IOT: The Competition for Attention

My most recent post on LinkedIn cross-posted here.


The Internet of Things (IOT) is everywhere. It was the only news at CES and it’s presently at the top of the hype curve with press attention on Apple’s new secret car,Homekit becoming available with iOS 8.1.3 and the Apple Watch launching soon!

But with IOT comes a whole new set of problems, and I (selfishly) believe that calendar and AI will be a key pillars in enabling the future IOT software platform.

Why? IOT devices are not meant to collect dust but to be actionable. Being actionable means being used and when your Amazon Echo sits on your counter and collects dust, it’s not actionable. To mitigate this, every IOT device needs to speak-up – they are competing for your attention.

To compete, they send notifications (email, SMS etc): “Cheerios are on sale,” “You’re running low on milk,” “You’re laundry machine will be done in 35 minutes,” “You haven’t walked enough steps today” and so forth.

As I alluded to in a previous post, your email is transforming from a collection of human communication to machine generated messages and tasks. And your calendar will follow because every great IOT notification is an actionable task which needs to be scheduled back into your calendar (eg “Oil change needed soon”).

So How Can We Fix This?

Well, first off, it’d be great if just 10 percent of the notifications I get on my phone were useful. And to do that would not be hard:

  1. Let me configure my notifications – Facebook and others give me too many notifications and IOT is going to fall into the same trap. Yes, I don’t want to be notified by my smart sink each morning that the water quality has negligibly gone up or down. It’s a novel concept at first but it gets old quickly — and even faster if you are sending it to me on every device!
  2. Learn which notifications I read and more importantly respond to – Email marketers are experts at this, they know when you open and click a link in an email. Notifications are the same thing, they are just another form of CRM. See Kahuna for example.
  3. Use the signals you have – More signals are not necessarily better, but the right signals can make a huge difference. Dear Nest, if you can tap into my calendar to better know when I’m home or not, and subsequently save me money on my heating bill, that’d be awesome!
  4. Be smart about when to bother me – Imagine a real-world assistant receiving a call while you were in a meeting. He wouldn’t interrupt your meeting unless he thought the call was important enough. This is hard and this is where AI and using the user’s own data (calendar, email, Facebook, Linkedin, location etc) to understand intent can make a huge difference. I do want to know “Cheerios are on sale” but only when I go shopping.

Next, distinguish between content and tasks. Most notifications are content and this can get overwhelming real fast. Two years ago, maybe one app told me whose birthday it was today but now I receive this notification from multiple apps on multiple channels each day this is annoying. Be smart about firing content notifications and focus on the unique, not the obvious.

Even better though are task-based notifications; telling me I need to buy milk is better but I don’t want that to clutter my email. Instead, map that task to my calendar, add it to my todo list, and present it to me at the right time.

The competition for attention is real – it’s happening and IOT will take it to a new level. Leveraging simple AI smarts, providing configuration and mapping those task notifications to your todo list and calendar will help. The winning IOT devices are those that are smart enough to keep my attention over the long run.

Ad Learnings from Recipe Search

As part of YumYum Labs, earlier this summer, we built Recipe Search (top 10 app in the Android Health category). Yesterday, I shared on a panel at AppNation, some of our ad learnings that I thought would be useful to you:

– We did a quick and dirty test to see if users were willing to upgrade for the removal of ads. Our test wasn’t perfect – the user could pay whatever they wanted (1 cent or 1 dollar etc) and the billing workflow was processed through mobile Paypal as opposed to Android’s in-app billing (this test was done before that was available). The result was abysmal, some 10 users converted amongst 300-400K (at the time). Certainly, the result may have be better today with in-app billing but my conclusion is that the expected conversion is going to be low unless the ads become annoying assuming your users don’t abandon you (eg has anyone else noticed the increase in ads on Pandora!). Cathy Edwards with Chomp had suggested that our user base was self-selected and if we instead had made a separate paid app , we would have seen better conversion (since those would have been users willing to pay as opposed to those who only download free apps). Note, there was consensus that the iPhone seemed to have more users willing to pay versus Android and I’m curious if anyone has done any analysis as to why (besides number of cached CCs)?

– We saw significant variance in eCPM amongst different ad networks. You can never do enough A/B testing and it is absolutely one of the biggest ad monetization mistakes I’ve made with previous projects in my early days (eg ToneThis, TinyTube etc). Moving from 1 ad network to multiple ad networks instantly resulted in a 30-50% improvement. The reality is that eCPMs are not the same and either building your own or using third party mediators such as Nexage may be your best bet. In addition, make sure to test as many placements as you can. With TinyTube, for a very longtime, we didn’t put ads on the homepage because we thought it would upset the user. One day we decided to do so an instantly doubled our revenue with no increase in user attrition!

– We saw interesting eCPM variance depending on the time of the day and/or the week. I understand variance based on season (eg end of the month quotas or end of year spend etc) but to see variance through the day was interesting and I wonder if that was based on the algorithms ad networks used (eg serve high value inventory first with a ad spend cap for the day). Curious if anyone else has seen similar?

– Passing location certainly improved the eCPM but resulted in a different problem. Our users were getting upset to the point of not installing the app wondering why we were asking for location. Unless your app needs location (eg specific feature) which a Recipe Search app does not, than asking for location may result in user backlash.

– There was some discussion/debate about whether to have ads in the product from day 1. I’m of the opinion that it’s best to introduce it initially because no matter what you do, you are going to upset users. If you do it initially, it’s expected – if you introduce it later, you will definitely see some negative commentary! Jean Hsu with Pulse News mentioned that they had a paid app and than an ad-supported version of Pulse and users started commenting that if they (Pulse) introduce ads, they want their $1 back 🙂

– There is significant distinction between mobile web and in-app advertising. With mobile web, I’ve absolutely seen a significantly better eCPM for the top ad placement (eg top of the page) usually because the CTR is almost 2X of text/banners that are displayed in the body or footer of the page. With in-app advertising, I’ve seen fairly consistent CTRs regardless of placement (floating header or floating footer). This is important because you can design your app accordingly and then place the ad where it is most convenient.

– We touched on this briefly yesterday but there are various tricks we used to SEO/SEM. Everything from titling your application with keywords to titling the name of the company with the keyword as well (probably all good practices generally). Rob Coneybeer with Shasta Ventures had also mentioned the importance of the icon and real use cases with Smule’s applications as to how usage varied based on having a cool or crappy icon.

– Keyword versus categorical targeting – keyword targeting with mobile ad networks is mostly unavailable but categorical targeting is definitely there. Curious what others have experienced but we found that choosing all categories as opposed to the category you are most relevant to results in a higher eCPM since the ad network presumably than delivers the highest value ad across all the categories as opposed to just the one category. I imagine this is only temporary until we reach a point where the ads available become so large that the eCPM has effectively normalized (as opposed to the high variance we see today). Really the big question is how do you index the information inside of an app so we can do keyword targeting (especially if apps are the new web!)

Would love your feedback!

(BTW, link to an Intel blog post summarizing another panel I was on at AppNation)