Part One: Strategy · Deciding what to build
New capability wins new customers and is easy to announce. The things existing customers already struggle with go on costing them every day.
A new feature gets a name, a launch and a line in the newsletter. A repair gets none of that. "Connecting a repository now takes two steps and not six" is an announcement nobody wants to write. Companies count what they shipped, and credit goes to launches. So the rough spots stay.
A rough spot is met again and again. The six-step connection is waiting for every developer on every team that takes up the tool. Existing customers pay that cost daily, and their opinion of the product is the one that reaches everyone else.
A product does need new things. A missing feature can lose a sale, and a product that only polishes will be passed by. So this is a rule about order, and the list of repairs never empties. In each stretch of work, repairs get the first share, because they are the surest work there is. You already know people use the thing, because they are complaining about it. You already know what is wrong. Very little new work comes with that much certainty.
In a large company the worst rough spots sit between teams, where one team's screen hands over to another's, and nobody owns them. Give each of those an owner.
Apple once built a release around this. Mac OS X Snow Leopard was presented in 2009 with a slide that read "0 New Features," half as a joke. It did add a few things, but the work went into speed and reliability.
Therefore:
When you choose what to work on next, look first at what people already use and struggle with. Give repairs the first share of every stretch of work, before anything new is chosen.
In practice. Look through the last month of support messages, or your own inbox, for the questions that begin "How do I...?" Each one that keeps recurring is a repair waiting to be made. Count how many people hit it. A repair with a number attached is hard to put off.
Goes with. You find the rough spots when you WATCH SOMEONE USE IT (59). Repairs should not move what people have learned: CHANGE SLOWLY WHAT PEOPLE KNOW (56). Mending the smallest flaws as you pass them is SOMEBODY LOOKED (85).