Failing to Learn

Last year, the top 15 US companies combined spent over $150 billion dollars on R&D. That’s an awful lot of money to maintain a competitive edge when the returns are so uncertain. Alphabet (Google’s parent company) spends some 15% of its net revenues on R&D, while non-tech consumer companies tend to spend less than 2%. The top 5 spenders were all tech companies.

Developing and delivering successful products is really, really hard. Ideas that seem very clever in the lab may not catch the public’s imagination, with the ever-present risk of your proud product launch quickly being the next Google Glass, Apple Newton or Amazon Fire Phone and becoming the topic of social media derision.

Jeff Bezos of Amazon acknowledging not everything they try works. But the runaway success of Alexa shows what can happen when R&D works.
The Simpsons ridicule the Apple Newton handwriting recognition (Beat up Martin -> Eat up Martha)

Even if your product is great, standing out is still difficult. At somewhere like Google that now has over 85,000 employees, worthy new products may get drowned out by stunningly successful siblings that already count their user base in billions, even if they’re in unrelated markets. And if your product isn’t immediately great, then you can expect harsh — though frequently honest — criticism from consumers venting their disappointment.

Hit and Miss

While some of its products are among the most commonly used in the world, Google is no stranger to failure. Of the 160 Google-developed apps I can see in the Play Store, 17 of them have been installed on over 1 billion Android devices. While Apple don’t show install figures in the App Store, it’s safe to assume that there are many hundreds of millions of iPhones also carrying the most popular Google apps, perhaps doubling the total installed base to truly global reach and undoubted success. Despite the fact that many of these are installed by default on devices rather than by user choice, they still set a high bar for success.

The top downloaded Google apps

But among the 160 apps, there are some with surprisingly low numbers. Two high profile recent launches from Google that focused on its R&D-driven credentials, the Google Clips (the AI-powered camera with just 1,000–5,000 installs) and Google Jacquard (the gesture-enabled smart jacket at less than 1,000), are two of the worst performing apps. Although both are niche/experimental products, clearly they are not (yet) resonating with consumers. They may not be working in their current guise or at their current price points, but the underlying technologies are likely to stick around as Google looks for product-market fit with its customary willingness to fail, learn and iterate.

The bottom; it’s a rough start for some Google innovations such as Clips & Jacquard. (The Play Store doesn’t currently show the number of downloads for several apps specific to Google Pixel Phones — shown here as N/A)

One other interesting insight among the data from the Play Store is the performance of Google’s apps when ranked by consumer rating. What do consumers think of the app after they’ve installed it — how many feel strongly enough (either positively or negatively) to return to the store and rate it? Tens of millions of installs may sound popular but it doesn’t necessarily equate to happy users. Across the entire 160 Google apps, the average rating is a fairly solid 4.08. But looking at the apps sorted by rating reveals a worrying consumer response to Google’s foray into the living room. The seven worst-rated apps all related to Android TV (six if you discount the now discontinued YouTube for Google TV app). It would seem that some R&D spend needs to be directed towards improving the Android TV experience if success is measured by consumer ratings.

Android TV’s hall of shame — all 7 lowest rated apps made by Google. Restaurant review app Zagat has recently been sold, while Hangouts Chat gets low ratings from users frustrated that it’s a business rather than consumer app. Google VR services tends to get low ratings from consumers who don’t use VR but find the app taking up space on their devices.

AI to the Rescue?

Spending huge sums of money on new products is of course no guarantee of success. Even with careful controls, massive resources and hugely clever people, the cost of breakthroughs is pushing R&D spending ever higher as technology becomes more complicated. Although the adoption of new technologies is happening faster than at any time in history, there may still be a requirement for patience on the part of companies. Historically, Edison said “I have not failed. I’ve just found 10,000 ways that won’t work”, but today’s widespread focus from both investors and consumers on instant gratification means there is little leeway for products that take too long to deliver on their promise. Knowing when to persevere or when to go back to the drawing board is tough. And deep pockets is no guarantee of success. Sometimes consumers just don’t want a technology (like 3D TV) however much its makers believe in it.

In an era of instant feedback and impatient investors, how will these powerhouse companies manage their R&D spending in the future? No doubt much of the current R&D focus from tech companies is on artificial intelligence technologies such as deep learning. What are the odds that one of the key applications for these technologies will be trying to better predict future investments based on AI-powered assessments? There’s something ironic about one of the early applications of AI perhaps being to assess where technology can be successfully added to products. Will R&D departments of the future consist of a computer pitching ideas to another computer which will reject them? Would an AI have recommended the launch of an AI-powered camera or a touch-enabled jacket? Will advances in AI lead to more product ideas being killed before they come to market or will consumers remain the ultimate judges? What about advances that are totally serendipitous — like the unintentional discovery of penicillin or microwave cooking — will AI models be able to predict these and their consumer appeal?

Those looking to AI to accelerate new product development will need to remember that no amount of algorithmic sophistication will compensate for a lack of data — positive results alone will not be sufficient for training AI — the analysis of failures will be a crucial input to identify the successes of the future, and companies whose cultures allow failures will have a crucial advantage.

— — — -

Disclosure: The author personally owns several of the products mentioned here including a Google Glass, Google Clips, Jacquard, Android TV & Amazon Fire Phone — and he quite likes them!

App download and ratings stats accurate as at April 14th 2018 in the US Play Store