although there are ways of gaming the result - I do believe that value for the provider and value for the customer are linked.
So here’s a wrapup. If you want to stay in touch, check me out on Twitter, @HAStark.