(Disclosure: I work for Google, but not on Android, and in any case, these are my personal opinions.)
Marco points out iPhones’ poor battery life, and wishes that Apple prioritises battery life over thinness. Indeed, battery life has been rated as the most important factor in deciding what phone to buy.
OEMs and OS vendors like Google should set themselves a goal that everybody makes it through the day without running out of power. We can define a day as lasting 17 hours — from the time the phone is taken off charger at 100% charge, it should last 17 hours. If it’s infeasible to make every single user’s phone last a day, we should at least aim for 99% of users making it through the day.
OEMs and OS vendors should collect data  from end-users’ phones as to when they run out of power and shut down, and how long ago they were charged to 100% . With this data, they can answer questions like, “What percentage of Galaxy S6 users run out of power during the day?”
This will make the whole exercise data-driven, as opposed to having to read subjective opinions like “it easily lasts a day”. Some sites like AnandTech do better, with battery tests, but a synthetic test is nowhere as accurate as actually measuring real-world data. In the absence of objective data, we don’t even know how well we are doing, so how can we improve the situation? Buyers won’t know how long-lasting a phone they are buying.
With real-world data available at scale, we can slice and dice it in various ways, such as at the company level (“What percentage of Samsung phones make it through the day?”), by when they were introduced (“What percentage of phones introduced in 2014 make it through the day? How fast is battery life increasing over the years?”), by the age of the phone (”What percentage of two-year-old phones make it through the day? How quickly do batteries degrade?”).
We can also see statistics by the country. For example, battery life is probably lower in India than in developed countries, given the poor signal strength. So, an Indian user buying a phone can see the statistics for India, which all that matters, anyway.
This data should be released publicly, to drive competition. Or companies can just release a list of their models sorted by battery life. If 90% of Note 4 users make it through the day without their phones running out of power, but only 80% of S6 users do, then the Note 4 would rank higher than the S6. Companies should rank all their devices this way, and publish the list. Since they don’t have to release the exact numbers, only the order, they don’t have to worry that they’ll lose customers to their competitors. If Samsung says that the Note 4 lasts longer than the S6, that doesn’t motivate me to buy an LG phone, so Samsung needn’t worry that they’re losing sales.
At a high level, companies should start doing real-world measurements at scale, so that we know where we stand and can measure our progress forward, and choose to buy phones that last long.
 Ideally, this data collection will be opt-out, but with a dedicated screen during the set-up flow explaining what data will be connected and why (”to make your phone run longer”), with a big, clear Opt Out button. Opt-in will result in far fewer data points.
 For this exercise, ignore phones that were charged partially. That is, taken off the charger before they reached 100%.